feat: add daily model statistics aggregation with stats_daily_model table

- 自动刷新改为按钮切换模式，移除独立Switch开关
- 自动刷新间隔从30s改为10s
- ESC关闭弹窗后blur焦点，避免样式残留

.env.example

@@ -1,8 +1,16 @@
 # ==================== 必须配置（启动前） ====================
 # 以下配置项必须在项目启动前设置
 
-# 数据库密码
+# 数据库配置
+DB_HOST=localhost
+DB_PORT=5432
+DB_USER=postgres
+DB_NAME=aether
 DB_PASSWORD=your_secure_password_here
+
+# Redis 配置
+REDIS_HOST=localhost
+REDIS_PORT=6379
 REDIS_PASSWORD=your_redis_password_here
 
 # JWT密钥（使用 python generate_keys.py 生成）

.github/workflows/docker-publish.yml

@@ -15,6 +15,8 @@ env:
   REGISTRY: ghcr.io
   BASE_IMAGE_NAME: fawney19/aether-base
   APP_IMAGE_NAME: fawney19/aether
+  # Files that affect base image - used for hash calculation
+  BASE_FILES: "Dockerfile.base pyproject.toml frontend/package.json frontend/package-lock.json"
 
 jobs:
   check-base-changes:
@@ -23,8 +25,13 @@ jobs:
       base_changed: ${{ steps.check.outputs.base_changed }}
     steps:
       - uses: actions/checkout@v4
+
+      - name: Log in to Container Registry
+        uses: docker/login-action@v3
         with:
-          fetch-depth: 2
+          registry: ${{ env.REGISTRY }}
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
 
       - name: Check if base image needs rebuild
         id: check
@@ -34,10 +41,26 @@ jobs:
             exit 0
           fi
 
-          # Check if base-related files changed
-          if git diff --name-only HEAD~1 HEAD | grep -qE '^(Dockerfile\.base|pyproject\.toml|frontend/package.*\.json)$'; then
+          # Calculate current hash of base-related files
+          CURRENT_HASH=$(cat ${{ env.BASE_FILES }} 2>/dev/null | sha256sum | cut -d' ' -f1)
+          echo "Current base files hash: $CURRENT_HASH"
+
+          # Try to get hash label from remote image config
+          # Pull the image config and extract labels
+          REMOTE_HASH=""
+          if docker pull ${{ env.REGISTRY }}/${{ env.BASE_IMAGE_NAME }}:latest 2>/dev/null; then
+            REMOTE_HASH=$(docker inspect ${{ env.REGISTRY }}/${{ env.BASE_IMAGE_NAME }}:latest --format '{{ index .Config.Labels "org.opencontainers.image.base.hash" }}' 2>/dev/null) || true
+          fi
+
+          if [ -z "$REMOTE_HASH" ] || [ "$REMOTE_HASH" == "<no value>" ]; then
+            # No remote image or no hash label, need to rebuild
+            echo "No remote base image or hash label found, need rebuild"
+            echo "base_changed=true" >> $GITHUB_OUTPUT
+          elif [ "$CURRENT_HASH" != "$REMOTE_HASH" ]; then
+            echo "Hash mismatch: remote=$REMOTE_HASH, current=$CURRENT_HASH"
             echo "base_changed=true" >> $GITHUB_OUTPUT
           else
+            echo "Hash matches, no rebuild needed"
             echo "base_changed=false" >> $GITHUB_OUTPUT
           fi
 
@@ -61,6 +84,12 @@ jobs:
           username: ${{ github.actor }}
           password: ${{ secrets.GITHUB_TOKEN }}
 
+      - name: Calculate base files hash
+        id: hash
+        run: |
+          HASH=$(cat ${{ env.BASE_FILES }} 2>/dev/null | sha256sum | cut -d' ' -f1)
+          echo "hash=$HASH" >> $GITHUB_OUTPUT
+
       - name: Extract metadata for base image
         id: meta
         uses: docker/metadata-action@v5
@@ -69,6 +98,8 @@ jobs:
           tags: |
             type=raw,value=latest
             type=sha,prefix=
+          labels: |
+            org.opencontainers.image.base.hash=${{ steps.hash.outputs.hash }}
 
       - name: Build and push base image
         uses: docker/build-push-action@v5
@@ -117,7 +148,7 @@ jobs:
 
       - name: Update Dockerfile.app to use registry base image
         run: |
-          sed -i "s|FROM aether-base:latest|FROM ${{ env.REGISTRY }}/${{ env.BASE_IMAGE_NAME }}:latest|g" Dockerfile.app
+          sed -i "s|FROM aether-base:latest AS builder|FROM ${{ env.REGISTRY }}/${{ env.BASE_IMAGE_NAME }}:latest AS builder|g" Dockerfile.app
 
       - name: Build and push app image
         uses: docker/build-push-action@v5

Dockerfile.app

@@ -1,16 +1,134 @@
-# 应用镜像：基于基础镜像，只复制代码（秒级构建）
+# 运行镜像：从 base 提取产物到精简运行时
 # 构建命令: docker build -f Dockerfile.app -t aether-app:latest .
-FROM aether-base:latest
+# 用于 GitHub Actions CI（官方源）
+FROM aether-base:latest AS builder
 
 WORKDIR /app
 
+# 复制前端源码并构建
+COPY frontend/ ./frontend/
+RUN cd frontend && npm run build
+
+# ==================== 运行时镜像 ====================
+FROM python:3.12-slim
+
+WORKDIR /app
+
+# 运行时依赖（无 gcc/nodejs/npm）
+RUN apt-get update && apt-get install -y \
+    nginx \
+    supervisor \
+    libpq5 \
+    curl \
+    && rm -rf /var/lib/apt/lists/*
+
+# 从 base 镜像复制 Python 包
+COPY --from=builder /usr/local/lib/python3.12/site-packages /usr/local/lib/python3.12/site-packages
+
+# 只复制需要的 Python 可执行文件
+COPY --from=builder /usr/local/bin/gunicorn /usr/local/bin/
+COPY --from=builder /usr/local/bin/uvicorn /usr/local/bin/
+COPY --from=builder /usr/local/bin/alembic /usr/local/bin/
+
+# 从 builder 阶段复制前端构建产物
+COPY --from=builder /app/frontend/dist /usr/share/nginx/html
+
 # 复制后端代码
 COPY src/ ./src/
 COPY alembic.ini ./
 COPY alembic/ ./alembic/
 
-# 构建前端（使用基础镜像中已安装的 node_modules）
-COPY frontend/ /tmp/frontend/
-RUN cd /tmp/frontend && npm run build && \
-    cp -r dist/* /usr/share/nginx/html/ && \
-    rm -rf /tmp/frontend
+# Nginx 配置模板
+RUN printf '%s\n' \
+'server {' \
+'    listen 80;' \
+'    server_name _;' \
+'    root /usr/share/nginx/html;' \
+'    index index.html;' \
+'    client_max_body_size 100M;' \
+'' \
+'    location ~* \.(js|css|png|jpg|jpeg|gif|ico|svg|woff|woff2|ttf|eot)$ {' \
+'        expires 1y;' \
+'        add_header Cache-Control "public, no-transform";' \
+'        try_files $uri =404;' \
+'    }' \
+'' \
+'    location ~ ^/(src|node_modules)/ {' \
+'        deny all;' \
+'        return 404;' \
+'    }' \
+'' \
+'    location ~ ^/(dashboard|admin|login)(/|$) {' \
+'        try_files $uri $uri/ /index.html;' \
+'    }' \
+'' \
+'    location / {' \
+'        try_files $uri $uri/ @backend;' \
+'    }' \
+'' \
+'    location @backend {' \
+'        proxy_pass http://127.0.0.1:PORT_PLACEHOLDER;' \
+'        proxy_http_version 1.1;' \
+'        proxy_set_header Host $host;' \
+'        proxy_set_header X-Real-IP $remote_addr;' \
+'        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;' \
+'        proxy_set_header X-Forwarded-Proto $scheme;' \
+'        proxy_set_header Connection "";' \
+'        proxy_set_header Accept $http_accept;' \
+'        proxy_set_header Content-Type $content_type;' \
+'        proxy_set_header Authorization $http_authorization;' \
+'        proxy_set_header X-Api-Key $http_x_api_key;' \
+'        proxy_buffering off;' \
+'        proxy_cache off;' \
+'        proxy_request_buffering off;' \
+'        chunked_transfer_encoding on;' \
+'        gzip off;' \
+'        add_header X-Accel-Buffering no;' \
+'        proxy_connect_timeout 600s;' \
+'        proxy_send_timeout 600s;' \
+'        proxy_read_timeout 600s;' \
+'    }' \
+'}' > /etc/nginx/sites-available/default.template
+
+# Supervisor 配置
+RUN printf '%s\n' \
+'[supervisord]' \
+'nodaemon=true' \
+'logfile=/var/log/supervisor/supervisord.log' \
+'pidfile=/var/run/supervisord.pid' \
+'' \
+'[program:nginx]' \
+'command=/bin/bash -c "sed \"s/PORT_PLACEHOLDER/${PORT:-8084}/g\" /etc/nginx/sites-available/default.template > /etc/nginx/sites-available/default && /usr/sbin/nginx -g \"daemon off;\""' \
+'autostart=true' \
+'autorestart=true' \
+'stdout_logfile=/var/log/nginx/access.log' \
+'stderr_logfile=/var/log/nginx/error.log' \
+'' \
+'[program:app]' \
+'command=gunicorn src.main:app -w %(ENV_GUNICORN_WORKERS)s -k uvicorn.workers.UvicornWorker --bind 0.0.0.0:%(ENV_PORT)s --timeout 120 --access-logfile - --error-logfile - --log-level info' \
+'directory=/app' \
+'autostart=true' \
+'autorestart=true' \
+'stdout_logfile=/dev/stdout' \
+'stdout_logfile_maxbytes=0' \
+'stderr_logfile=/dev/stderr' \
+'stderr_logfile_maxbytes=0' \
+'environment=PYTHONUNBUFFERED=1,PYTHONIOENCODING=utf-8,LANG=C.UTF-8,LC_ALL=C.UTF-8,DOCKER_CONTAINER=true' > /etc/supervisor/conf.d/supervisord.conf
+
+# 创建目录
+RUN mkdir -p /var/log/supervisor /app/logs /app/data
+
+# 环境变量
+ENV PYTHONUNBUFFERED=1 \
+    PYTHONDONTWRITEBYTECODE=1 \
+    PYTHONIOENCODING=utf-8 \
+    LANG=C.UTF-8 \
+    LC_ALL=C.UTF-8 \
+    PORT=8084
+
+EXPOSE 80
+
+HEALTHCHECK --interval=30s --timeout=10s --start-period=5s --retries=3 \
+    CMD curl -f http://localhost/health || exit 1
+
+CMD ["/usr/bin/supervisord", "-c", "/etc/supervisor/conf.d/supervisord.conf"]

Dockerfile.app.local

@@ -0,0 +1,135 @@
+# 运行镜像：从 base 提取产物到精简运行时（国内镜像源版本）
+# 构建命令: docker build -f Dockerfile.app.local -t aether-app:latest .
+# 用于本地/国内服务器部署
+FROM aether-base:latest AS builder
+
+WORKDIR /app
+
+# 复制前端源码并构建
+COPY frontend/ ./frontend/
+RUN cd frontend && npm run build
+
+# ==================== 运行时镜像 ====================
+FROM python:3.12-slim
+
+WORKDIR /app
+
+# 运行时依赖（使用清华镜像源）
+RUN sed -i 's/deb.debian.org/mirrors.tuna.tsinghua.edu.cn/g' /etc/apt/sources.list.d/debian.sources && \
+    apt-get update && apt-get install -y \
+    nginx \
+    supervisor \
+    libpq5 \
+    curl \
+    && rm -rf /var/lib/apt/lists/*
+
+# 从 base 镜像复制 Python 包
+COPY --from=builder /usr/local/lib/python3.12/site-packages /usr/local/lib/python3.12/site-packages
+
+# 只复制需要的 Python 可执行文件
+COPY --from=builder /usr/local/bin/gunicorn /usr/local/bin/
+COPY --from=builder /usr/local/bin/uvicorn /usr/local/bin/
+COPY --from=builder /usr/local/bin/alembic /usr/local/bin/
+
+# 从 builder 阶段复制前端构建产物
+COPY --from=builder /app/frontend/dist /usr/share/nginx/html
+
+# 复制后端代码
+COPY src/ ./src/
+COPY alembic.ini ./
+COPY alembic/ ./alembic/
+
+# Nginx 配置模板
+RUN printf '%s\n' \
+'server {' \
+'    listen 80;' \
+'    server_name _;' \
+'    root /usr/share/nginx/html;' \
+'    index index.html;' \
+'    client_max_body_size 100M;' \
+'' \
+'    location ~* \.(js|css|png|jpg|jpeg|gif|ico|svg|woff|woff2|ttf|eot)$ {' \
+'        expires 1y;' \
+'        add_header Cache-Control "public, no-transform";' \
+'        try_files $uri =404;' \
+'    }' \
+'' \
+'    location ~ ^/(src|node_modules)/ {' \
+'        deny all;' \
+'        return 404;' \
+'    }' \
+'' \
+'    location ~ ^/(dashboard|admin|login)(/|$) {' \
+'        try_files $uri $uri/ /index.html;' \
+'    }' \
+'' \
+'    location / {' \
+'        try_files $uri $uri/ @backend;' \
+'    }' \
+'' \
+'    location @backend {' \
+'        proxy_pass http://127.0.0.1:PORT_PLACEHOLDER;' \
+'        proxy_http_version 1.1;' \
+'        proxy_set_header Host $host;' \
+'        proxy_set_header X-Real-IP $remote_addr;' \
+'        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;' \
+'        proxy_set_header X-Forwarded-Proto $scheme;' \
+'        proxy_set_header Connection "";' \
+'        proxy_set_header Accept $http_accept;' \
+'        proxy_set_header Content-Type $content_type;' \
+'        proxy_set_header Authorization $http_authorization;' \
+'        proxy_set_header X-Api-Key $http_x_api_key;' \
+'        proxy_buffering off;' \
+'        proxy_cache off;' \
+'        proxy_request_buffering off;' \
+'        chunked_transfer_encoding on;' \
+'        gzip off;' \
+'        add_header X-Accel-Buffering no;' \
+'        proxy_connect_timeout 600s;' \
+'        proxy_send_timeout 600s;' \
+'        proxy_read_timeout 600s;' \
+'    }' \
+'}' > /etc/nginx/sites-available/default.template
+
+# Supervisor 配置
+RUN printf '%s\n' \
+'[supervisord]' \
+'nodaemon=true' \
+'logfile=/var/log/supervisor/supervisord.log' \
+'pidfile=/var/run/supervisord.pid' \
+'' \
+'[program:nginx]' \
+'command=/bin/bash -c "sed \"s/PORT_PLACEHOLDER/${PORT:-8084}/g\" /etc/nginx/sites-available/default.template > /etc/nginx/sites-available/default && /usr/sbin/nginx -g \"daemon off;\""' \
+'autostart=true' \
+'autorestart=true' \
+'stdout_logfile=/var/log/nginx/access.log' \
+'stderr_logfile=/var/log/nginx/error.log' \
+'' \
+'[program:app]' \
+'command=gunicorn src.main:app -w %(ENV_GUNICORN_WORKERS)s -k uvicorn.workers.UvicornWorker --bind 0.0.0.0:%(ENV_PORT)s --timeout 120 --access-logfile - --error-logfile - --log-level info' \
+'directory=/app' \
+'autostart=true' \
+'autorestart=true' \
+'stdout_logfile=/dev/stdout' \
+'stdout_logfile_maxbytes=0' \
+'stderr_logfile=/dev/stderr' \
+'stderr_logfile_maxbytes=0' \
+'environment=PYTHONUNBUFFERED=1,PYTHONIOENCODING=utf-8,LANG=C.UTF-8,LC_ALL=C.UTF-8,DOCKER_CONTAINER=true' > /etc/supervisor/conf.d/supervisord.conf
+
+# 创建目录
+RUN mkdir -p /var/log/supervisor /app/logs /app/data
+
+# 环境变量
+ENV PYTHONUNBUFFERED=1 \
+    PYTHONDONTWRITEBYTECODE=1 \
+    PYTHONIOENCODING=utf-8 \
+    LANG=C.UTF-8 \
+    LC_ALL=C.UTF-8 \
+    PORT=8084
+
+EXPOSE 80
+
+HEALTHCHECK --interval=30s --timeout=10s --start-period=5s --retries=3 \
+    CMD curl -f http://localhost/health || exit 1
+
+CMD ["/usr/bin/supervisord", "-c", "/etc/supervisor/conf.d/supervisord.conf"]

Dockerfile.base

@@ -1,122 +1,25 @@
-# 基础镜像：包含所有依赖，只在依赖变化时需要重建
+# 构建镜像：编译环境 + 预编译的依赖
 # 用于 GitHub Actions CI 构建（不使用国内镜像源）
+# 构建命令: docker build -f Dockerfile.base -t aether-base:latest .
+# 只在 pyproject.toml 或 frontend/package*.json 变化时需要重建
 FROM python:3.12-slim
 
 WORKDIR /app
 
-# 系统依赖
+# 构建工具
 RUN apt-get update && apt-get install -y \
-    nginx \
-    supervisor \
     libpq-dev \
     gcc \
-    curl \
-    gettext-base \
     nodejs \
     npm \
     && rm -rf /var/lib/apt/lists/*
 
-# Python 依赖（安装到系统，不用 -e 模式）
+# Python 依赖
 COPY pyproject.toml README.md ./
 RUN mkdir -p src && touch src/__init__.py && \
-    pip install --no-cache-dir .
+    SETUPTOOLS_SCM_PRETEND_VERSION=0.1.0 pip install --no-cache-dir . && \
+    pip cache purge
 
-# 前端依赖
-COPY frontend/package*.json /tmp/frontend/
-WORKDIR /tmp/frontend
-RUN npm ci
-
-# Nginx 配置模板
-RUN printf '%s\n' \
-'server {' \
-'    listen 80;' \
-'    server_name _;' \
-'    root /usr/share/nginx/html;' \
-'    index index.html;' \
-'    client_max_body_size 100M;' \
-'' \
-'    location ~* \.(js|css|png|jpg|jpeg|gif|ico|svg|woff|woff2|ttf|eot)$ {' \
-'        expires 1y;' \
-'        add_header Cache-Control "public, no-transform";' \
-'        try_files $uri =404;' \
-'    }' \
-'' \
-'    location ~ ^/(src|node_modules)/ {' \
-'        deny all;' \
-'        return 404;' \
-'    }' \
-'' \
-'    location ~ ^/(dashboard|admin|login)(/|$) {' \
-'        try_files $uri $uri/ /index.html;' \
-'    }' \
-'' \
-'    location / {' \
-'        try_files $uri $uri/ @backend;' \
-'    }' \
-'' \
-'    location @backend {' \
-'        proxy_pass http://127.0.0.1:PORT_PLACEHOLDER;' \
-'        proxy_http_version 1.1;' \
-'        proxy_set_header Host $host;' \
-'        proxy_set_header X-Real-IP $remote_addr;' \
-'        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;' \
-'        proxy_set_header X-Forwarded-Proto $scheme;' \
-'        proxy_set_header Connection "";' \
-'        proxy_set_header Accept $http_accept;' \
-'        proxy_set_header Content-Type $content_type;' \
-'        proxy_set_header Authorization $http_authorization;' \
-'        proxy_set_header X-Api-Key $http_x_api_key;' \
-'        proxy_buffering off;' \
-'        proxy_cache off;' \
-'        proxy_request_buffering off;' \
-'        chunked_transfer_encoding on;' \
-'        proxy_connect_timeout 600s;' \
-'        proxy_send_timeout 600s;' \
-'        proxy_read_timeout 600s;' \
-'    }' \
-'}' > /etc/nginx/sites-available/default.template
-
-# Supervisor 配置
-RUN printf '%s\n' \
-'[supervisord]' \
-'nodaemon=true' \
-'logfile=/var/log/supervisor/supervisord.log' \
-'pidfile=/var/run/supervisord.pid' \
-'' \
-'[program:nginx]' \
-'command=/bin/bash -c "sed \"s/PORT_PLACEHOLDER/${PORT:-8084}/g\" /etc/nginx/sites-available/default.template > /etc/nginx/sites-available/default && /usr/sbin/nginx -g \"daemon off;\""' \
-'autostart=true' \
-'autorestart=true' \
-'stdout_logfile=/var/log/nginx/access.log' \
-'stderr_logfile=/var/log/nginx/error.log' \
-'' \
-'[program:app]' \
-'command=gunicorn src.main:app -w %(ENV_GUNICORN_WORKERS)s -k uvicorn.workers.UvicornWorker --bind 0.0.0.0:%(ENV_PORT)s --timeout 120 --access-logfile - --error-logfile - --log-level info' \
-'directory=/app' \
-'autostart=true' \
-'autorestart=true' \
-'stdout_logfile=/dev/stdout' \
-'stdout_logfile_maxbytes=0' \
-'stderr_logfile=/dev/stderr' \
-'stderr_logfile_maxbytes=0' \
-'environment=PYTHONUNBUFFERED=1,PYTHONIOENCODING=utf-8,LANG=C.UTF-8,LC_ALL=C.UTF-8,DOCKER_CONTAINER=true' > /etc/supervisor/conf.d/supervisord.conf
-
-# 创建目录
-RUN mkdir -p /var/log/supervisor /app/logs /app/data /usr/share/nginx/html
-
-WORKDIR /app
-
-# 环境变量
-ENV PYTHONUNBUFFERED=1 \
-    PYTHONDONTWRITEBYTECODE=1 \
-    PYTHONIOENCODING=utf-8 \
-    LANG=C.UTF-8 \
-    LC_ALL=C.UTF-8 \
-    PORT=8084
-
-EXPOSE 80
-
-HEALTHCHECK --interval=30s --timeout=10s --start-period=5s --retries=3 \
-    CMD curl -f http://localhost/health || exit 1
-
-CMD ["/usr/bin/supervisord", "-c", "/etc/supervisor/conf.d/supervisord.conf"]
+# 前端依赖（只安装，不构建）
+COPY frontend/package*.json ./frontend/
+RUN cd frontend && npm ci

Dockerfile.base.local

@@ -1,18 +1,15 @@
-# 基础镜像：包含所有依赖，只在依赖变化时需要重建
-# 构建命令: docker build -f Dockerfile.base -t aether-base:latest .
+# 构建镜像：编译环境 + 预编译的依赖（国内镜像源版本）
+# 构建命令: docker build -f Dockerfile.base.local -t aether-base:latest .
+# 只在 pyproject.toml 或 frontend/package*.json 变化时需要重建
 FROM python:3.12-slim
 
 WORKDIR /app
 
-# 系统依赖
+# 构建工具（使用清华镜像源）
 RUN sed -i 's/deb.debian.org/mirrors.tuna.tsinghua.edu.cn/g' /etc/apt/sources.list.d/debian.sources && \
     apt-get update && apt-get install -y \
-    nginx \
-    supervisor \
     libpq-dev \
     gcc \
-    curl \
-    gettext-base \
     nodejs \
     npm \
     && rm -rf /var/lib/apt/lists/*
@@ -20,107 +17,12 @@ RUN sed -i 's/deb.debian.org/mirrors.tuna.tsinghua.edu.cn/g' /etc/apt/sources.li
 # pip 镜像源
 RUN pip config set global.index-url https://pypi.tuna.tsinghua.edu.cn/simple
 
-# Python 依赖（安装到系统，不用 -e 模式）
+# Python 依赖
 COPY pyproject.toml README.md ./
 RUN mkdir -p src && touch src/__init__.py && \
-    pip install --no-cache-dir .
+    SETUPTOOLS_SCM_PRETEND_VERSION=0.1.0 pip install --no-cache-dir . && \
+    pip cache purge
 
-# 前端依赖
-COPY frontend/package*.json /tmp/frontend/
-WORKDIR /tmp/frontend
-RUN npm config set registry https://registry.npmmirror.com && npm ci
-
-# Nginx 配置模板
-RUN printf '%s\n' \
-'server {' \
-'    listen 80;' \
-'    server_name _;' \
-'    root /usr/share/nginx/html;' \
-'    index index.html;' \
-'    client_max_body_size 100M;' \
-'' \
-'    location ~* \.(js|css|png|jpg|jpeg|gif|ico|svg|woff|woff2|ttf|eot)$ {' \
-'        expires 1y;' \
-'        add_header Cache-Control "public, no-transform";' \
-'        try_files $uri =404;' \
-'    }' \
-'' \
-'    location ~ ^/(src|node_modules)/ {' \
-'        deny all;' \
-'        return 404;' \
-'    }' \
-'' \
-'    location ~ ^/(dashboard|admin|login)(/|$) {' \
-'        try_files $uri $uri/ /index.html;' \
-'    }' \
-'' \
-'    location / {' \
-'        try_files $uri $uri/ @backend;' \
-'    }' \
-'' \
-'    location @backend {' \
-'        proxy_pass http://127.0.0.1:PORT_PLACEHOLDER;' \
-'        proxy_http_version 1.1;' \
-'        proxy_set_header Host $host;' \
-'        proxy_set_header X-Real-IP $remote_addr;' \
-'        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;' \
-'        proxy_set_header X-Forwarded-Proto $scheme;' \
-'        proxy_set_header Connection "";' \
-'        proxy_set_header Accept $http_accept;' \
-'        proxy_set_header Content-Type $content_type;' \
-'        proxy_set_header Authorization $http_authorization;' \
-'        proxy_set_header X-Api-Key $http_x_api_key;' \
-'        proxy_buffering off;' \
-'        proxy_cache off;' \
-'        proxy_request_buffering off;' \
-'        chunked_transfer_encoding on;' \
-'        proxy_connect_timeout 600s;' \
-'        proxy_send_timeout 600s;' \
-'        proxy_read_timeout 600s;' \
-'    }' \
-'}' > /etc/nginx/sites-available/default.template
-
-# Supervisor 配置
-RUN printf '%s\n' \
-'[supervisord]' \
-'nodaemon=true' \
-'logfile=/var/log/supervisor/supervisord.log' \
-'pidfile=/var/run/supervisord.pid' \
-'' \
-'[program:nginx]' \
-'command=/bin/bash -c "sed \"s/PORT_PLACEHOLDER/${PORT:-8084}/g\" /etc/nginx/sites-available/default.template > /etc/nginx/sites-available/default && /usr/sbin/nginx -g \"daemon off;\""' \
-'autostart=true' \
-'autorestart=true' \
-'stdout_logfile=/var/log/nginx/access.log' \
-'stderr_logfile=/var/log/nginx/error.log' \
-'' \
-'[program:app]' \
-'command=gunicorn src.main:app -w %(ENV_GUNICORN_WORKERS)s -k uvicorn.workers.UvicornWorker --bind 0.0.0.0:%(ENV_PORT)s --timeout 120 --access-logfile - --error-logfile - --log-level info' \
-'directory=/app' \
-'autostart=true' \
-'autorestart=true' \
-'stdout_logfile=/dev/stdout' \
-'stdout_logfile_maxbytes=0' \
-'stderr_logfile=/dev/stderr' \
-'stderr_logfile_maxbytes=0' \
-'environment=PYTHONUNBUFFERED=1,PYTHONIOENCODING=utf-8,LANG=C.UTF-8,LC_ALL=C.UTF-8,DOCKER_CONTAINER=true' > /etc/supervisor/conf.d/supervisord.conf
-
-# 创建目录
-RUN mkdir -p /var/log/supervisor /app/logs /app/data /usr/share/nginx/html
-
-WORKDIR /app
-
-# 环境变量
-ENV PYTHONUNBUFFERED=1 \
-    PYTHONDONTWRITEBYTECODE=1 \
-    PYTHONIOENCODING=utf-8 \
-    LANG=C.UTF-8 \
-    LC_ALL=C.UTF-8 \
-    PORT=8084
-
-EXPOSE 80
-
-HEALTHCHECK --interval=30s --timeout=10s --start-period=5s --retries=3 \
-    CMD curl -f http://localhost/health || exit 1
-
-CMD ["/usr/bin/supervisord", "-c", "/etc/supervisor/conf.d/supervisord.conf"]
+# 前端依赖（只安装，不构建，使用淘宝镜像源）
+COPY frontend/package*.json ./frontend/
+RUN cd frontend && npm config set registry https://registry.npmmirror.com && npm ci

README.md

@@ -60,8 +60,11 @@ python generate_keys.py  # 生成密钥, 并将生成的密钥填入 .env
 # 3. 部署
 docker-compose up -d
 
-# 4. 更新
-docker-compose pull && docker-compose up -d
+# 4. 首次部署时, 初始化数据库
+./migrate.sh
+
+# 5. 更新
+docker-compose pull && docker-compose up -d && ./migrate.sh
 ```
 
 ### Docker Compose（本地构建镜像）

alembic/versions/20251210_baseline.py

@@ -20,10 +20,10 @@ depends_on = None
 
 
 def upgrade() -> None:
-    # Create ENUM types
-    op.execute("CREATE TYPE userrole AS ENUM ('admin', 'user')")
+    # Create ENUM types (with IF NOT EXISTS for idempotency)
+    op.execute("DO $$ BEGIN CREATE TYPE userrole AS ENUM ('admin', 'user'); EXCEPTION WHEN duplicate_object THEN NULL; END $$")
     op.execute(
-        "CREATE TYPE providerbillingtype AS ENUM ('monthly_quota', 'pay_as_you_go', 'free_tier')"
+        "DO $$ BEGIN CREATE TYPE providerbillingtype AS ENUM ('monthly_quota', 'pay_as_you_go', 'free_tier'); EXCEPTION WHEN duplicate_object THEN NULL; END $$"
     )
 
     # ==================== users ====================
@@ -35,7 +35,7 @@ def upgrade() -> None:
         sa.Column("password_hash", sa.String(255), nullable=False),
         sa.Column(
             "role",
-            sa.Enum("admin", "user", name="userrole", create_type=False),
+            postgresql.ENUM("admin", "user", name="userrole", create_type=False),
             nullable=False,
             server_default="user",
         ),
@@ -67,7 +67,7 @@ def upgrade() -> None:
         sa.Column("website", sa.String(500), nullable=True),
         sa.Column(
             "billing_type",
-            sa.Enum(
+            postgresql.ENUM(
                 "monthly_quota", "pay_as_you_go", "free_tier", name="providerbillingtype", create_type=False
             ),
             nullable=False,

alembic/versions/20251214_1300_e9b3d63f0cbf_remove_model_mappings_add_aliases.py

@@ -0,0 +1,315 @@
+"""remove_model_mappings_add_aliases
+
+合并迁移：
+1. 添加 provider_model_aliases 字段到 models 表
+2. 迁移 model_mappings 数据到 provider_model_aliases
+3. 删除 model_mappings 表
+4. 添加索引优化别名解析性能
+
+Revision ID: e9b3d63f0cbf
+Revises: 20251210_baseline
+Create Date: 2025-12-14 13:00:22.828183+00:00
+
+"""
+import json
+from datetime import datetime, timezone
+
+import sqlalchemy as sa
+from alembic import op
+from sqlalchemy.orm import Session
+
+
+# revision identifiers, used by Alembic.
+revision = 'e9b3d63f0cbf'
+down_revision = '20251210_baseline'
+branch_labels = None
+depends_on = None
+
+
+def column_exists(bind, table_name: str, column_name: str) -> bool:
+    """检查列是否存在"""
+    result = bind.execute(
+        sa.text(
+            """
+            SELECT EXISTS (
+                SELECT 1 FROM information_schema.columns
+                WHERE table_name = :table_name AND column_name = :column_name
+            )
+            """
+        ),
+        {"table_name": table_name, "column_name": column_name},
+    )
+    return result.scalar()
+
+
+def table_exists(bind, table_name: str) -> bool:
+    """检查表是否存在"""
+    result = bind.execute(
+        sa.text(
+            """
+            SELECT EXISTS (
+                SELECT 1 FROM information_schema.tables
+                WHERE table_name = :table_name
+            )
+            """
+        ),
+        {"table_name": table_name},
+    )
+    return result.scalar()
+
+
+def index_exists(bind, index_name: str) -> bool:
+    """检查索引是否存在"""
+    result = bind.execute(
+        sa.text(
+            """
+            SELECT EXISTS (
+                SELECT 1 FROM pg_indexes
+                WHERE indexname = :index_name
+            )
+            """
+        ),
+        {"index_name": index_name},
+    )
+    return result.scalar()
+
+
+def upgrade() -> None:
+    """添加 provider_model_aliases 字段，迁移数据，删除 model_mappings 表"""
+    bind = op.get_bind()
+
+    # 1. 添加 provider_model_aliases 字段（如果不存在）
+    if not column_exists(bind, "models", "provider_model_aliases"):
+        op.add_column(
+            'models',
+            sa.Column('provider_model_aliases', sa.JSON(), nullable=True)
+        )
+
+    # 2. 迁移 model_mappings 数据（如果表存在）
+    session = Session(bind=bind)
+
+    model_mappings_table = sa.table(
+        "model_mappings",
+        sa.column("source_model", sa.String),
+        sa.column("target_global_model_id", sa.String),
+        sa.column("provider_id", sa.String),
+        sa.column("mapping_type", sa.String),
+        sa.column("is_active", sa.Boolean),
+    )
+
+    models_table = sa.table(
+        "models",
+        sa.column("id", sa.String),
+        sa.column("provider_id", sa.String),
+        sa.column("global_model_id", sa.String),
+        sa.column("provider_model_aliases", sa.JSON),
+        sa.column("updated_at", sa.DateTime(timezone=True)),
+    )
+
+    def normalize_alias_list(value) -> list[dict]:
+        """将 DB 返回的 JSON 值规范化为 list[{'name': str, 'priority': int}]"""
+        if value is None:
+            return []
+
+        if isinstance(value, str):
+            try:
+                value = json.loads(value) if value else []
+            except Exception:
+                return []
+
+        if not isinstance(value, list):
+            return []
+
+        normalized: list[dict] = []
+        for item in value:
+            if not isinstance(item, dict):
+                continue
+
+            raw_name = item.get("name")
+            if not isinstance(raw_name, str):
+                continue
+            name = raw_name.strip()
+            if not name:
+                continue
+
+            raw_priority = item.get("priority", 1)
+            try:
+                priority = int(raw_priority)
+            except Exception:
+                priority = 1
+            if priority < 1:
+                priority = 1
+
+            normalized.append({"name": name, "priority": priority})
+
+        return normalized
+
+    # 查询所有活跃的 provider 级别 alias（只迁移 is_active=True 且 mapping_type='alias' 的）
+    # 全局别名/映射不迁移（新架构不再支持 source_model -> GlobalModel.name 的解析）
+    # 仅当 model_mappings 表存在时执行迁移
+    if table_exists(bind, "model_mappings"):
+        mappings = session.execute(
+            sa.select(
+                model_mappings_table.c.source_model,
+                model_mappings_table.c.target_global_model_id,
+                model_mappings_table.c.provider_id,
+            )
+            .where(
+                model_mappings_table.c.is_active.is_(True),
+                model_mappings_table.c.provider_id.isnot(None),
+                model_mappings_table.c.mapping_type == "alias",
+            )
+            .order_by(model_mappings_table.c.provider_id, model_mappings_table.c.source_model)
+        ).all()
+
+        # 按 (provider_id, target_global_model_id) 分组，收集别名
+        alias_groups: dict = {}
+        for source_model, target_global_model_id, provider_id in mappings:
+            if not isinstance(source_model, str):
+                continue
+            source_model = source_model.strip()
+            if not source_model:
+                continue
+            if not isinstance(provider_id, str) or not provider_id:
+                continue
+            if not isinstance(target_global_model_id, str) or not target_global_model_id:
+                continue
+
+            key = (provider_id, target_global_model_id)
+            if key not in alias_groups:
+                alias_groups[key] = []
+            priority = len(alias_groups[key]) + 1
+            alias_groups[key].append({"name": source_model, "priority": priority})
+
+        # 更新对应的 models 记录
+        for (provider_id, global_model_id), aliases in alias_groups.items():
+            model_row = session.execute(
+                sa.select(models_table.c.id, models_table.c.provider_model_aliases)
+                .where(
+                    models_table.c.provider_id == provider_id,
+                    models_table.c.global_model_id == global_model_id,
+                )
+                .limit(1)
+            ).first()
+
+            if model_row:
+                model_id = model_row[0]
+                existing_aliases = normalize_alias_list(model_row[1])
+
+                existing_names = {a["name"] for a in existing_aliases}
+                merged_aliases = list(existing_aliases)
+                for alias in aliases:
+                    name = alias.get("name")
+                    if not isinstance(name, str):
+                        continue
+                    name = name.strip()
+                    if not name or name in existing_names:
+                        continue
+
+                    merged_aliases.append(
+                        {
+                            "name": name,
+                            "priority": len(merged_aliases) + 1,
+                        }
+                    )
+                    existing_names.add(name)
+
+                session.execute(
+                    models_table.update()
+                    .where(models_table.c.id == model_id)
+                    .values(
+                        provider_model_aliases=merged_aliases if merged_aliases else None,
+                        updated_at=datetime.now(timezone.utc),
+                    )
+                )
+
+        session.commit()
+
+        # 3. 删除 model_mappings 表
+        op.drop_table('model_mappings')
+
+    # 4. 添加索引优化别名解析性能
+    # provider_model_name 索引（支持精确匹配，如果不存在）
+    if not index_exists(bind, "idx_model_provider_model_name"):
+        op.create_index(
+            "idx_model_provider_model_name",
+            "models",
+            ["provider_model_name"],
+            unique=False,
+            postgresql_where=sa.text("is_active = true"),
+        )
+
+    # provider_model_aliases GIN 索引（支持 JSONB 查询，仅 PostgreSQL）
+    if bind.dialect.name == "postgresql":
+        # 将 json 列转为 jsonb（jsonb 性能更好且支持 GIN 索引）
+        # 使用 IF NOT EXISTS 风格的检查来避免重复转换
+        op.execute(
+            """
+            DO $$
+            BEGIN
+                IF EXISTS (
+                    SELECT 1 FROM information_schema.columns
+                    WHERE table_name = 'models'
+                    AND column_name = 'provider_model_aliases'
+                    AND data_type = 'json'
+                ) THEN
+                    ALTER TABLE models
+                    ALTER COLUMN provider_model_aliases TYPE jsonb
+                    USING provider_model_aliases::jsonb;
+                END IF;
+            END $$;
+            """
+        )
+        # 创建 GIN 索引
+        op.execute(
+            """
+            CREATE INDEX IF NOT EXISTS idx_model_provider_model_aliases_gin
+            ON models USING gin(provider_model_aliases jsonb_path_ops)
+            WHERE is_active = true
+            """
+        )
+
+
+def downgrade() -> None:
+    """恢复 model_mappings 表，移除 provider_model_aliases 字段和索引"""
+    bind = op.get_bind()
+
+    # 1. 删除索引
+    op.drop_index("idx_model_provider_model_name", table_name="models")
+
+    if bind.dialect.name == "postgresql":
+        op.execute("DROP INDEX IF EXISTS idx_model_provider_model_aliases_gin")
+        # 将 jsonb 列还原为 json
+        op.execute(
+            """
+            ALTER TABLE models
+            ALTER COLUMN provider_model_aliases TYPE json
+            USING provider_model_aliases::json
+            """
+        )
+
+    # 2. 恢复 model_mappings 表
+    op.create_table(
+        'model_mappings',
+        sa.Column('id', sa.String(36), primary_key=True),
+        sa.Column('source_model', sa.String(200), nullable=False),
+        sa.Column(
+            'target_global_model_id',
+            sa.String(36),
+            sa.ForeignKey('global_models.id', ondelete='CASCADE'),
+            nullable=False,
+        ),
+        sa.Column('provider_id', sa.String(36), sa.ForeignKey('providers.id'), nullable=True),
+        sa.Column('mapping_type', sa.String(20), nullable=False, server_default='alias'),
+        sa.Column('is_active', sa.Boolean(), nullable=False, server_default='true'),
+        sa.Column('created_at', sa.DateTime(timezone=True), server_default=sa.func.now(), nullable=False),
+        sa.Column('updated_at', sa.DateTime(timezone=True), server_default=sa.func.now(), nullable=False),
+        sa.UniqueConstraint('source_model', 'provider_id', name='uq_model_mapping_source_provider'),
+    )
+    op.create_index('ix_model_mappings_source_model', 'model_mappings', ['source_model'])
+    op.create_index('ix_model_mappings_target_global_model_id', 'model_mappings', ['target_global_model_id'])
+    op.create_index('ix_model_mappings_provider_id', 'model_mappings', ['provider_id'])
+    op.create_index('ix_model_mappings_mapping_type', 'model_mappings', ['mapping_type'])
+
+    # 3. 移除 provider_model_aliases 字段
+    op.drop_column('models', 'provider_model_aliases')

alembic/versions/20251215_1707_180e63a9c83a_add_first_byte_time_ms_to_usage_table.py

@@ -0,0 +1,47 @@
+"""add first_byte_time_ms to usage table
+
+Revision ID: 180e63a9c83a
+Revises: e9b3d63f0cbf
+Create Date: 2025-12-15 17:07:44.631032+00:00
+
+"""
+import sqlalchemy as sa
+from alembic import op
+
+
+# revision identifiers, used by Alembic.
+revision = '180e63a9c83a'
+down_revision = 'e9b3d63f0cbf'
+branch_labels = None
+depends_on = None
+
+
+def column_exists(bind, table_name: str, column_name: str) -> bool:
+    """检查列是否存在"""
+    result = bind.execute(
+        sa.text(
+            """
+            SELECT EXISTS (
+                SELECT 1 FROM information_schema.columns
+                WHERE table_name = :table_name AND column_name = :column_name
+            )
+            """
+        ),
+        {"table_name": table_name, "column_name": column_name},
+    )
+    return result.scalar()
+
+
+def upgrade() -> None:
+    """应用迁移：升级到新版本"""
+    bind = op.get_bind()
+
+    # 添加首字时间字段到 usage 表（如果不存在）
+    if not column_exists(bind, "usage", "first_byte_time_ms"):
+        op.add_column('usage', sa.Column('first_byte_time_ms', sa.Integer(), nullable=True))
+
+
+def downgrade() -> None:
+    """回滚迁移：降级到旧版本"""
+    # 删除首字时间字段
+    op.drop_column('usage', 'first_byte_time_ms')

alembic/versions/20251216_0311_1cc6942cf06f_refactor_global_model_to_use_config_.py

@@ -0,0 +1,110 @@
+"""refactor global_model to use config json field
+
+Revision ID: 1cc6942cf06f
+Revises: 180e63a9c83a
+Create Date: 2025-12-16 03:11:32.480976+00:00
+
+"""
+import sqlalchemy as sa
+from alembic import op
+from sqlalchemy.dialects import postgresql
+
+# revision identifiers, used by Alembic.
+revision = '1cc6942cf06f'
+down_revision = '180e63a9c83a'
+branch_labels = None
+depends_on = None
+
+
+def column_exists(bind, table_name: str, column_name: str) -> bool:
+    """检查列是否存在"""
+    result = bind.execute(
+        sa.text(
+            """
+            SELECT EXISTS (
+                SELECT 1 FROM information_schema.columns
+                WHERE table_name = :table_name AND column_name = :column_name
+            )
+            """
+        ),
+        {"table_name": table_name, "column_name": column_name},
+    )
+    return result.scalar()
+
+
+def upgrade() -> None:
+    """应用迁移：升级到新版本
+
+    1. 添加 config 列
+    2. 把旧数据迁移到 config
+    3. 删除旧列
+    """
+    bind = op.get_bind()
+
+    # 检查是否已经迁移过（config 列存在且旧列不存在）
+    has_config = column_exists(bind, "global_models", "config")
+    has_old_columns = column_exists(bind, "global_models", "default_supports_streaming")
+
+    if has_config and not has_old_columns:
+        # 已完成迁移，跳过
+        return
+
+    # 1. 添加 config 列（使用 JSONB 类型，支持索引和更高效的查询）
+    if not has_config:
+        op.add_column('global_models', sa.Column('config', postgresql.JSONB(), nullable=True))
+
+    # 2. 迁移数据：把旧字段合并到 config JSON（仅当旧列存在时）
+    if has_old_columns:
+        op.execute("""
+            UPDATE global_models
+            SET config = jsonb_strip_nulls(jsonb_build_object(
+                'streaming', COALESCE(default_supports_streaming, true),
+                'vision', CASE WHEN COALESCE(default_supports_vision, false) THEN true ELSE NULL END,
+                'function_calling', CASE WHEN COALESCE(default_supports_function_calling, false) THEN true ELSE NULL END,
+                'extended_thinking', CASE WHEN COALESCE(default_supports_extended_thinking, false) THEN true ELSE NULL END,
+                'image_generation', CASE WHEN COALESCE(default_supports_image_generation, false) THEN true ELSE NULL END,
+                'description', description,
+                'icon_url', icon_url,
+                'official_url', official_url
+            ))
+        """)
+
+        # 3. 删除旧列
+        op.drop_column('global_models', 'default_supports_streaming')
+        op.drop_column('global_models', 'default_supports_vision')
+        op.drop_column('global_models', 'default_supports_function_calling')
+        op.drop_column('global_models', 'default_supports_extended_thinking')
+        op.drop_column('global_models', 'default_supports_image_generation')
+        op.drop_column('global_models', 'description')
+        op.drop_column('global_models', 'icon_url')
+        op.drop_column('global_models', 'official_url')
+
+
+def downgrade() -> None:
+    """回滚迁移：降级到旧版本"""
+    # 1. 添加旧列
+    op.add_column('global_models', sa.Column('icon_url', sa.VARCHAR(length=500), nullable=True))
+    op.add_column('global_models', sa.Column('official_url', sa.VARCHAR(length=500), nullable=True))
+    op.add_column('global_models', sa.Column('description', sa.TEXT(), nullable=True))
+    op.add_column('global_models', sa.Column('default_supports_streaming', sa.BOOLEAN(), nullable=True))
+    op.add_column('global_models', sa.Column('default_supports_vision', sa.BOOLEAN(), nullable=True))
+    op.add_column('global_models', sa.Column('default_supports_function_calling', sa.BOOLEAN(), nullable=True))
+    op.add_column('global_models', sa.Column('default_supports_extended_thinking', sa.BOOLEAN(), nullable=True))
+    op.add_column('global_models', sa.Column('default_supports_image_generation', sa.BOOLEAN(), nullable=True))
+
+    # 2. 从 config 恢复数据
+    op.execute("""
+        UPDATE global_models
+        SET
+            default_supports_streaming = COALESCE((config->>'streaming')::boolean, true),
+            default_supports_vision = COALESCE((config->>'vision')::boolean, false),
+            default_supports_function_calling = COALESCE((config->>'function_calling')::boolean, false),
+            default_supports_extended_thinking = COALESCE((config->>'extended_thinking')::boolean, false),
+            default_supports_image_generation = COALESCE((config->>'image_generation')::boolean, false),
+            description = config->>'description',
+            icon_url = config->>'icon_url',
+            official_url = config->>'official_url'
+    """)
+
+    # 3. 删除 config 列
+    op.drop_column('global_models', 'config')

alembic/versions/20251218_0631_f30f9936f6a2_add_proxy_field_to_provider_endpoints.py

@@ -0,0 +1,57 @@
+"""add proxy field to provider_endpoints
+
+Revision ID: f30f9936f6a2
+Revises: 1cc6942cf06f
+Create Date: 2025-12-18 06:31:58.451112+00:00
+
+"""
+from alembic import op
+import sqlalchemy as sa
+from sqlalchemy.dialects.postgresql import JSONB
+from sqlalchemy import inspect
+
+# revision identifiers, used by Alembic.
+revision = 'f30f9936f6a2'
+down_revision = '1cc6942cf06f'
+branch_labels = None
+depends_on = None
+
+
+def column_exists(table_name: str, column_name: str) -> bool:
+    """检查列是否存在"""
+    bind = op.get_bind()
+    inspector = inspect(bind)
+    columns = [col['name'] for col in inspector.get_columns(table_name)]
+    return column_name in columns
+
+
+def get_column_type(table_name: str, column_name: str) -> str:
+    """获取列的类型"""
+    bind = op.get_bind()
+    inspector = inspect(bind)
+    for col in inspector.get_columns(table_name):
+        if col['name'] == column_name:
+            return str(col['type']).upper()
+    return ''
+
+
+def upgrade() -> None:
+    """添加 proxy 字段到 provider_endpoints 表"""
+    if not column_exists('provider_endpoints', 'proxy'):
+        # 字段不存在，直接添加 JSONB 类型
+        op.add_column('provider_endpoints', sa.Column('proxy', JSONB(), nullable=True))
+    else:
+        # 字段已存在，检查是否需要转换类型
+        col_type = get_column_type('provider_endpoints', 'proxy')
+        if 'JSONB' not in col_type:
+            # 如果是 JSON 类型，转换为 JSONB
+            op.execute(
+                'ALTER TABLE provider_endpoints '
+                'ALTER COLUMN proxy TYPE JSONB USING proxy::jsonb'
+            )
+
+
+def downgrade() -> None:
+    """移除 proxy 字段"""
+    if column_exists('provider_endpoints', 'proxy'):
+        op.drop_column('provider_endpoints', 'proxy')

alembic/versions/20251220_1200_add_stats_daily_model_table.py

@@ -0,0 +1,86 @@
+"""add stats_daily_model table and rename provider_model_aliases
+
+Revision ID: a1b2c3d4e5f6
+Revises: f30f9936f6a2
+Create Date: 2025-12-20 12:00:00.000000+00:00
+
+"""
+from alembic import op
+import sqlalchemy as sa
+from sqlalchemy import inspect
+
+# revision identifiers, used by Alembic.
+revision = 'a1b2c3d4e5f6'
+down_revision = 'f30f9936f6a2'
+branch_labels = None
+depends_on = None
+
+
+def table_exists(table_name: str) -> bool:
+    """检查表是否存在"""
+    bind = op.get_bind()
+    inspector = inspect(bind)
+    return table_name in inspector.get_table_names()
+
+
+def column_exists(table_name: str, column_name: str) -> bool:
+    """检查列是否存在"""
+    bind = op.get_bind()
+    inspector = inspect(bind)
+    columns = [col['name'] for col in inspector.get_columns(table_name)]
+    return column_name in columns
+
+
+def upgrade() -> None:
+    """创建 stats_daily_model 表，重命名 provider_model_aliases 为 provider_model_mappings"""
+    # 1. 创建 stats_daily_model 表
+    if not table_exists('stats_daily_model'):
+        op.create_table(
+            'stats_daily_model',
+            sa.Column('id', sa.String(36), primary_key=True),
+            sa.Column('date', sa.DateTime(timezone=True), nullable=False),
+            sa.Column('model', sa.String(100), nullable=False),
+            sa.Column('total_requests', sa.Integer(), nullable=False, default=0),
+            sa.Column('input_tokens', sa.BigInteger(), nullable=False, default=0),
+            sa.Column('output_tokens', sa.BigInteger(), nullable=False, default=0),
+            sa.Column('cache_creation_tokens', sa.BigInteger(), nullable=False, default=0),
+            sa.Column('cache_read_tokens', sa.BigInteger(), nullable=False, default=0),
+            sa.Column('total_cost', sa.Float(), nullable=False, default=0.0),
+            sa.Column('avg_response_time_ms', sa.Float(), nullable=False, default=0.0),
+            sa.Column('created_at', sa.DateTime(timezone=True), nullable=False,
+                      server_default=sa.func.now()),
+            sa.Column('updated_at', sa.DateTime(timezone=True), nullable=False,
+                      server_default=sa.func.now(), onupdate=sa.func.now()),
+            sa.UniqueConstraint('date', 'model', name='uq_stats_daily_model'),
+        )
+
+        # 创建索引
+        op.create_index('idx_stats_daily_model_date', 'stats_daily_model', ['date'])
+        op.create_index('idx_stats_daily_model_date_model', 'stats_daily_model', ['date', 'model'])
+
+    # 2. 重命名 models 表的 provider_model_aliases 为 provider_model_mappings
+    if column_exists('models', 'provider_model_aliases') and not column_exists('models', 'provider_model_mappings'):
+        op.alter_column('models', 'provider_model_aliases', new_column_name='provider_model_mappings')
+
+
+def index_exists(table_name: str, index_name: str) -> bool:
+    """检查索引是否存在"""
+    bind = op.get_bind()
+    inspector = inspect(bind)
+    indexes = [idx['name'] for idx in inspector.get_indexes(table_name)]
+    return index_name in indexes
+
+
+def downgrade() -> None:
+    """删除 stats_daily_model 表，恢复 provider_model_aliases 列名"""
+    # 恢复列名
+    if column_exists('models', 'provider_model_mappings') and not column_exists('models', 'provider_model_aliases'):
+        op.alter_column('models', 'provider_model_mappings', new_column_name='provider_model_aliases')
+
+    # 删除表
+    if table_exists('stats_daily_model'):
+        if index_exists('stats_daily_model', 'idx_stats_daily_model_date_model'):
+            op.drop_index('idx_stats_daily_model_date_model', table_name='stats_daily_model')
+        if index_exists('stats_daily_model', 'idx_stats_daily_model_date'):
+            op.drop_index('idx_stats_daily_model_date', table_name='stats_daily_model')
+        op.drop_table('stats_daily_model')

deploy.sh

@@ -21,9 +21,9 @@ HASH_FILE=".deps-hash"
 CODE_HASH_FILE=".code-hash"
 MIGRATION_HASH_FILE=".migration-hash"
 
-# 计算依赖文件的哈希值
+# 计算依赖文件的哈希值（包含 Dockerfile.base.local）
 calc_deps_hash() {
-    cat pyproject.toml frontend/package.json frontend/package-lock.json 2>/dev/null | md5sum | cut -d' ' -f1
+    cat pyproject.toml frontend/package.json frontend/package-lock.json Dockerfile.base.local 2>/dev/null | md5sum | cut -d' ' -f1
 }
 
 # 计算代码文件的哈希值
@@ -88,7 +88,7 @@ build_base() {
 # 构建应用镜像
 build_app() {
     echo ">>> Building app image (code only)..."
-    docker build -f Dockerfile.app -t aether-app:latest .
+    docker build -f Dockerfile.app.local -t aether-app:latest .
     save_code_hash
 }
 
@@ -162,25 +162,32 @@ git pull
 
 # 标记是否需要重启
 NEED_RESTART=false
+BASE_REBUILT=false
 
 # 检查基础镜像是否存在，或依赖是否变化
 if ! docker image inspect aether-base:latest >/dev/null 2>&1; then
     echo ">>> Base image not found, building..."
     build_base
+    BASE_REBUILT=true
     NEED_RESTART=true
 elif check_deps_changed; then
     echo ">>> Dependencies changed, rebuilding base image..."
     build_base
+    BASE_REBUILT=true
     NEED_RESTART=true
 else
     echo ">>> Dependencies unchanged."
 fi
 
-# 检查代码是否变化
+# 检查代码是否变化，或者 base 重建了（app 依赖 base）
 if ! docker image inspect aether-app:latest >/dev/null 2>&1; then
     echo ">>> App image not found, building..."
     build_app
     NEED_RESTART=true
+elif [ "$BASE_REBUILT" = true ]; then
+    echo ">>> Base image rebuilt, rebuilding app image..."
+    build_app
+    NEED_RESTART=true
 elif check_code_changed; then
     echo ">>> Code changed, rebuilding app image..."
     build_app

dev.sh

@@ -8,7 +8,8 @@ source .env
 set +a
 
 # 构建 DATABASE_URL
-export DATABASE_URL="postgresql://postgres:${DB_PASSWORD}@localhost:5432/aether"
+export DATABASE_URL="postgresql://${DB_USER:-postgres}:${DB_PASSWORD}@${DB_HOST:-localhost}:${DB_PORT:-5432}/${DB_NAME:-aether}"
+export REDIS_URL=redis://:${REDIS_PASSWORD}@${REDIS_HOST:-localhost}:${REDIS_PORT:-6379}/0
 
 # 启动 uvicorn（热重载模式）
 echo "🚀 启动本地开发服务器..."

docker-compose.build.yml

@@ -41,7 +41,7 @@ services:
   app:
     build:
       context: .
-      dockerfile: Dockerfile.app
+      dockerfile: Dockerfile.app.local
     image: aether-app:latest
     container_name: aether-app
     environment:

docker-compose.yml

@@ -12,8 +12,6 @@ services:
       TZ: Asia/Shanghai
     volumes:
       - postgres_data:/var/lib/postgresql/data
-    ports:
-      - "${DB_PORT:-5432}:5432"
     healthcheck:
       test: ["CMD-SHELL", "pg_isready -U postgres"]
       interval: 5s
@@ -27,8 +25,6 @@ services:
     command: redis-server --appendonly yes --requirepass ${REDIS_PASSWORD}
     volumes:
       - redis_data:/data
-    ports:
-      - "${REDIS_PORT:-6379}:6379"
     healthcheck:
       test: ["CMD", "redis-cli", "--raw", "incr", "ping"]
       interval: 5s

frontend/src/api/admin.ts

@@ -1,5 +1,179 @@
 import apiClient from './client'
 
+// 配置导出数据结构
+export interface ConfigExportData {
+  version: string
+  exported_at: string
+  global_models: GlobalModelExport[]
+  providers: ProviderExport[]
+}
+
+// 用户导出数据结构
+export interface UsersExportData {
+  version: string
+  exported_at: string
+  users: UserExport[]
+}
+
+export interface UserExport {
+  email: string
+  username: string
+  password_hash: string
+  role: string
+  allowed_providers?: string[] | null
+  allowed_endpoints?: string[] | null
+  allowed_models?: string[] | null
+  model_capability_settings?: any
+  quota_usd?: number | null
+  used_usd?: number
+  total_usd?: number
+  is_active: boolean
+  api_keys: UserApiKeyExport[]
+}
+
+export interface UserApiKeyExport {
+  key_hash: string
+  key_encrypted?: string | null
+  name?: string | null
+  is_standalone: boolean
+  balance_used_usd?: number
+  current_balance_usd?: number | null
+  allowed_providers?: string[] | null
+  allowed_endpoints?: string[] | null
+  allowed_api_formats?: string[] | null
+  allowed_models?: string[] | null
+  rate_limit?: number
+  concurrent_limit?: number | null
+  force_capabilities?: any
+  is_active: boolean
+  auto_delete_on_expiry?: boolean
+  total_requests?: number
+  total_cost_usd?: number
+}
+
+export interface GlobalModelExport {
+  name: string
+  display_name: string
+  default_price_per_request?: number | null
+  default_tiered_pricing: any
+  supported_capabilities?: string[] | null
+  config?: any
+  is_active: boolean
+}
+
+export interface ProviderExport {
+  name: string
+  display_name: string
+  description?: string | null
+  website?: string | null
+  billing_type?: string | null
+  monthly_quota_usd?: number | null
+  quota_reset_day?: number
+  rpm_limit?: number | null
+  provider_priority?: number
+  is_active: boolean
+  rate_limit?: number | null
+  concurrent_limit?: number | null
+  config?: any
+  endpoints: EndpointExport[]
+  models: ModelExport[]
+}
+
+export interface EndpointExport {
+  api_format: string
+  base_url: string
+  headers?: any
+  timeout?: number
+  max_retries?: number
+  max_concurrent?: number | null
+  rate_limit?: number | null
+  is_active: boolean
+  custom_path?: string | null
+  config?: any
+  keys: KeyExport[]
+}
+
+export interface KeyExport {
+  api_key: string
+  name?: string | null
+  note?: string | null
+  rate_multiplier?: number
+  internal_priority?: number
+  global_priority?: number | null
+  max_concurrent?: number | null
+  rate_limit?: number | null
+  daily_limit?: number | null
+  monthly_limit?: number | null
+  allowed_models?: string[] | null
+  capabilities?: any
+  is_active: boolean
+}
+
+export interface ModelExport {
+  global_model_name: string | null
+  provider_model_name: string
+  provider_model_mappings?: any
+  price_per_request?: number | null
+  tiered_pricing?: any
+  supports_vision?: boolean | null
+  supports_function_calling?: boolean | null
+  supports_streaming?: boolean | null
+  supports_extended_thinking?: boolean | null
+  supports_image_generation?: boolean | null
+  is_active: boolean
+  config?: any
+}
+
+// Provider 模型查询响应
+export interface ProviderModelsQueryResponse {
+  success: boolean
+  data: {
+    models: Array<{
+      id: string
+      object?: string
+      created?: number
+      owned_by?: string
+      display_name?: string
+      api_format?: string
+    }>
+    error?: string
+  }
+  provider: {
+    id: string
+    name: string
+    display_name: string
+  }
+}
+
+export interface ConfigImportRequest extends ConfigExportData {
+  merge_mode: 'skip' | 'overwrite' | 'error'
+}
+
+export interface UsersImportRequest extends UsersExportData {
+  merge_mode: 'skip' | 'overwrite' | 'error'
+}
+
+export interface UsersImportResponse {
+  message: string
+  stats: {
+    users: { created: number; updated: number; skipped: number }
+    api_keys: { created: number; skipped: number }
+    errors: string[]
+  }
+}
+
+export interface ConfigImportResponse {
+  message: string
+  stats: {
+    global_models: { created: number; updated: number; skipped: number }
+    providers: { created: number; updated: number; skipped: number }
+    endpoints: { created: number; updated: number; skipped: number }
+    keys: { created: number; updated: number; skipped: number }
+    models: { created: number; updated: number; skipped: number }
+    errors: string[]
+  }
+}
+
 // API密钥管理相关接口定义
 export interface AdminApiKey {
   id: string // UUID
@@ -173,5 +347,44 @@ export const adminApi = {
       '/api/admin/system/api-formats'
     )
     return response.data
+  },
+
+  // 导出配置
+  async exportConfig(): Promise<ConfigExportData> {
+    const response = await apiClient.get<ConfigExportData>('/api/admin/system/config/export')
+    return response.data
+  },
+
+  // 导入配置
+  async importConfig(data: ConfigImportRequest): Promise<ConfigImportResponse> {
+    const response = await apiClient.post<ConfigImportResponse>(
+      '/api/admin/system/config/import',
+      data
+    )
+    return response.data
+  },
+
+  // 导出用户数据
+  async exportUsers(): Promise<UsersExportData> {
+    const response = await apiClient.get<UsersExportData>('/api/admin/system/users/export')
+    return response.data
+  },
+
+  // 导入用户数据
+  async importUsers(data: UsersImportRequest): Promise<UsersImportResponse> {
+    const response = await apiClient.post<UsersImportResponse>(
+      '/api/admin/system/users/import',
+      data
+    )
+    return response.data
+  },
+
+  // 查询 Provider 可用模型（从上游 API 获取）
+  async queryProviderModels(providerId: string, apiKeyId?: string): Promise<ProviderModelsQueryResponse> {
+    const response = await apiClient.post<ProviderModelsQueryResponse>(
+      '/api/admin/provider-query/models',
+      { provider_id: providerId, api_key_id: apiKeyId }
+    )
+    return response.data
   }
 }

frontend/src/api/cache.ts

@@ -66,6 +66,7 @@ export interface UserAffinity {
   key_name: string | null
   key_prefix: string | null  // Provider Key 脱敏显示（前4...后4）
   rate_multiplier: number
+  global_model_id: string | null  // 原始的 global_model_id（用于删除）
   model_name: string | null  // 模型名称（如 claude-haiku-4-5-20250514）
   model_display_name: string | null  // 模型显示名称（如 Claude Haiku 4.5）
   api_format: string | null  // API 格式 (claude/openai)
@@ -119,6 +120,18 @@ export const cacheApi = {
     await api.delete(`/api/admin/monitoring/cache/users/${userIdentifier}`)
   },
 
+  /**
+   * 清除单条缓存亲和性
+   *
+   * @param affinityKey API Key ID
+   * @param endpointId Endpoint ID
+   * @param modelId GlobalModel ID
+   * @param apiFormat API 格式 (claude/openai)
+   */
+  async clearSingleAffinity(affinityKey: string, endpointId: string, modelId: string, apiFormat: string): Promise<void> {
+    await api.delete(`/api/admin/monitoring/cache/affinity/${affinityKey}/${endpointId}/${modelId}/${apiFormat}`)
+  },
+
   /**
    * 清除所有缓存
    */
@@ -271,3 +284,93 @@ export const cacheAnalysisApi = {
     return response.data
   }
 }
+
+// ==================== 模型映射缓存管理 API ====================
+
+// 映射条目
+export interface ModelMappingItem {
+  mapping_name: string
+  global_model_name: string | null
+  global_model_display_name: string | null
+  providers: string[]
+  ttl: number | null
+}
+
+// 未映射的条目（NOT_FOUND、invalid、error）
+export interface UnmappedEntry {
+  mapping_name: string
+  status: 'not_found' | 'invalid' | 'error'
+  ttl: number | null
+}
+
+// Provider 模型映射缓存（Redis 缓存）
+export interface ProviderModelMapping {
+  provider_id: string
+  provider_name: string
+  global_model_id: string
+  global_model_name: string
+  global_model_display_name: string | null
+  provider_model_name: string
+  aliases: string[] | null
+  ttl: number | null
+  hit_count: number
+}
+
+export interface ModelMappingCacheStats {
+  available: boolean
+  message?: string
+  ttl_seconds?: number
+  total_keys?: number
+  breakdown?: {
+    model_by_id: number
+    model_by_provider_global: number
+    global_model_by_id: number
+    global_model_by_name: number
+    global_model_resolve: number
+  }
+  mappings?: ModelMappingItem[]
+  provider_model_mappings?: ProviderModelMapping[] | null
+  unmapped?: UnmappedEntry[] | null
+}
+
+export interface ClearModelMappingCacheResponse {
+  status: string
+  message: string
+  deleted_count?: number
+  model_name?: string
+  deleted_keys?: string[]
+}
+
+export const modelMappingCacheApi = {
+  /**
+   * 获取模型映射缓存统计
+   */
+  async getStats(): Promise<ModelMappingCacheStats> {
+    const response = await api.get('/api/admin/monitoring/cache/model-mapping/stats')
+    return response.data.data
+  },
+
+  /**
+   * 清除所有模型映射缓存
+   */
+  async clearAll(): Promise<ClearModelMappingCacheResponse> {
+    const response = await api.delete('/api/admin/monitoring/cache/model-mapping')
+    return response.data
+  },
+
+  /**
+   * 清除指定模型名称的映射缓存
+   */
+  async clearByName(modelName: string): Promise<ClearModelMappingCacheResponse> {
+    const response = await api.delete(`/api/admin/monitoring/cache/model-mapping/${encodeURIComponent(modelName)}`)
+    return response.data
+  },
+
+  /**
+   * 清除指定 Provider 和 GlobalModel 的映射缓存
+   */
+  async clearProviderModel(providerId: string, globalModelId: string): Promise<ClearModelMappingCacheResponse> {
+    const response = await api.delete(`/api/admin/monitoring/cache/model-mapping/provider/${providerId}/${globalModelId}`)
+    return response.data
+  }
+}

frontend/src/api/endpoints/aliases.ts

@@ -1,121 +0,0 @@
-/**
- * 模型别名管理 API
- */
-
-import client from '../client'
-import type { ModelMapping, ModelMappingCreate, ModelMappingUpdate } from './types'
-
-export interface ModelAlias {
-  id: string
-  alias: string
-  global_model_id: string
-  global_model_name: string | null
-  global_model_display_name: string | null
-  provider_id: string | null
-  provider_name: string | null
-  scope: 'global' | 'provider'
-  mapping_type: 'alias' | 'mapping'
-  is_active: boolean
-  created_at: string
-  updated_at: string
-}
-
-export interface CreateModelAliasRequest {
-  alias: string
-  global_model_id: string
-  provider_id?: string | null
-  mapping_type?: 'alias' | 'mapping'
-  is_active?: boolean
-}
-
-export interface UpdateModelAliasRequest {
-  alias?: string
-  global_model_id?: string
-  provider_id?: string | null
-  mapping_type?: 'alias' | 'mapping'
-  is_active?: boolean
-}
-
-function transformMapping(mapping: ModelMapping): ModelAlias {
-  return {
-    id: mapping.id,
-    alias: mapping.source_model,
-    global_model_id: mapping.target_global_model_id,
-    global_model_name: mapping.target_global_model_name,
-    global_model_display_name: mapping.target_global_model_display_name,
-    provider_id: mapping.provider_id ?? null,
-    provider_name: mapping.provider_name ?? null,
-    scope: mapping.scope,
-    mapping_type: mapping.mapping_type || 'alias',
-    is_active: mapping.is_active,
-    created_at: mapping.created_at,
-    updated_at: mapping.updated_at
-  }
-}
-
-/**
- * 获取别名列表
- */
-export async function getAliases(params?: {
-  provider_id?: string
-  global_model_id?: string
-  is_active?: boolean
-  skip?: number
-  limit?: number
-}): Promise<ModelAlias[]> {
-  const response = await client.get('/api/admin/models/mappings', {
-    params: {
-      provider_id: params?.provider_id,
-      target_global_model_id: params?.global_model_id,
-      is_active: params?.is_active,
-      skip: params?.skip,
-      limit: params?.limit
-    }
-  })
-  return (response.data as ModelMapping[]).map(transformMapping)
-}
-
-/**
- * 获取单个别名
- */
-export async function getAlias(id: string): Promise<ModelAlias> {
-  const response = await client.get(`/api/admin/models/mappings/${id}`)
-  return transformMapping(response.data)
-}
-
-/**
- * 创建别名
- */
-export async function createAlias(data: CreateModelAliasRequest): Promise<ModelAlias> {
-  const payload: ModelMappingCreate = {
-    source_model: data.alias,
-    target_global_model_id: data.global_model_id,
-    provider_id: data.provider_id ?? null,
-    mapping_type: data.mapping_type ?? 'alias',
-    is_active: data.is_active ?? true
-  }
-  const response = await client.post('/api/admin/models/mappings', payload)
-  return transformMapping(response.data)
-}
-
-/**
- * 更新别名
- */
-export async function updateAlias(id: string, data: UpdateModelAliasRequest): Promise<ModelAlias> {
-  const payload: ModelMappingUpdate = {
-    source_model: data.alias,
-    target_global_model_id: data.global_model_id,
-    provider_id: data.provider_id ?? null,
-    mapping_type: data.mapping_type,
-    is_active: data.is_active
-  }
-  const response = await client.patch(`/api/admin/models/mappings/${id}`, payload)
-  return transformMapping(response.data)
-}
-
-/**
- * 删除别名
- */
-export async function deleteAlias(id: string): Promise<void> {
-  await client.delete(`/api/admin/models/mappings/${id}`)
-}

frontend/src/api/endpoints/endpoints.ts

@@ -1,5 +1,5 @@
 import client from '../client'
-import type { ProviderEndpoint } from './types'
+import type { ProviderEndpoint, ProxyConfig } from './types'
 
 /**
  * 获取指定 Provider 的所有 Endpoints
@@ -38,6 +38,7 @@ export async function createEndpoint(
     rate_limit?: number
     is_active?: boolean
     config?: Record<string, any>
+    proxy?: ProxyConfig | null
   }
 ): Promise<ProviderEndpoint> {
   const response = await client.post(`/api/admin/endpoints/providers/${providerId}/endpoints`, data)
@@ -63,6 +64,7 @@ export async function updateEndpoint(
     rate_limit: number
     is_active: boolean
     config: Record<string, any>
+    proxy: ProxyConfig | null
   }>
 ): Promise<ProviderEndpoint> {
   const response = await client.put(`/api/admin/endpoints/${endpointId}`, data)

frontend/src/api/endpoints/index.ts

@@ -4,6 +4,5 @@ export * from './endpoints'
 export * from './keys'
 export * from './health'
 export * from './models'
-export * from './aliases'
 export * from './adaptive'
 export * from './global-models'

frontend/src/api/endpoints/models.ts

@@ -5,9 +5,8 @@ import type {
   ModelUpdate,
   ModelCatalogResponse,
   ProviderAvailableSourceModelsResponse,
-  UpdateModelMappingRequest,
-  UpdateModelMappingResponse,
-  DeleteModelMappingResponse
+  UpstreamModel,
+  ImportFromUpstreamResponse,
 } from './types'
 
 /**
@@ -99,27 +98,6 @@ export async function getProviderAvailableSourceModels(
   return response.data
 }
 
-/**
- * 更新目录中的模型映射
- */
-export async function updateCatalogMapping(
-  mappingId: string,
-  data: UpdateModelMappingRequest
-): Promise<UpdateModelMappingResponse> {
-  const response = await client.put(`/api/admin/models/catalog/mappings/${mappingId}`, data)
-  return response.data
-}
-
-/**
- * 删除目录中的模型映射
- */
-export async function deleteCatalogMapping(
-  mappingId: string
-): Promise<DeleteModelMappingResponse> {
-  const response = await client.delete(`/api/admin/models/catalog/mappings/${mappingId}`)
-  return response.data
-}
-
 /**
  * 批量为 Provider 关联 GlobalModels
  */
@@ -143,3 +121,40 @@ export async function batchAssignModelsToProvider(
   )
   return response.data
 }
+
+/**
+ * 查询提供商的上游模型列表
+ */
+export async function queryProviderUpstreamModels(
+  providerId: string
+): Promise<{
+  success: boolean
+  data: {
+    models: UpstreamModel[]
+    error: string | null
+  }
+  provider: {
+    id: string
+    name: string
+    display_name: string
+  }
+}> {
+  const response = await client.post('/api/admin/provider-query/models', {
+    provider_id: providerId,
+  })
+  return response.data
+}
+
+/**
+ * 从上游提供商导入模型
+ */
+export async function importModelsFromUpstream(
+  providerId: string,
+  modelIds: string[]
+): Promise<ImportFromUpstreamResponse> {
+  const response = await client.post(
+    `/api/admin/providers/${providerId}/import-from-upstream`,
+    { model_ids: modelIds }
+  )
+  return response.data
+}

frontend/src/api/endpoints/types.ts

@@ -1,3 +1,35 @@
+// API 格式常量
+export const API_FORMATS = {
+  CLAUDE: 'CLAUDE',
+  CLAUDE_CLI: 'CLAUDE_CLI',
+  OPENAI: 'OPENAI',
+  OPENAI_CLI: 'OPENAI_CLI',
+  GEMINI: 'GEMINI',
+  GEMINI_CLI: 'GEMINI_CLI',
+} as const
+
+export type APIFormat = typeof API_FORMATS[keyof typeof API_FORMATS]
+
+// API 格式显示名称映射（按品牌分组：API 在前，CLI 在后）
+export const API_FORMAT_LABELS: Record<string, string> = {
+  [API_FORMATS.CLAUDE]: 'Claude',
+  [API_FORMATS.CLAUDE_CLI]: 'Claude CLI',
+  [API_FORMATS.OPENAI]: 'OpenAI',
+  [API_FORMATS.OPENAI_CLI]: 'OpenAI CLI',
+  [API_FORMATS.GEMINI]: 'Gemini',
+  [API_FORMATS.GEMINI_CLI]: 'Gemini CLI',
+}
+
+/**
+ * 代理配置类型
+ */
+export interface ProxyConfig {
+  url: string
+  username?: string
+  password?: string
+  enabled?: boolean  // 是否启用代理（false 时保留配置但不使用）
+}
+
 export interface ProviderEndpoint {
   id: string
   provider_id: string
@@ -19,6 +51,7 @@ export interface ProviderEndpoint {
   last_failure_at?: string
   is_active: boolean
   config?: Record<string, any>
+  proxy?: ProxyConfig | null
   total_keys: number
   active_keys: number
   created_at: string
@@ -211,11 +244,21 @@ export interface ConcurrencyStatus {
   key_max_concurrent?: number
 }
 
+export interface ProviderModelMapping {
+  name: string
+  priority: number  // 优先级（数字越小优先级越高）
+  api_formats?: string[]  // 作用域（适用的 API 格式），为空表示对所有格式生效
+}
+
+// 保留别名以保持向后兼容
+export type ProviderModelAlias = ProviderModelMapping
+
 export interface Model {
   id: string
   provider_id: string
   global_model_id?: string  // 关联的 GlobalModel ID
-  provider_model_name: string  // Provider 侧的模型名称（原 name）
+  provider_model_name: string  // Provider 侧的主模型名称
+  provider_model_mappings?: ProviderModelMapping[] | null  // 模型名称映射列表（带优先级）
   // 原始配置值（可能为空，为空时使用 GlobalModel 默认值）
   price_per_request?: number | null  // 按次计费价格
   tiered_pricing?: TieredPricingConfig | null  // 阶梯计费配置
@@ -244,7 +287,8 @@ export interface Model {
 }
 
 export interface ModelCreate {
-  provider_model_name: string  // Provider 侧的模型名称（原 name）
+  provider_model_name: string  // Provider 侧的主模型名称
+  provider_model_mappings?: ProviderModelMapping[]  // 模型名称映射列表（带优先级）
   global_model_id: string  // 关联的 GlobalModel ID（必填）
   // 计费配置（可选，为空时使用 GlobalModel 默认值）
   price_per_request?: number  // 按次计费价格
@@ -261,6 +305,7 @@ export interface ModelCreate {
 
 export interface ModelUpdate {
   provider_model_name?: string
+  provider_model_mappings?: ProviderModelMapping[] | null  // 模型名称映射列表（带优先级）
   global_model_id?: string
   price_per_request?: number | null  // 按次计费价格（null 表示清空/使用默认值）
   tiered_pricing?: TieredPricingConfig | null  // 阶梯计费配置
@@ -273,21 +318,6 @@ export interface ModelUpdate {
   is_available?: boolean
 }
 
-export interface ModelMapping {
-  id: string
-  source_model: string  // 别名/源模型名
-  target_global_model_id: string  // 目标 GlobalModel ID
-  target_global_model_name: string | null
-  target_global_model_display_name: string | null
-  provider_id: string | null
-  provider_name: string | null
-  scope: 'global' | 'provider'
-  mapping_type: 'alias' | 'mapping'
-  is_active: boolean
-  created_at: string
-  updated_at: string
-}
-
 export interface ModelCapabilities {
   supports_vision: boolean
   supports_function_calling: boolean
@@ -335,7 +365,6 @@ export interface ModelCatalogItem {
   global_model_name: string  // GlobalModel.name（原 source_model）
   display_name: string  // GlobalModel.display_name
   description?: string | null  // GlobalModel.description
-  aliases: string[]  // 所有指向该 GlobalModel 的别名列表
   providers: ModelCatalogProviderDetail[]  // 支持该模型的 Provider 列表
   price_range: ModelPriceRange  // 价格区间
   total_providers: number
@@ -351,8 +380,6 @@ export interface ProviderAvailableSourceModel {
   global_model_name: string  // GlobalModel.name（原 source_model）
   display_name: string  // GlobalModel.display_name
   provider_model_name: string  // Model.provider_model_name（Provider 侧的模型名）
-  has_alias: boolean  // 是否有别名指向该 GlobalModel
-  aliases: string[]  // 别名列表
   model_id?: string | null  // Model.id
   price: ProviderModelPriceInfo
   capabilities: ModelCapabilities
@@ -371,65 +398,6 @@ export interface BatchAssignProviderConfig {
   model_id?: string
 }
 
-export interface BatchAssignModelMappingRequest {
-  global_model_id: string  // 要分配的 GlobalModel ID（原 source_model）
-  providers: BatchAssignProviderConfig[]
-}
-
-export interface BatchAssignProviderResult {
-  provider_id: string
-  mapping_id?: string | null
-  created_model: boolean
-  model_id?: string | null
-  updated: boolean
-}
-
-export interface BatchAssignError {
-  provider_id: string
-  error: string
-}
-
-export interface BatchAssignModelMappingResponse {
-  success: boolean
-  created_mappings: BatchAssignProviderResult[]
-  errors: BatchAssignError[]
-}
-
-export interface ModelMappingCreate {
-  source_model: string  // 源模型名或别名
-  target_global_model_id: string  // 目标 GlobalModel ID
-  provider_id?: string | null
-  mapping_type?: 'alias' | 'mapping'
-  is_active?: boolean
-}
-
-export interface ModelMappingUpdate {
-  source_model?: string  // 源模型名或别名
-  target_global_model_id?: string  // 目标 GlobalModel ID
-  provider_id?: string | null
-  mapping_type?: 'alias' | 'mapping'
-  is_active?: boolean
-}
-
-export interface UpdateModelMappingRequest {
-  source_model?: string
-  target_global_model_id?: string
-  provider_id?: string | null
-  mapping_type?: 'alias' | 'mapping'
-  is_active?: boolean
-}
-
-export interface UpdateModelMappingResponse {
-  success: boolean
-  mapping_id: string
-  message?: string
-}
-
-export interface DeleteModelMappingResponse {
-  success: boolean
-  message?: string
-}
-
 export interface AdaptiveStatsResponse {
   adaptive_mode: boolean
   current_limit: number | null
@@ -476,67 +444,45 @@ export interface TieredPricingConfig {
 export interface GlobalModelCreate {
   name: string
   display_name: string
-  description?: string
-  official_url?: string
-  icon_url?: string
   // 按次计费配置（可选，与阶梯计费叠加）
   default_price_per_request?: number
   // 阶梯计费配置（必填，固定价格用单阶梯表示）
   default_tiered_pricing: TieredPricingConfig
-  // 默认能力配置
-  default_supports_vision?: boolean
-  default_supports_function_calling?: boolean
-  default_supports_streaming?: boolean
-  default_supports_extended_thinking?: boolean
-  default_supports_image_generation?: boolean
   // Key 能力配置 - 模型支持的能力列表
   supported_capabilities?: string[]
+  // 模型配置（JSON格式）- 包含能力、规格、元信息等
+  config?: Record<string, any>
   is_active?: boolean
 }
 
 export interface GlobalModelUpdate {
   display_name?: string
-  description?: string
-  official_url?: string
-  icon_url?: string
   is_active?: boolean
   // 按次计费配置
   default_price_per_request?: number | null  // null 表示清空
   // 阶梯计费配置
   default_tiered_pricing?: TieredPricingConfig
-  // 默认能力配置
-  default_supports_vision?: boolean
-  default_supports_function_calling?: boolean
-  default_supports_streaming?: boolean
-  default_supports_extended_thinking?: boolean
-  default_supports_image_generation?: boolean
   // Key 能力配置 - 模型支持的能力列表
   supported_capabilities?: string[] | null
+  // 模型配置（JSON格式）- 包含能力、规格、元信息等
+  config?: Record<string, any> | null
 }
 
 export interface GlobalModelResponse {
   id: string
   name: string
   display_name: string
-  description?: string
-  official_url?: string
-  icon_url?: string
   is_active: boolean
   // 按次计费配置
   default_price_per_request?: number
   // 阶梯计费配置（必填）
   default_tiered_pricing: TieredPricingConfig
-  // 默认能力配置
-  default_supports_vision?: boolean
-  default_supports_function_calling?: boolean
-  default_supports_streaming?: boolean
-  default_supports_extended_thinking?: boolean
-  default_supports_image_generation?: boolean
   // Key 能力配置 - 模型支持的能力列表
   supported_capabilities?: string[] | null
+  // 模型配置（JSON格式）
+  config?: Record<string, any> | null
   // 统计数据
   provider_count?: number
-  alias_count?: number
   usage_count?: number
   created_at: string
   updated_at?: string
@@ -552,3 +498,42 @@ export interface GlobalModelListResponse {
   models: GlobalModelResponse[]
   total: number
 }
+
+// ==================== 上游模型导入相关 ====================
+
+/**
+ * 上游模型（从提供商 API 获取的原始模型）
+ */
+export interface UpstreamModel {
+  id: string
+  owned_by?: string
+  display_name?: string
+  api_format?: string
+}
+
+/**
+ * 导入成功的模型信息
+ */
+export interface ImportFromUpstreamSuccessItem {
+  model_id: string
+  global_model_id: string
+  global_model_name: string
+  provider_model_id: string
+  created_global_model: boolean
+}
+
+/**
+ * 导入失败的模型信息
+ */
+export interface ImportFromUpstreamErrorItem {
+  model_id: string
+  error: string
+}
+
+/**
+ * 从上游提供商导入模型响应
+ */
+export interface ImportFromUpstreamResponse {
+  success: ImportFromUpstreamSuccessItem[]
+  errors: ImportFromUpstreamErrorItem[]
+}

frontend/src/api/models-dev.ts

@@ -0,0 +1,288 @@
+/**
+ * Models.dev API 服务
+ * 通过后端代理获取 models.dev 数据（解决跨域问题）
+ */
+
+import api from './client'
+
+// 缓存配置
+const CACHE_KEY = 'models_dev_cache'
+const CACHE_DURATION = 15 * 60 * 1000 // 15 分钟
+
+// Models.dev API 数据结构
+export interface ModelsDevCost {
+  input?: number
+  output?: number
+  reasoning?: number
+  cache_read?: number
+}
+
+export interface ModelsDevLimit {
+  context?: number
+  output?: number
+}
+
+export interface ModelsDevModel {
+  id: string
+  name: string
+  family?: string
+  reasoning?: boolean
+  tool_call?: boolean
+  structured_output?: boolean
+  temperature?: boolean
+  attachment?: boolean
+  knowledge?: string
+  release_date?: string
+  last_updated?: string
+  input?: string[] // 输入模态: text, image, audio, video, pdf
+  output?: string[] // 输出模态: text, image, audio
+  open_weights?: boolean
+  cost?: ModelsDevCost
+  limit?: ModelsDevLimit
+  deprecated?: boolean
+}
+
+export interface ModelsDevProvider {
+  id: string
+  env?: string[]
+  npm?: string
+  api?: string
+  name: string
+  doc?: string
+  models: Record<string, ModelsDevModel>
+  official?: boolean // 是否为官方提供商
+}
+
+export type ModelsDevData = Record<string, ModelsDevProvider>
+
+// 扁平化的模型列表项（用于搜索和选择）
+export interface ModelsDevModelItem {
+  providerId: string
+  providerName: string
+  modelId: string
+  modelName: string
+  family?: string
+  inputPrice?: number
+  outputPrice?: number
+  contextLimit?: number
+  outputLimit?: number
+  supportsVision?: boolean
+  supportsToolCall?: boolean
+  supportsReasoning?: boolean
+  supportsStructuredOutput?: boolean
+  supportsTemperature?: boolean
+  supportsAttachment?: boolean
+  openWeights?: boolean
+  deprecated?: boolean
+  official?: boolean // 是否来自官方提供商
+  // 用于 display_metadata 的额外字段
+  knowledgeCutoff?: string
+  releaseDate?: string
+  inputModalities?: string[]
+  outputModalities?: string[]
+}
+
+interface CacheData {
+  timestamp: number
+  data: ModelsDevData
+}
+
+// 内存缓存
+let memoryCache: CacheData | null = null
+
+function hasOfficialFlag(data: ModelsDevData): boolean {
+  return Object.values(data).some(provider => typeof provider?.official === 'boolean')
+}
+
+/**
+ * 获取 models.dev 数据（带缓存）
+ */
+export async function getModelsDevData(): Promise<ModelsDevData> {
+  // 1. 检查内存缓存
+  if (memoryCache && Date.now() - memoryCache.timestamp < CACHE_DURATION) {
+    // 兼容旧缓存：没有 official 字段时丢弃，强制刷新一次
+    if (hasOfficialFlag(memoryCache.data)) {
+      return memoryCache.data
+    }
+    memoryCache = null
+  }
+
+  // 2. 检查 localStorage 缓存
+  try {
+    const cached = localStorage.getItem(CACHE_KEY)
+    if (cached) {
+      const cacheData: CacheData = JSON.parse(cached)
+      if (Date.now() - cacheData.timestamp < CACHE_DURATION) {
+        // 兼容旧缓存：没有 official 字段时丢弃，强制刷新一次
+        if (hasOfficialFlag(cacheData.data)) {
+          memoryCache = cacheData
+          return cacheData.data
+        }
+        localStorage.removeItem(CACHE_KEY)
+      }
+    }
+  } catch {
+    // 缓存解析失败，忽略
+  }
+
+  // 3. 从后端代理获取新数据
+  const response = await api.get<ModelsDevData>('/api/admin/models/external')
+  const data = response.data
+
+  // 4. 更新缓存
+  const cacheData: CacheData = {
+    timestamp: Date.now(),
+    data,
+  }
+  memoryCache = cacheData
+  try {
+    localStorage.setItem(CACHE_KEY, JSON.stringify(cacheData))
+  } catch {
+    // localStorage 写入失败，忽略
+  }
+
+  return data
+}
+
+// 模型列表缓存（避免重复转换）
+let modelsListCache: ModelsDevModelItem[] | null = null
+let modelsListCacheTimestamp: number | null = null
+
+/**
+ * 获取扁平化的模型列表
+ * 数据只加载一次，通过参数过滤官方/全部
+ */
+export async function getModelsDevList(officialOnly: boolean = true): Promise<ModelsDevModelItem[]> {
+  const data = await getModelsDevData()
+  const currentTimestamp = memoryCache?.timestamp ?? 0
+
+  // 如果缓存为空或数据已刷新，构建一次
+  if (!modelsListCache || modelsListCacheTimestamp !== currentTimestamp) {
+    const items: ModelsDevModelItem[] = []
+
+    for (const [providerId, provider] of Object.entries(data)) {
+      if (!provider.models) continue
+
+      for (const [modelId, model] of Object.entries(provider.models)) {
+        items.push({
+          providerId,
+          providerName: provider.name,
+          modelId,
+          modelName: model.name || modelId,
+          family: model.family,
+          inputPrice: model.cost?.input,
+          outputPrice: model.cost?.output,
+          contextLimit: model.limit?.context,
+          outputLimit: model.limit?.output,
+          supportsVision: model.input?.includes('image'),
+          supportsToolCall: model.tool_call,
+          supportsReasoning: model.reasoning,
+          supportsStructuredOutput: model.structured_output,
+          supportsTemperature: model.temperature,
+          supportsAttachment: model.attachment,
+          openWeights: model.open_weights,
+          deprecated: model.deprecated,
+          official: provider.official,
+          // display_metadata 相关字段
+          knowledgeCutoff: model.knowledge,
+          releaseDate: model.release_date,
+          inputModalities: model.input,
+          outputModalities: model.output,
+        })
+      }
+    }
+
+    // 按 provider 名称和模型名称排序
+    items.sort((a, b) => {
+      const providerCompare = a.providerName.localeCompare(b.providerName)
+      if (providerCompare !== 0) return providerCompare
+      return a.modelName.localeCompare(b.modelName)
+    })
+
+    modelsListCache = items
+    modelsListCacheTimestamp = currentTimestamp
+  }
+
+  // 根据参数过滤
+  if (officialOnly) {
+    return modelsListCache.filter(m => m.official)
+  }
+  return modelsListCache
+}
+
+/**
+ * 搜索模型
+ * 搜索时包含所有提供商（包括第三方）
+ */
+export async function searchModelsDevModels(
+  query: string,
+  options?: {
+    limit?: number
+    excludeDeprecated?: boolean
+  }
+): Promise<ModelsDevModelItem[]> {
+  // 搜索时包含全部提供商
+  const allModels = await getModelsDevList(false)
+  const { limit = 50, excludeDeprecated = true } = options || {}
+
+  const queryLower = query.toLowerCase()
+
+  const filtered = allModels.filter((model) => {
+    if (excludeDeprecated && model.deprecated) return false
+
+    // 搜索模型 ID、名称、provider 名称、family
+    return (
+      model.modelId.toLowerCase().includes(queryLower) ||
+      model.modelName.toLowerCase().includes(queryLower) ||
+      model.providerName.toLowerCase().includes(queryLower) ||
+      model.family?.toLowerCase().includes(queryLower)
+    )
+  })
+
+  // 排序：精确匹配优先
+  filtered.sort((a, b) => {
+    const aExact =
+      a.modelId.toLowerCase() === queryLower ||
+      a.modelName.toLowerCase() === queryLower
+    const bExact =
+      b.modelId.toLowerCase() === queryLower ||
+      b.modelName.toLowerCase() === queryLower
+    if (aExact && !bExact) return -1
+    if (!aExact && bExact) return 1
+    return 0
+  })
+
+  return filtered.slice(0, limit)
+}
+
+/**
+ * 获取特定模型详情
+ */
+export async function getModelsDevModel(
+  providerId: string,
+  modelId: string
+): Promise<ModelsDevModel | null> {
+  const data = await getModelsDevData()
+  return data[providerId]?.models?.[modelId] || null
+}
+
+/**
+ * 获取 provider logo URL
+ */
+export function getProviderLogoUrl(providerId: string): string {
+  return `https://models.dev/logos/${providerId}.svg`
+}
+
+/**
+ * 清除缓存
+ */
+export function clearModelsDevCache(): void {
+  memoryCache = null
+  modelsListCache = null
+  modelsListCacheTimestamp = null
+  try {
+    localStorage.removeItem(CACHE_KEY)
+  } catch {
+    // 忽略错误
+  }
+}

frontend/src/api/public-models.ts

@@ -9,20 +9,14 @@ export interface PublicGlobalModel {
   id: string
   name: string
   display_name: string | null
-  description: string | null
-  icon_url: string | null
   is_active: boolean
   // 阶梯计费配置
   default_tiered_pricing: TieredPricingConfig
   default_price_per_request: number | null  // 按次计费价格
-  // 能力
-  default_supports_vision: boolean
-  default_supports_function_calling: boolean
-  default_supports_streaming: boolean
-  default_supports_extended_thinking: boolean
-  default_supports_image_generation: boolean
   // Key 能力支持
   supported_capabilities: string[] | null
+  // 模型配置（JSON）
+  config: Record<string, any> | null
 }
 
 export interface PublicGlobalModelListResponse {

frontend/src/components/charts/ScatterChart.vue

@@ -299,7 +299,7 @@ function formatDuration(ms: number): string {
   const hours = Math.floor(ms / (1000 * 60 * 60))
   const minutes = Math.floor((ms % (1000 * 60 * 60)) / (1000 * 60))
   if (hours > 0) {
-    return `${hours}h${minutes > 0 ? minutes + 'm' : ''}`
+    return `${hours}h${minutes > 0 ? `${minutes}m` : ''}`
   }
   return `${minutes}m`
 }

frontend/src/components/ui/button.vue

@@ -34,11 +34,10 @@ const buttonClass = computed(() => {
     'inline-flex items-center justify-center rounded-xl text-sm font-semibold transition-all duration-200 ring-offset-background focus-visible:outline-none focus-visible:ring-2 focus-visible:ring-ring focus-visible:ring-offset-2 disabled:pointer-events-none disabled:opacity-50 active:scale-[0.98]'
 
   const variantClasses = {
-    default:
-      'bg-primary text-white shadow-[0_20px_35px_rgba(204,120,92,0.35)] hover:bg-primary/90 hover:shadow-[0_25px_45px_rgba(204,120,92,0.45)]',
-    destructive: 'bg-destructive text-destructive-foreground hover:bg-destructive/85 shadow-sm',
+    default: 'bg-primary text-white hover:bg-primary/90',
+    destructive: 'bg-destructive text-destructive-foreground hover:bg-destructive/85',
     outline:
-      'border border-border/60 bg-card/60 text-foreground hover:border-primary/60 hover:text-primary hover:bg-primary/10 shadow-sm backdrop-blur transition-all',
+      'border border-border/60 bg-card/60 text-foreground hover:border-primary/60 hover:text-primary hover:bg-primary/10 backdrop-blur transition-all',
     secondary:
       'bg-secondary text-secondary-foreground shadow-inner hover:bg-secondary/80',
     ghost: 'hover:bg-accent hover:text-accent-foreground',

frontend/src/components/ui/dialog/Dialog.vue

@@ -2,7 +2,7 @@
   <Teleport to="body">
     <div
       v-if="isOpen"
-      class="fixed inset-0 overflow-y-auto"
+      class="fixed inset-0 overflow-y-auto pointer-events-none"
       :style="{ zIndex: containerZIndex }"
     >
       <!-- 背景遮罩 -->
@@ -16,13 +16,13 @@
       >
         <div
           v-if="isOpen"
-          class="fixed inset-0 bg-black/40 backdrop-blur-sm transition-opacity"
+          class="fixed inset-0 bg-black/40 backdrop-blur-sm transition-opacity pointer-events-auto"
           :style="{ zIndex: backdropZIndex }"
           @click="handleClose"
         />
       </Transition>
 
-      <div class="relative flex min-h-full items-end justify-center p-4 text-center sm:items-center sm:p-0">
+      <div class="relative flex min-h-full items-end justify-center p-4 text-center sm:items-center sm:p-0 pointer-events-none">
         <!-- 对话框内容 -->
         <Transition
           enter-active-class="duration-300 ease-out"
@@ -34,7 +34,7 @@
         >
           <div
             v-if="isOpen"
-            class="relative transform rounded-lg bg-background text-left shadow-2xl transition-all sm:my-8 sm:w-full border border-border"
+            class="relative transform rounded-lg bg-background text-left shadow-2xl transition-all sm:my-8 sm:w-full border border-border pointer-events-auto"
             :style="{ zIndex: contentZIndex }"
             :class="maxWidthClass"
             @click.stop
@@ -92,6 +92,7 @@
 
 <script setup lang="ts">
 import { computed, useSlots, type Component } from 'vue'
+import { useEscapeKey } from '@/composables/useEscapeKey'
 
 // Props 定义
 const props = defineProps<{
@@ -157,4 +158,14 @@ const maxWidthClass = computed(() => {
 const containerZIndex = computed(() => props.zIndex || 60)
 const backdropZIndex = computed(() => props.zIndex || 60)
 const contentZIndex = computed(() => (props.zIndex || 60) + 10)
+
+// 添加 ESC 键监听
+useEscapeKey(() => {
+  if (isOpen.value) {
+    handleClose()
+  }
+}, {
+  disableOnInput: true,
+  once: false
+})
 </script>

frontend/src/components/ui/select-content.vue

@@ -45,7 +45,7 @@ const props = withDefaults(defineProps<Props>(), {
 
 const contentClass = computed(() =>
   cn(
-    'z-[100] max-h-96 min-w-[8rem] overflow-hidden rounded-2xl border border-border bg-card text-foreground shadow-2xl backdrop-blur-xl pointer-events-auto',
+    'z-[200] max-h-96 min-w-[8rem] overflow-hidden rounded-2xl border border-border bg-card text-foreground shadow-2xl backdrop-blur-xl pointer-events-auto',
     'data-[state=open]:animate-in data-[state=closed]:animate-out data-[state=closed]:fade-out-0 data-[state=open]:fade-in-0 data-[state=closed]:zoom-out-95 data-[state=open]:zoom-in-95',
     'data-[side=bottom]:slide-in-from-top-2 data-[side=left]:slide-in-from-right-2 data-[side=right]:slide-in-from-left-2 data-[side=top]:slide-in-from-bottom-2',
     props.class

frontend/src/composables/useEscapeKey.ts

@@ -0,0 +1,80 @@
+import { onMounted, onUnmounted, ref } from 'vue'
+
+/**
+ * ESC 键监听 Composable（简化版本，直接使用独立监听器）
+ * 用于按 ESC 键关闭弹窗或其他可关闭的组件
+ *
+ * @param callback - 按 ESC 键时执行的回调函数
+ * @param options - 配置选项
+ */
+export function useEscapeKey(
+  callback: () => void,
+  options: {
+    /** 是否在输入框获得焦点时禁用 ESC 键，默认 true */
+    disableOnInput?: boolean
+    /** 是否只监听一次，默认 false */
+    once?: boolean
+  } = {}
+) {
+  const { disableOnInput = true, once = false } = options
+  const isActive = ref(true)
+
+  function handleKeyDown(event: KeyboardEvent) {
+    // 只处理 ESC 键
+    if (event.key !== 'Escape') return
+
+    // 检查组件是否还活跃
+    if (!isActive.value) return
+
+    // 如果配置了在输入框获得焦点时禁用，则检查当前焦点元素
+    if (disableOnInput) {
+      const activeElement = document.activeElement
+      const isInputElement = activeElement && (
+        activeElement.tagName === 'INPUT' ||
+        activeElement.tagName === 'TEXTAREA' ||
+        activeElement.tagName === 'SELECT' ||
+        activeElement.contentEditable === 'true' ||
+        activeElement.getAttribute('role') === 'textbox' ||
+        activeElement.getAttribute('role') === 'combobox'
+      )
+
+      // 如果焦点在输入框中，不处理 ESC 键
+      if (isInputElement) return
+    }
+
+    // 执行回调
+    callback()
+
+    // 移除当前元素的焦点，避免残留样式
+    if (document.activeElement instanceof HTMLElement) {
+      document.activeElement.blur()
+    }
+
+    // 如果只监听一次，则移除监听器
+    if (once) {
+      removeEventListener()
+    }
+  }
+
+  function addEventListener() {
+    document.addEventListener('keydown', handleKeyDown)
+  }
+
+  function removeEventListener() {
+    document.removeEventListener('keydown', handleKeyDown)
+  }
+
+  onMounted(() => {
+    addEventListener()
+  })
+
+  onUnmounted(() => {
+    isActive.value = false
+    removeEventListener()
+  })
+
+  return {
+    addEventListener,
+    removeEventListener
+  }
+}

frontend/src/features/api-keys/components/StandaloneKeyFormDialog.vue

@@ -132,7 +132,7 @@
               type="number"
               min="1"
               max="10000"
-              placeholder="100"
+              placeholder="留空不限制"
               class="h-10"
               @update:model-value="(v) => form.rate_limit = parseNumberInput(v, { min: 1, max: 10000 })"
             />
@@ -376,7 +376,7 @@ const form = ref<StandaloneKeyFormData>({
   initial_balance_usd: 10,
   expire_days: undefined,
   never_expire: true,
-  rate_limit: 100,
+  rate_limit: undefined,
   auto_delete_on_expiry: false,
   allowed_providers: [],
   allowed_api_formats: [],
@@ -389,7 +389,7 @@ function resetForm() {
     initial_balance_usd: 10,
     expire_days: undefined,
     never_expire: true,
-    rate_limit: 100,
+    rate_limit: undefined,
     auto_delete_on_expiry: false,
     allowed_providers: [],
     allowed_api_formats: [],
@@ -408,7 +408,7 @@ function loadKeyData() {
     initial_balance_usd: props.apiKey.initial_balance_usd,
     expire_days: props.apiKey.expire_days,
     never_expire: props.apiKey.never_expire,
-    rate_limit: props.apiKey.rate_limit || 100,
+    rate_limit: props.apiKey.rate_limit,
     auto_delete_on_expiry: props.apiKey.auto_delete_on_expiry,
     allowed_providers: props.apiKey.allowed_providers || [],
     allowed_api_formats: props.apiKey.allowed_api_formats || [],

frontend/src/features/models/components/AliasDialog.vue

@@ -1,384 +0,0 @@
-<template>
-  <Dialog
-    :model-value="open"
-    :title="dialogTitle"
-    :description="dialogDescription"
-    :icon="dialogIcon"
-    size="md"
-    @update:model-value="handleDialogUpdate"
-  >
-    <form
-      class="space-y-4"
-      @submit.prevent="handleSubmit"
-    >
-      <!-- 模式选择（仅创建时显示） -->
-      <div
-        v-if="!isEditMode"
-        class="space-y-2"
-      >
-        <Label>创建类型 *</Label>
-        <div class="grid grid-cols-2 gap-3">
-          <button
-            type="button"
-            class="p-3 rounded-lg border-2 text-left transition-all"
-            :class="[
-              form.mapping_type === 'alias'
-                ? 'border-primary bg-primary/5'
-                : 'border-border hover:border-primary/50'
-            ]"
-            @click="form.mapping_type = 'alias'"
-          >
-            <div class="font-medium text-sm">
-              别名
-            </div>
-            <div class="text-xs text-muted-foreground mt-1">
-              名称简写，按目标模型计费
-            </div>
-          </button>
-          <button
-            type="button"
-            class="p-3 rounded-lg border-2 text-left transition-all"
-            :class="[
-              form.mapping_type === 'mapping'
-                ? 'border-primary bg-primary/5'
-                : 'border-border hover:border-primary/50'
-            ]"
-            @click="form.mapping_type = 'mapping'"
-          >
-            <div class="font-medium text-sm">
-              映射
-            </div>
-            <div class="text-xs text-muted-foreground mt-1">
-              模型降级，按源模型计费
-            </div>
-          </button>
-        </div>
-      </div>
-
-      <!-- 模式说明 -->
-      <div class="rounded-lg border border-border bg-muted/50 p-3 text-sm">
-        <p class="text-foreground font-medium mb-1">
-          {{ form.mapping_type === 'alias' ? '别名模式' : '映射模式' }}
-        </p>
-        <p class="text-muted-foreground text-xs">
-          {{ form.mapping_type === 'alias'
-            ? '用户请求此别名时，会路由到目标模型，并按目标模型价格计费。'
-            : '将源模型的请求转发到目标模型处理，按源模型价格计费。' }}
-        </p>
-      </div>
-
-      <!-- Provider 选择/作用范围 -->
-      <div
-        v-if="showProviderSelect"
-        class="space-y-2"
-      >
-        <Label>作用范围</Label>
-        <!-- 固定 Provider 时显示只读 -->
-        <div
-          v-if="fixedProvider"
-          class="px-3 py-2 border rounded-md bg-muted/50 text-sm"
-        >
-          仅 {{ fixedProvider.display_name || fixedProvider.name }}
-        </div>
-        <!-- 否则显示可选择的下拉 -->
-        <Select
-          v-else
-          v-model:open="providerSelectOpen"
-          :model-value="form.provider_id || 'global'"
-          @update:model-value="handleProviderChange"
-        >
-          <SelectTrigger class="w-full">
-            <SelectValue placeholder="选择作用范围" />
-          </SelectTrigger>
-          <SelectContent>
-            <SelectItem value="global">
-              全局（所有 Provider）
-            </SelectItem>
-            <SelectItem
-              v-for="p in providers"
-              :key="p.id"
-              :value="p.id"
-            >
-              仅 {{ p.display_name || p.name }}
-            </SelectItem>
-          </SelectContent>
-        </Select>
-      </div>
-
-      <!-- 别名模式：别名名称 -->
-      <div
-        v-if="form.mapping_type === 'alias'"
-        class="space-y-2"
-      >
-        <Label for="alias-name">别名名称 *</Label>
-        <Input
-          id="alias-name"
-          v-model="form.alias"
-          placeholder="如：sonnet, opus"
-          :disabled="isEditMode"
-          required
-        />
-        <p class="text-xs text-muted-foreground">
-          {{ isEditMode ? '创建后不可修改' : '用户将使用此名称请求模型' }}
-        </p>
-      </div>
-
-      <!-- 映射模式：选择源模型 -->
-      <div
-        v-else
-        class="space-y-2"
-      >
-        <Label>源模型 (用户请求的模型) *</Label>
-        <Select
-          v-model:open="sourceModelSelectOpen"
-          :model-value="form.alias"
-          :disabled="isEditMode"
-          @update:model-value="form.alias = $event"
-        >
-          <SelectTrigger
-            class="w-full"
-            :class="{ 'opacity-50': isEditMode }"
-          >
-            <SelectValue placeholder="请选择源模型" />
-          </SelectTrigger>
-          <SelectContent>
-            <SelectItem
-              v-for="model in availableSourceModels"
-              :key="model.id"
-              :value="model.name"
-            >
-              {{ model.display_name }} ({{ model.name }})
-            </SelectItem>
-          </SelectContent>
-        </Select>
-        <p class="text-xs text-muted-foreground">
-          {{ isEditMode ? '创建后不可修改' : '选择要被映射的源模型，计费将按此模型价格' }}
-        </p>
-      </div>
-
-      <!-- 目标模型选择 -->
-      <div class="space-y-2">
-        <Label>
-          {{ form.mapping_type === 'alias' ? '目标模型 *' : '目标模型 (实际处理请求) *' }}
-        </Label>
-        <!-- 固定目标模型时显示只读信息 -->
-        <div
-          v-if="fixedTargetModel"
-          class="px-3 py-2 border rounded-md bg-muted/50"
-        >
-          <span class="font-medium">{{ fixedTargetModel.display_name }}</span>
-          <span class="text-muted-foreground ml-1">({{ fixedTargetModel.name }})</span>
-        </div>
-        <!-- 否则显示下拉选择 -->
-        <Select
-          v-else
-          v-model:open="targetModelSelectOpen"
-          :model-value="form.global_model_id"
-          @update:model-value="form.global_model_id = $event"
-        >
-          <SelectTrigger class="w-full">
-            <SelectValue placeholder="请选择模型" />
-          </SelectTrigger>
-          <SelectContent>
-            <SelectItem
-              v-for="model in availableTargetModels"
-              :key="model.id"
-              :value="model.id"
-            >
-              {{ model.display_name }} ({{ model.name }})
-            </SelectItem>
-          </SelectContent>
-        </Select>
-      </div>
-    </form>
-
-    <template #footer>
-      <Button
-        type="button"
-        variant="outline"
-        @click="handleCancel"
-      >
-        取消
-      </Button>
-      <Button
-        :disabled="submitting"
-        @click="handleSubmit"
-      >
-        <Loader2
-          v-if="submitting"
-          class="w-4 h-4 mr-2 animate-spin"
-        />
-        {{ isEditMode ? '保存' : '创建' }}
-      </Button>
-    </template>
-  </Dialog>
-</template>
-
-<script setup lang="ts">
-import { ref, computed } from 'vue'
-import { Loader2, Tag, SquarePen } from 'lucide-vue-next'
-import { Dialog, Select, SelectTrigger, SelectValue, SelectContent, SelectItem } from '@/components/ui'
-import Button from '@/components/ui/button.vue'
-import Input from '@/components/ui/input.vue'
-import Label from '@/components/ui/label.vue'
-import { useToast } from '@/composables/useToast'
-import { useFormDialog } from '@/composables/useFormDialog'
-import type { ModelAlias, CreateModelAliasRequest, UpdateModelAliasRequest } from '@/api/endpoints/aliases'
-import type { GlobalModelResponse } from '@/api/global-models'
-
-export interface ProviderOption {
-  id: string
-  name: string
-  display_name?: string
-}
-
-interface AliasFormData {
-  alias: string
-  global_model_id: string
-  provider_id: string | null
-  mapping_type: 'alias' | 'mapping'
-  is_active: boolean
-}
-
-const props = withDefaults(defineProps<{
-  open: boolean
-  editingAlias?: ModelAlias | null
-  globalModels: GlobalModelResponse[]
-  providers?: ProviderOption[]
-  fixedTargetModel?: GlobalModelResponse | null  // 用于从模型详情抽屉打开时固定目标模型
-  fixedProvider?: ProviderOption | null  // 用于 Provider 特定别名固定 Provider
-  showProviderSelect?: boolean  // 是否显示 Provider 选择（默认 true）
-}>(), {
-  editingAlias: null,
-  providers: () => [],
-  fixedTargetModel: null,
-  fixedProvider: null,
-  showProviderSelect: true
-})
-
-const emit = defineEmits<{
-  'update:open': [value: boolean]
-  'submit': [data: CreateModelAliasRequest | UpdateModelAliasRequest, isEdit: boolean]
-}>()
-
-const { error: showError } = useToast()
-
-// 状态
-const submitting = ref(false)
-const providerSelectOpen = ref(false)
-const sourceModelSelectOpen = ref(false)
-const targetModelSelectOpen = ref(false)
-const form = ref<AliasFormData>({
-  alias: '',
-  global_model_id: '',
-  provider_id: null,
-  mapping_type: 'alias',
-  is_active: true,
-})
-
-// 处理 Provider 选择变化
-function handleProviderChange(value: string) {
-  form.value.provider_id = value === 'global' ? null : value
-}
-
-// 重置表单
-function resetForm() {
-  form.value = {
-    alias: '',
-    global_model_id: props.fixedTargetModel?.id || '',
-    provider_id: props.fixedProvider?.id || null,
-    mapping_type: 'alias',
-    is_active: true,
-  }
-}
-
-// 加载别名数据（编辑模式）
-function loadAliasData() {
-  if (!props.editingAlias) return
-  form.value = {
-    alias: props.editingAlias.alias,
-    global_model_id: props.editingAlias.global_model_id,
-    provider_id: props.editingAlias.provider_id,
-    mapping_type: props.editingAlias.mapping_type || 'alias',
-    is_active: props.editingAlias.is_active,
-  }
-}
-
-// 使用 useFormDialog 统一处理对话框逻辑
-const { isEditMode, handleDialogUpdate, handleCancel } = useFormDialog({
-  isOpen: () => props.open,
-  entity: () => props.editingAlias,
-  isLoading: submitting,
-  onClose: () => emit('update:open', false),
-  loadData: loadAliasData,
-  resetForm,
-})
-
-// 对话框标题
-const dialogTitle = computed(() => {
-  if (isEditMode.value) {
-    return form.value.mapping_type === 'mapping' ? '编辑映射' : '编辑别名'
-  }
-  if (props.fixedProvider) {
-    return `创建 ${props.fixedProvider.display_name || props.fixedProvider.name} 特定别名/映射`
-  }
-  return '创建别名/映射'
-})
-
-// 对话框描述
-const dialogDescription = computed(() => {
-  if (isEditMode.value) {
-    return form.value.mapping_type === 'mapping' ? '修改模型映射配置' : '修改别名设置'
-  }
-  return '为模型创建别名或映射规则'
-})
-
-// 对话框图标
-const dialogIcon = computed(() => isEditMode.value ? SquarePen : Tag)
-
-// 映射模式下可选的源模型（排除已选择的目标模型）
-const availableSourceModels = computed(() => {
-  return props.globalModels.filter(m => m.id !== form.value.global_model_id)
-})
-
-// 可选的目标模型（映射模式下排除已选择的源模型）
-const availableTargetModels = computed(() => {
-  if (form.value.mapping_type === 'mapping' && form.value.alias) {
-    // 找到源模型对应的 GlobalModel
-    const sourceModel = props.globalModels.find(m => m.name === form.value.alias)
-    if (sourceModel) {
-      return props.globalModels.filter(m => m.id !== sourceModel.id)
-    }
-  }
-  return props.globalModels
-})
-
-// 提交表单
-async function handleSubmit() {
-  if (!form.value.alias) {
-    showError(form.value.mapping_type === 'alias' ? '请输入别名名称' : '请选择源模型', '错误')
-    return
-  }
-
-  const targetModelId = props.fixedTargetModel?.id || form.value.global_model_id
-  if (!targetModelId) {
-    showError('请选择目标模型', '错误')
-    return
-  }
-
-  submitting.value = true
-  try {
-    const data: CreateModelAliasRequest | UpdateModelAliasRequest = {
-      alias: form.value.alias,
-      global_model_id: targetModelId,
-      provider_id: props.fixedProvider?.id || form.value.provider_id,
-      mapping_type: form.value.mapping_type,
-      is_active: form.value.is_active,
-    }
-
-    emit('submit', data, !!props.editingAlias)
-  } finally {
-    submitting.value = false
-  }
-}
-</script>

frontend/src/features/models/components/GlobalModelFormDialog.vue

@@ -2,174 +2,304 @@
   <Dialog
     :model-value="open"
     :title="isEditMode ? '编辑模型' : '创建统一模型'"
-    :description="isEditMode ? '修改模型配置和价格信息' : '添加一个新的全局模型定义'"
+    :description="isEditMode ? '修改模型配置和价格信息' : ''"
     :icon="isEditMode ? SquarePen : Layers"
-    size="xl"
+    size="3xl"
     @update:model-value="handleDialogUpdate"
   >
-    <form
-      class="space-y-5 max-h-[70vh] overflow-y-auto pr-1"
-      @submit.prevent="handleSubmit"
+    <div
+      class="flex gap-4"
+      :class="isEditMode ? '' : 'h-[500px]'"
     >
-      <!-- 基本信息 -->
-      <section class="space-y-3">
-        <h4 class="font-medium text-sm">
-          基本信息
-        </h4>
-
-        <div class="grid grid-cols-2 gap-3">
-          <div class="space-y-1.5">
-            <Label
-              for="model-name"
-              class="text-xs"
-            >模型名称 *</Label>
-            <Input
-              id="model-name"
-              v-model="form.name"
-              placeholder="claude-3-5-sonnet-20241022"
-              :disabled="isEditMode"
-              required
-            />
-            <p
-              v-if="!isEditMode"
-              class="text-xs text-muted-foreground"
-            >
-              创建后不可修改
-            </p>
-          </div>
-          <div class="space-y-1.5">
-            <Label
-              for="model-display-name"
-              class="text-xs"
-            >显示名称 *</Label>
-            <Input
-              id="model-display-name"
-              v-model="form.display_name"
-              placeholder="Claude 3.5 Sonnet"
-              required
-            />
-          </div>
-        </div>
-
-        <div class="space-y-1.5">
-          <Label
-            for="model-description"
-            class="text-xs"
-          >描述</Label>
-          <Input
-            id="model-description"
-            v-model="form.description"
-            placeholder="简短描述此模型的特点"
-          />
-        </div>
-      </section>
-
-      <!-- 能力配置 -->
-      <section class="space-y-2">
-        <h4 class="font-medium text-sm">
-          默认能力
-        </h4>
-        <div class="flex flex-wrap gap-2">
-          <label class="flex items-center gap-2 px-3 py-1.5 rounded-md border border-border bg-muted/30 cursor-pointer text-sm">
-            <input
-              v-model="form.default_supports_streaming"
-              type="checkbox"
-              class="rounded"
-            >
-            <Zap class="w-3.5 h-3.5 text-muted-foreground" />
-            <span>流式输出</span>
-          </label>
-          <label class="flex items-center gap-2 px-3 py-1.5 rounded-md border border-border bg-muted/30 cursor-pointer text-sm">
-            <input
-              v-model="form.default_supports_vision"
-              type="checkbox"
-              class="rounded"
-            >
-            <Eye class="w-3.5 h-3.5 text-muted-foreground" />
-            <span>视觉理解</span>
-          </label>
-          <label class="flex items-center gap-2 px-3 py-1.5 rounded-md border border-border bg-muted/30 cursor-pointer text-sm">
-            <input
-              v-model="form.default_supports_function_calling"
-              type="checkbox"
-              class="rounded"
-            >
-            <Wrench class="w-3.5 h-3.5 text-muted-foreground" />
-            <span>工具调用</span>
-          </label>
-          <label class="flex items-center gap-2 px-3 py-1.5 rounded-md border border-border bg-muted/30 cursor-pointer text-sm">
-            <input
-              v-model="form.default_supports_extended_thinking"
-              type="checkbox"
-              class="rounded"
-            >
-            <Brain class="w-3.5 h-3.5 text-muted-foreground" />
-            <span>深度思考</span>
-          </label>
-          <label class="flex items-center gap-2 px-3 py-1.5 rounded-md border border-border bg-muted/30 cursor-pointer text-sm">
-            <input
-              v-model="form.default_supports_image_generation"
-              type="checkbox"
-              class="rounded"
-            >
-            <Image class="w-3.5 h-3.5 text-muted-foreground" />
-            <span>图像生成</span>
-          </label>
-        </div>
-      </section>
-
-      <!-- Key 能力配置 -->
-      <section
-        v-if="availableCapabilities.length > 0"
-        class="space-y-2"
+      <!-- 左侧：模型选择（仅创建模式） -->
+      <div
+        v-if="!isEditMode"
+        class="w-[260px] shrink-0 flex flex-col h-full"
       >
-        <h4 class="font-medium text-sm">
-          Key 能力支持
-        </h4>
-        <div class="flex flex-wrap gap-2">
-          <label
-            v-for="cap in availableCapabilities"
-            :key="cap.name"
-            class="flex items-center gap-2 px-3 py-1.5 rounded-md border border-border bg-muted/30 cursor-pointer text-sm"
-          >
-            <input
-              type="checkbox"
-              :checked="form.supported_capabilities?.includes(cap.name)"
-              class="rounded"
-              @change="toggleCapability(cap.name)"
-            >
-            <span>{{ cap.display_name }}</span>
-          </label>
-        </div>
-      </section>
-
-      <!-- 价格配置 -->
-      <section class="space-y-3">
-        <h4 class="font-medium text-sm">
-          价格配置
-        </h4>
-        <TieredPricingEditor
-          ref="tieredPricingEditorRef"
-          v-model="tieredPricing"
-          :show-cache1h="form.supported_capabilities?.includes('cache_1h')"
-        />
-
-        <!-- 按次计费 -->
-        <div class="flex items-center gap-3 pt-2 border-t">
-          <Label class="text-xs whitespace-nowrap">按次计费 ($/次)</Label>
+        <!-- 搜索框 -->
+        <div class="relative mb-3">
+          <Search class="absolute left-2.5 top-1/2 -translate-y-1/2 h-4 w-4 text-muted-foreground" />
           <Input
-            :model-value="form.default_price_per_request ?? ''"
-            type="number"
-            step="0.001"
-            min="0"
-            class="w-32"
-            placeholder="留空不启用"
-            @update:model-value="(v) => form.default_price_per_request = parseNumberInput(v, { allowFloat: true })"
+            v-model="searchQuery"
+            type="text"
+            placeholder="搜索模型、提供商..."
+            class="pl-8 h-8 text-sm"
           />
-          <span class="text-xs text-muted-foreground">每次请求固定费用，可与 Token 计费叠加</span>
         </div>
-      </section>
-    </form>
+
+        <!-- 模型列表（两级结构） -->
+        <div class="flex-1 overflow-y-auto border rounded-lg min-h-0 scrollbar-thin">
+          <div
+            v-if="loading"
+            class="flex items-center justify-center h-32"
+          >
+            <Loader2 class="w-5 h-5 animate-spin text-muted-foreground" />
+          </div>
+          <template v-else>
+            <!-- 提供商分组 -->
+            <div
+              v-for="group in groupedModels"
+              :key="group.providerId"
+              class="border-b last:border-b-0"
+            >
+              <!-- 提供商标题行 -->
+              <div
+                class="flex items-center gap-2 px-2.5 py-2 cursor-pointer hover:bg-muted text-sm"
+                @click="toggleProvider(group.providerId)"
+              >
+                <ChevronRight
+                  class="w-3.5 h-3.5 text-muted-foreground transition-transform shrink-0"
+                  :class="expandedProvider === group.providerId ? 'rotate-90' : ''"
+                />
+                <img
+                  :src="getProviderLogoUrl(group.providerId)"
+                  :alt="group.providerName"
+                  class="w-4 h-4 rounded shrink-0 dark:invert dark:brightness-90"
+                  @error="handleLogoError"
+                >
+                <span class="truncate font-medium text-xs flex-1">{{ group.providerName }}</span>
+                <span class="text-[10px] text-muted-foreground shrink-0">{{ group.models.length }}</span>
+              </div>
+              <!-- 模型列表 -->
+              <div
+                v-if="expandedProvider === group.providerId"
+                class="bg-muted/30"
+              >
+                <div
+                  v-for="model in group.models"
+                  :key="model.modelId"
+                  class="flex flex-col gap-0.5 pl-7 pr-2.5 py-1.5 cursor-pointer text-xs border-t"
+                  :class="selectedModel?.modelId === model.modelId && selectedModel?.providerId === model.providerId
+                    ? 'bg-primary text-primary-foreground'
+                    : 'hover:bg-muted'"
+                  @click="selectModel(model)"
+                >
+                  <span class="truncate font-medium">{{ model.modelName }}</span>
+                  <span
+                    class="truncate text-[10px]"
+                    :class="selectedModel?.modelId === model.modelId && selectedModel?.providerId === model.providerId
+                      ? 'text-primary-foreground/70'
+                      : 'text-muted-foreground'"
+                  >{{ model.modelId }}</span>
+                </div>
+              </div>
+            </div>
+            <div
+              v-if="groupedModels.length === 0"
+              class="text-center py-8 text-sm text-muted-foreground"
+            >
+              {{ searchQuery ? '未找到模型' : '加载中...' }}
+            </div>
+          </template>
+        </div>
+      </div>
+
+      <!-- 右侧：表单 -->
+      <div
+        class="flex-1 overflow-y-auto h-full scrollbar-thin"
+        :class="isEditMode ? 'max-h-[70vh]' : ''"
+      >
+        <form
+          class="space-y-5"
+          @submit.prevent="handleSubmit"
+        >
+          <!-- 基本信息 -->
+          <section class="space-y-3">
+            <h4 class="font-medium text-sm">
+              基本信息
+            </h4>
+            <div class="grid grid-cols-2 gap-3">
+              <div class="space-y-1.5">
+                <Label
+                  for="model-name"
+                  class="text-xs"
+                >模型名称 *</Label>
+                <Input
+                  id="model-name"
+                  v-model="form.name"
+                  placeholder="claude-3-5-sonnet-20241022"
+                  :disabled="isEditMode"
+                  required
+                />
+              </div>
+              <div class="space-y-1.5">
+                <Label
+                  for="model-display-name"
+                  class="text-xs"
+                >显示名称 *</Label>
+                <Input
+                  id="model-display-name"
+                  v-model="form.display_name"
+                  placeholder="Claude 3.5 Sonnet"
+                  required
+                />
+              </div>
+            </div>
+            <div class="space-y-1.5">
+              <Label
+                for="model-description"
+                class="text-xs"
+              >描述</Label>
+              <Input
+                id="model-description"
+                :model-value="form.config?.description || ''"
+                placeholder="简短描述此模型的特点"
+                @update:model-value="(v) => setConfigField('description', v || undefined)"
+              />
+            </div>
+            <div class="grid grid-cols-3 gap-3">
+              <div class="space-y-1.5">
+                <Label
+                  for="model-family"
+                  class="text-xs"
+                >模型系列</Label>
+                <Input
+                  id="model-family"
+                  :model-value="form.config?.family || ''"
+                  placeholder="如 GPT-4、Claude 3"
+                  @update:model-value="(v) => setConfigField('family', v || undefined)"
+                />
+              </div>
+              <div class="space-y-1.5">
+                <Label
+                  for="model-context-limit"
+                  class="text-xs"
+                >上下文限制</Label>
+                <Input
+                  id="model-context-limit"
+                  type="number"
+                  :model-value="form.config?.context_limit ?? ''"
+                  placeholder="如 128000"
+                  @update:model-value="(v) => setConfigField('context_limit', v ? Number(v) : undefined)"
+                />
+              </div>
+              <div class="space-y-1.5">
+                <Label
+                  for="model-output-limit"
+                  class="text-xs"
+                >输出限制</Label>
+                <Input
+                  id="model-output-limit"
+                  type="number"
+                  :model-value="form.config?.output_limit ?? ''"
+                  placeholder="如 8192"
+                  @update:model-value="(v) => setConfigField('output_limit', v ? Number(v) : undefined)"
+                />
+              </div>
+            </div>
+          </section>
+
+          <!-- 能力配置 -->
+          <section class="space-y-2">
+            <h4 class="font-medium text-sm">
+              默认能力
+            </h4>
+            <div class="flex flex-wrap gap-2">
+              <label class="flex items-center gap-2 px-2.5 py-1 rounded-md border bg-muted/30 cursor-pointer text-sm">
+                <input
+                  type="checkbox"
+                  :checked="form.config?.streaming !== false"
+                  class="rounded"
+                  @change="setConfigField('streaming', ($event.target as HTMLInputElement).checked)"
+                >
+                <Zap class="w-3.5 h-3.5 text-muted-foreground" />
+                <span>流式</span>
+              </label>
+              <label class="flex items-center gap-2 px-2.5 py-1 rounded-md border bg-muted/30 cursor-pointer text-sm">
+                <input
+                  type="checkbox"
+                  :checked="form.config?.vision === true"
+                  class="rounded"
+                  @change="setConfigField('vision', ($event.target as HTMLInputElement).checked)"
+                >
+                <Eye class="w-3.5 h-3.5 text-muted-foreground" />
+                <span>视觉</span>
+              </label>
+              <label class="flex items-center gap-2 px-2.5 py-1 rounded-md border bg-muted/30 cursor-pointer text-sm">
+                <input
+                  type="checkbox"
+                  :checked="form.config?.function_calling === true"
+                  class="rounded"
+                  @change="setConfigField('function_calling', ($event.target as HTMLInputElement).checked)"
+                >
+                <Wrench class="w-3.5 h-3.5 text-muted-foreground" />
+                <span>工具</span>
+              </label>
+              <label class="flex items-center gap-2 px-2.5 py-1 rounded-md border bg-muted/30 cursor-pointer text-sm">
+                <input
+                  type="checkbox"
+                  :checked="form.config?.extended_thinking === true"
+                  class="rounded"
+                  @change="setConfigField('extended_thinking', ($event.target as HTMLInputElement).checked)"
+                >
+                <Brain class="w-3.5 h-3.5 text-muted-foreground" />
+                <span>思考</span>
+              </label>
+              <label class="flex items-center gap-2 px-2.5 py-1 rounded-md border bg-muted/30 cursor-pointer text-sm">
+                <input
+                  type="checkbox"
+                  :checked="form.config?.image_generation === true"
+                  class="rounded"
+                  @change="setConfigField('image_generation', ($event.target as HTMLInputElement).checked)"
+                >
+                <Image class="w-3.5 h-3.5 text-muted-foreground" />
+                <span>生图</span>
+              </label>
+            </div>
+          </section>
+
+          <!-- Key 能力配置 -->
+          <section
+            v-if="availableCapabilities.length > 0"
+            class="space-y-2"
+          >
+            <h4 class="font-medium text-sm">
+              Key 能力支持
+            </h4>
+            <div class="flex flex-wrap gap-2">
+              <label
+                v-for="cap in availableCapabilities"
+                :key="cap.name"
+                class="flex items-center gap-2 px-2.5 py-1 rounded-md border bg-muted/30 cursor-pointer text-sm"
+              >
+                <input
+                  type="checkbox"
+                  :checked="form.supported_capabilities?.includes(cap.name)"
+                  class="rounded"
+                  @change="toggleCapability(cap.name)"
+                >
+                <span>{{ cap.display_name }}</span>
+              </label>
+            </div>
+          </section>
+
+          <!-- 价格配置 -->
+          <section class="space-y-3">
+            <h4 class="font-medium text-sm">
+              价格配置
+            </h4>
+            <TieredPricingEditor
+              ref="tieredPricingEditorRef"
+              v-model="tieredPricing"
+              :show-cache1h="form.supported_capabilities?.includes('cache_1h')"
+            />
+            <div class="flex items-center gap-3 pt-2 border-t">
+              <Label class="text-xs whitespace-nowrap">按次计费</Label>
+              <Input
+                :model-value="form.default_price_per_request ?? ''"
+                type="number"
+                step="0.001"
+                min="0"
+                class="w-24"
+                placeholder="$/次"
+                @update:model-value="(v) => form.default_price_per_request = parseNumberInput(v, { allowFloat: true })"
+              />
+              <span class="text-xs text-muted-foreground">可与 Token 计费叠加</span>
+            </div>
+          </section>
+        </form>
+      </div>
+    </div>
 
     <template #footer>
       <Button
@@ -180,7 +310,7 @@
         取消
       </Button>
       <Button
-        :disabled="submitting"
+        :disabled="submitting || !form.name || !form.display_name"
         @click="handleSubmit"
       >
         <Loader2
@@ -189,19 +319,35 @@
         />
         {{ isEditMode ? '保存' : '创建' }}
       </Button>
+      <Button
+        v-if="selectedModel && !isEditMode"
+        type="button"
+        variant="ghost"
+        @click="clearSelection"
+      >
+        清空
+      </Button>
     </template>
   </Dialog>
 </template>
 
 <script setup lang="ts">
-import { ref, onMounted } from 'vue'
-import { Eye, Wrench, Brain, Zap, Image, Loader2, Layers, SquarePen } from 'lucide-vue-next'
+import { ref, computed, onMounted, watch } from 'vue'
+import {
+  Eye, Wrench, Brain, Zap, Image, Loader2, Layers, SquarePen,
+  Search, ChevronRight
+} from 'lucide-vue-next'
 import { Dialog, Button, Input, Label } from '@/components/ui'
 import { useToast } from '@/composables/useToast'
 import { useFormDialog } from '@/composables/useFormDialog'
 import { parseNumberInput } from '@/utils/form'
 import { log } from '@/utils/logger'
 import TieredPricingEditor from './TieredPricingEditor.vue'
+import {
+  getModelsDevList,
+  getProviderLogoUrl,
+  type ModelsDevModelItem,
+} from '@/api/models-dev'
 import {
   createGlobalModel,
   updateGlobalModel,
@@ -226,42 +372,147 @@ const { success, error: showError } = useToast()
 const submitting = ref(false)
 const tieredPricingEditorRef = ref<InstanceType<typeof TieredPricingEditor> | null>(null)
 
-// 阶梯计费配置（统一使用，固定价格就是单阶梯）
+// 模型列表相关
+const loading = ref(false)
+const searchQuery = ref('')
+const allModelsCache = ref<ModelsDevModelItem[]>([]) // 全部模型（缓存）
+const selectedModel = ref<ModelsDevModelItem | null>(null)
+const expandedProvider = ref<string | null>(null)
+
+// 当前显示的模型列表：有搜索词时用全部，否则只用官方
+const allModels = computed(() => {
+  if (searchQuery.value) {
+    return allModelsCache.value
+  }
+  return allModelsCache.value.filter(m => m.official)
+})
+
+// 按提供商分组的模型
+interface ProviderGroup {
+  providerId: string
+  providerName: string
+  models: ModelsDevModelItem[]
+}
+
+const groupedModels = computed(() => {
+  let models = allModels.value.filter(m => !m.deprecated)
+  // 搜索（支持空格分隔的多关键词 AND 搜索）
+  if (searchQuery.value) {
+    const keywords = searchQuery.value.toLowerCase().split(/\s+/).filter(k => k.length > 0)
+    models = models.filter(model => {
+      const searchableText = `${model.providerId} ${model.providerName} ${model.modelId} ${model.modelName} ${model.family || ''}`.toLowerCase()
+      return keywords.every(keyword => searchableText.includes(keyword))
+    })
+  }
+
+  // 按提供商分组
+  const groups = new Map<string, ProviderGroup>()
+  for (const model of models) {
+    if (!groups.has(model.providerId)) {
+      groups.set(model.providerId, {
+        providerId: model.providerId,
+        providerName: model.providerName,
+        models: []
+      })
+    }
+    groups.get(model.providerId)!.models.push(model)
+  }
+
+  // 转换为数组并排序
+  const result = Array.from(groups.values())
+
+  // 如果有搜索词，把提供商名称/ID匹配的排在前面
+  if (searchQuery.value) {
+    const keywords = searchQuery.value.toLowerCase().split(/\s+/).filter(k => k.length > 0)
+    result.sort((a, b) => {
+      const aText = `${a.providerId} ${a.providerName}`.toLowerCase()
+      const bText = `${b.providerId} ${b.providerName}`.toLowerCase()
+      const aProviderMatch = keywords.some(k => aText.includes(k))
+      const bProviderMatch = keywords.some(k => bText.includes(k))
+      if (aProviderMatch && !bProviderMatch) return -1
+      if (!aProviderMatch && bProviderMatch) return 1
+      return a.providerName.localeCompare(b.providerName)
+    })
+  } else {
+    result.sort((a, b) => a.providerName.localeCompare(b.providerName))
+  }
+
+  return result
+})
+
+// 搜索时如果只有一个提供商，自动展开
+watch(groupedModels, (groups) => {
+  if (searchQuery.value && groups.length === 1) {
+    expandedProvider.value = groups[0].providerId
+  }
+})
+
+// 切换提供商展开状态
+function toggleProvider(providerId: string) {
+  expandedProvider.value = expandedProvider.value === providerId ? null : providerId
+}
+
+// 阶梯计费配置
 const tieredPricing = ref<TieredPricingConfig | null>(null)
 
 interface FormData {
   name: string
   display_name: string
-  description?: string
   default_price_per_request?: number
-  default_supports_streaming?: boolean
-  default_supports_image_generation?: boolean
-  default_supports_vision?: boolean
-  default_supports_function_calling?: boolean
-  default_supports_extended_thinking?: boolean
   supported_capabilities?: string[]
+  config?: Record<string, any>
   is_active?: boolean
 }
 
 const defaultForm = (): FormData => ({
   name: '',
   display_name: '',
-  description: '',
   default_price_per_request: undefined,
-  default_supports_streaming: true,
-  default_supports_image_generation: false,
-  default_supports_vision: false,
-  default_supports_function_calling: false,
-  default_supports_extended_thinking: false,
   supported_capabilities: [],
+  config: { streaming: true },
   is_active: true,
 })
 
 const form = ref<FormData>(defaultForm())
 
+const KEEP_FALSE_CONFIG_KEYS = new Set(['streaming'])
+
+// 设置 config 字段
+function setConfigField(key: string, value: any) {
+  if (!form.value.config) {
+    form.value.config = {}
+  }
+  if (value === undefined || value === '' || (value === false && !KEEP_FALSE_CONFIG_KEYS.has(key))) {
+    delete form.value.config[key]
+  } else {
+    form.value.config[key] = value
+  }
+}
+
 // Key 能力选项
 const availableCapabilities = ref<CapabilityDefinition[]>([])
 
+// 加载模型列表
+async function loadModels() {
+  if (allModelsCache.value.length > 0) return
+  loading.value = true
+  try {
+    // 只加载一次全部模型，过滤在 computed 中完成
+    allModelsCache.value = await getModelsDevList(false)
+  } catch (err) {
+    log.error('Failed to load models:', err)
+  } finally {
+    loading.value = false
+  }
+}
+
+// 打开对话框时加载数据
+watch(() => props.open, (isOpen) => {
+  if (isOpen && !props.model) {
+    loadModels()
+  }
+})
+
 // 加载可用能力列表
 async function loadCapabilities() {
   try {
@@ -284,38 +535,92 @@ function toggleCapability(capName: string) {
   }
 }
 
-// 组件挂载时加载能力列表
 onMounted(() => {
   loadCapabilities()
 })
 
+// 选择模型并填充表单
+function selectModel(model: ModelsDevModelItem) {
+  selectedModel.value = model
+  expandedProvider.value = model.providerId
+  form.value.name = model.modelId
+  form.value.display_name = model.modelName
+
+  // 构建 config
+  const config: Record<string, any> = {
+    streaming: true,
+  }
+  if (model.supportsVision) config.vision = true
+  if (model.supportsToolCall) config.function_calling = true
+  if (model.supportsReasoning) config.extended_thinking = true
+  if (model.supportsStructuredOutput) config.structured_output = true
+  if (model.supportsTemperature !== false) config.temperature = model.supportsTemperature
+  if (model.supportsAttachment) config.attachment = true
+  if (model.openWeights) config.open_weights = true
+  if (model.contextLimit) config.context_limit = model.contextLimit
+  if (model.outputLimit) config.output_limit = model.outputLimit
+  if (model.knowledgeCutoff) config.knowledge_cutoff = model.knowledgeCutoff
+  if (model.family) config.family = model.family
+  if (model.releaseDate) config.release_date = model.releaseDate
+  if (model.inputModalities?.length) config.input_modalities = model.inputModalities
+  if (model.outputModalities?.length) config.output_modalities = model.outputModalities
+  form.value.config = config
+
+  if (model.inputPrice !== undefined || model.outputPrice !== undefined) {
+    tieredPricing.value = {
+      tiers: [{
+        up_to: null,
+        input_price_per_1m: model.inputPrice || 0,
+        output_price_per_1m: model.outputPrice || 0,
+      }]
+    }
+  } else {
+    tieredPricing.value = null
+  }
+}
+
+// 清除选择（手动填写）
+function clearSelection() {
+  selectedModel.value = null
+  form.value = defaultForm()
+  tieredPricing.value = null
+}
+
+// Logo 加载失败处理
+function handleLogoError(event: Event) {
+  const img = event.target as HTMLImageElement
+  img.style.display = 'none'
+}
+
 // 重置表单
 function resetForm() {
   form.value = defaultForm()
   tieredPricing.value = null
+  searchQuery.value = ''
+  selectedModel.value = null
+  expandedProvider.value = null
 }
 
 // 加载模型数据（编辑模式）
 function loadModelData() {
   if (!props.model) return
+  // 先重置创建模式的残留状态
+  selectedModel.value = null
+  searchQuery.value = ''
+  expandedProvider.value = null
+
   form.value = {
     name: props.model.name,
     display_name: props.model.display_name,
-    description: props.model.description,
     default_price_per_request: props.model.default_price_per_request,
-    default_supports_streaming: props.model.default_supports_streaming,
-    default_supports_image_generation: props.model.default_supports_image_generation,
-    default_supports_vision: props.model.default_supports_vision,
-    default_supports_function_calling: props.model.default_supports_function_calling,
-    default_supports_extended_thinking: props.model.default_supports_extended_thinking,
     supported_capabilities: [...(props.model.supported_capabilities || [])],
+    config: props.model.config ? { ...props.model.config } : { streaming: true },
     is_active: props.model.is_active,
   }
-
-  // 加载阶梯计费配置（深拷贝）
-  if (props.model.default_tiered_pricing) {
-    tieredPricing.value = JSON.parse(JSON.stringify(props.model.default_tiered_pricing))
-  }
+  // 确保 tieredPricing 也被正确设置或重置
+  tieredPricing.value = props.model.default_tiered_pricing
+    ? JSON.parse(JSON.stringify(props.model.default_tiered_pricing))
+    : null
 }
 
 // 使用 useFormDialog 统一处理对话框逻辑
@@ -339,24 +644,22 @@ async function handleSubmit() {
     return
   }
 
-  // 获取包含自动计算缓存价格的最终数据
   const finalTiers = tieredPricingEditorRef.value?.getFinalTiers()
   const finalTieredPricing = finalTiers ? { tiers: finalTiers } : tieredPricing.value
 
+  // 清理空的 config
+  const cleanConfig = form.value.config && Object.keys(form.value.config).length > 0
+    ? form.value.config
+    : undefined
+
   submitting.value = true
   try {
     if (isEditMode.value && props.model) {
       const updateData: GlobalModelUpdate = {
         display_name: form.value.display_name,
-        description: form.value.description,
-        // 使用 null 而不是 undefined 来显式清空字段
+        config: cleanConfig || null,
         default_price_per_request: form.value.default_price_per_request ?? null,
         default_tiered_pricing: finalTieredPricing,
-        default_supports_streaming: form.value.default_supports_streaming,
-        default_supports_image_generation: form.value.default_supports_image_generation,
-        default_supports_vision: form.value.default_supports_vision,
-        default_supports_function_calling: form.value.default_supports_function_calling,
-        default_supports_extended_thinking: form.value.default_supports_extended_thinking,
         supported_capabilities: form.value.supported_capabilities?.length ? form.value.supported_capabilities : null,
         is_active: form.value.is_active,
       }
@@ -366,14 +669,9 @@ async function handleSubmit() {
       const createData: GlobalModelCreate = {
         name: form.value.name!,
         display_name: form.value.display_name!,
-        description: form.value.description,
-        default_price_per_request: form.value.default_price_per_request || undefined,
+        config: cleanConfig,
+        default_price_per_request: form.value.default_price_per_request ?? undefined,
         default_tiered_pricing: finalTieredPricing,
-        default_supports_streaming: form.value.default_supports_streaming,
-        default_supports_image_generation: form.value.default_supports_image_generation,
-        default_supports_vision: form.value.default_supports_vision,
-        default_supports_function_calling: form.value.default_supports_function_calling,
-        default_supports_extended_thinking: form.value.default_supports_extended_thinking,
         supported_capabilities: form.value.supported_capabilities?.length ? form.value.supported_capabilities : undefined,
         is_active: form.value.is_active,
       }

frontend/src/features/models/components/ModelDetailDrawer.vue

@@ -38,12 +38,12 @@
                   >
                     <Copy class="w-3 h-3" />
                   </button>
-                  <template v-if="model.description">
+                  <template v-if="model.config?.description">
                     <span class="shrink-0">·</span>
                     <span
                       class="text-xs truncate"
-                      :title="model.description"
-                    >{{ model.description }}</span>
+                      :title="model.config?.description"
+                    >{{ model.config?.description }}</span>
                   </template>
                 </div>
               </div>
@@ -104,19 +104,6 @@
                 <span class="hidden sm:inline">关联提供商</span>
                 <span class="sm:hidden">提供商</span>
               </button>
-              <button
-                type="button"
-                class="flex-1 px-2 sm:px-4 py-2 text-xs sm:text-sm font-medium rounded-md transition-all duration-200"
-                :class="[
-                  detailTab === 'aliases'
-                    ? 'bg-primary text-primary-foreground shadow-sm'
-                    : 'text-muted-foreground hover:text-foreground hover:bg-background/50'
-                ]"
-                @click="detailTab = 'aliases'"
-              >
-                <span class="hidden sm:inline">别名/映射</span>
-                <span class="sm:hidden">别名</span>
-              </button>
             </div>
 
             <!-- Tab 内容 -->
@@ -156,10 +143,10 @@
                       </p>
                     </div>
                     <Badge
-                      :variant="model.default_supports_streaming ?? false ? 'default' : 'secondary'"
+                      :variant="model.config?.streaming !== false ? 'default' : 'secondary'"
                       class="text-xs"
                     >
-                      {{ model.default_supports_streaming ?? false ? '支持' : '不支持' }}
+                      {{ model.config?.streaming !== false ? '支持' : '不支持' }}
                     </Badge>
                   </div>
                   <div class="flex items-center gap-2 p-3 rounded-lg border">
@@ -173,10 +160,10 @@
                       </p>
                     </div>
                     <Badge
-                      :variant="model.default_supports_image_generation ?? false ? 'default' : 'secondary'"
+                      :variant="model.config?.image_generation === true ? 'default' : 'secondary'"
                       class="text-xs"
                     >
-                      {{ model.default_supports_image_generation ?? false ? '支持' : '不支持' }}
+                      {{ model.config?.image_generation === true ? '支持' : '不支持' }}
                     </Badge>
                   </div>
                   <div class="flex items-center gap-2 p-3 rounded-lg border">
@@ -190,10 +177,10 @@
                       </p>
                     </div>
                     <Badge
-                      :variant="model.default_supports_vision ?? false ? 'default' : 'secondary'"
+                      :variant="model.config?.vision === true ? 'default' : 'secondary'"
                       class="text-xs"
                     >
-                      {{ model.default_supports_vision ?? false ? '支持' : '不支持' }}
+                      {{ model.config?.vision === true ? '支持' : '不支持' }}
                     </Badge>
                   </div>
                   <div class="flex items-center gap-2 p-3 rounded-lg border">
@@ -207,10 +194,10 @@
                       </p>
                     </div>
                     <Badge
-                      :variant="model.default_supports_function_calling ?? false ? 'default' : 'secondary'"
+                      :variant="model.config?.function_calling === true ? 'default' : 'secondary'"
                       class="text-xs"
                     >
-                      {{ model.default_supports_function_calling ?? false ? '支持' : '不支持' }}
+                      {{ model.config?.function_calling === true ? '支持' : '不支持' }}
                     </Badge>
                   </div>
                   <div class="flex items-center gap-2 p-3 rounded-lg border">
@@ -224,10 +211,10 @@
                       </p>
                     </div>
                     <Badge
-                      :variant="model.default_supports_extended_thinking ?? false ? 'default' : 'secondary'"
+                      :variant="model.config?.extended_thinking === true ? 'default' : 'secondary'"
                       class="text-xs"
                     >
-                      {{ model.default_supports_extended_thinking ?? false ? '支持' : '不支持' }}
+                      {{ model.config?.extended_thinking === true ? '支持' : '不支持' }}
                     </Badge>
                   </div>
                 </div>
@@ -409,11 +396,11 @@
                   </div>
                   <div class="p-3 rounded-lg border bg-muted/20">
                     <div class="flex items-center justify-between">
-                      <Label class="text-xs text-muted-foreground">别名数量</Label>
-                      <Tag class="w-4 h-4 text-muted-foreground" />
+                      <Label class="text-xs text-muted-foreground">调用次数</Label>
+                      <BarChart3 class="w-4 h-4 text-muted-foreground" />
                     </div>
                     <p class="text-2xl font-bold mt-1">
-                      {{ model.alias_count || 0 }}
+                      {{ model.usage_count || 0 }}
                     </p>
                   </div>
                 </div>
@@ -468,105 +455,153 @@
                 <template v-else-if="providers.length > 0">
                   <!-- 桌面端表格 -->
                   <Table class="hidden sm:table">
-                  <TableHeader>
-                    <TableRow class="border-b border-border/60 hover:bg-transparent">
-                      <TableHead class="h-10 font-semibold">
-                        Provider
-                      </TableHead>
-                      <TableHead class="w-[120px] h-10 font-semibold">
-                        能力
-                      </TableHead>
-                      <TableHead class="w-[180px] h-10 font-semibold">
-                        价格 ($/M)
-                      </TableHead>
-                      <TableHead class="w-[80px] h-10 font-semibold text-center">
-                        操作
-                      </TableHead>
-                    </TableRow>
-                  </TableHeader>
-                  <TableBody>
-                    <TableRow
+                    <TableHeader>
+                      <TableRow class="border-b border-border/60 hover:bg-transparent">
+                        <TableHead class="h-10 font-semibold">
+                          Provider
+                        </TableHead>
+                        <TableHead class="w-[120px] h-10 font-semibold">
+                          能力
+                        </TableHead>
+                        <TableHead class="w-[180px] h-10 font-semibold">
+                          价格 ($/M)
+                        </TableHead>
+                        <TableHead class="w-[80px] h-10 font-semibold text-center">
+                          操作
+                        </TableHead>
+                      </TableRow>
+                    </TableHeader>
+                    <TableBody>
+                      <TableRow
+                        v-for="provider in providers"
+                        :key="provider.id"
+                        class="border-b border-border/40 hover:bg-muted/30 transition-colors"
+                      >
+                        <TableCell class="py-3">
+                          <div class="flex items-center gap-2">
+                            <span
+                              class="w-2 h-2 rounded-full shrink-0"
+                              :class="provider.is_active ? 'bg-green-500' : 'bg-gray-300'"
+                              :title="provider.is_active ? '活跃' : '停用'"
+                            />
+                            <span class="font-medium truncate">{{ provider.display_name }}</span>
+                          </div>
+                        </TableCell>
+                        <TableCell class="py-3">
+                          <div class="flex gap-0.5">
+                            <Zap
+                              v-if="provider.supports_streaming"
+                              class="w-3.5 h-3.5 text-muted-foreground"
+                              title="流式输出"
+                            />
+                            <Eye
+                              v-if="provider.supports_vision"
+                              class="w-3.5 h-3.5 text-muted-foreground"
+                              title="视觉理解"
+                            />
+                            <Wrench
+                              v-if="provider.supports_function_calling"
+                              class="w-3.5 h-3.5 text-muted-foreground"
+                              title="工具调用"
+                            />
+                          </div>
+                        </TableCell>
+                        <TableCell class="py-3">
+                          <div class="text-xs font-mono space-y-0.5">
+                            <!-- Token 计费：输入/输出 -->
+                            <div v-if="(provider.input_price_per_1m || 0) > 0 || (provider.output_price_per_1m || 0) > 0">
+                              <span class="text-muted-foreground">输入/输出:</span>
+                              <span class="ml-1">${{ (provider.input_price_per_1m || 0).toFixed(1) }}/${{ (provider.output_price_per_1m || 0).toFixed(1) }}</span>
+                              <!-- 阶梯标记 -->
+                              <span
+                                v-if="(provider.tier_count || 1) > 1"
+                                class="ml-1 text-muted-foreground"
+                                title="阶梯计费"
+                              >[阶梯]</span>
+                            </div>
+                            <!-- 缓存价格 -->
+                            <div
+                              v-if="(provider.cache_creation_price_per_1m || 0) > 0 || (provider.cache_read_price_per_1m || 0) > 0"
+                              class="text-muted-foreground"
+                            >
+                              <span>缓存:</span>
+                              <span class="ml-1">${{ (provider.cache_creation_price_per_1m || 0).toFixed(2) }}/${{ (provider.cache_read_price_per_1m || 0).toFixed(2) }}</span>
+                            </div>
+                            <!-- 1h 缓存价格 -->
+                            <div
+                              v-if="(provider.cache_1h_creation_price_per_1m || 0) > 0"
+                              class="text-muted-foreground"
+                            >
+                              <span>1h 缓存:</span>
+                              <span class="ml-1">${{ (provider.cache_1h_creation_price_per_1m || 0).toFixed(2) }}</span>
+                            </div>
+                            <!-- 按次计费 -->
+                            <div v-if="(provider.price_per_request || 0) > 0">
+                              <span class="text-muted-foreground">按次:</span>
+                              <span class="ml-1">${{ (provider.price_per_request || 0).toFixed(3) }}/次</span>
+                            </div>
+                            <!-- 无定价 -->
+                            <span
+                              v-if="!(provider.input_price_per_1m || 0) && !(provider.output_price_per_1m || 0) && !(provider.price_per_request || 0)"
+                              class="text-muted-foreground"
+                            >-</span>
+                          </div>
+                        </TableCell>
+                        <TableCell class="py-3 text-center">
+                          <div class="flex items-center justify-center gap-1">
+                            <Button
+                              variant="ghost"
+                              size="icon"
+                              class="h-7 w-7"
+                              title="编辑此关联"
+                              @click="$emit('editProvider', provider)"
+                            >
+                              <Edit class="w-3.5 h-3.5" />
+                            </Button>
+                            <Button
+                              variant="ghost"
+                              size="icon"
+                              class="h-7 w-7"
+                              :title="provider.is_active ? '停用此关联' : '启用此关联'"
+                              @click="$emit('toggleProviderStatus', provider)"
+                            >
+                              <Power class="w-3.5 h-3.5" />
+                            </Button>
+                            <Button
+                              variant="ghost"
+                              size="icon"
+                              class="h-7 w-7"
+                              title="删除此关联"
+                              @click="$emit('deleteProvider', provider)"
+                            >
+                              <Trash2 class="w-3.5 h-3.5" />
+                            </Button>
+                          </div>
+                        </TableCell>
+                      </TableRow>
+                    </TableBody>
+                  </Table>
+
+                  <!-- 移动端卡片列表 -->
+                  <div class="sm:hidden divide-y divide-border/40">
+                    <div
                       v-for="provider in providers"
                       :key="provider.id"
-                      class="border-b border-border/40 hover:bg-muted/30 transition-colors"
+                      class="p-4 space-y-3"
                     >
-                      <TableCell class="py-3">
-                        <div class="flex items-center gap-2">
+                      <div class="flex items-start justify-between gap-3">
+                        <div class="flex items-center gap-2 min-w-0">
                           <span
                             class="w-2 h-2 rounded-full shrink-0"
                             :class="provider.is_active ? 'bg-green-500' : 'bg-gray-300'"
-                            :title="provider.is_active ? '活跃' : '停用'"
                           />
                           <span class="font-medium truncate">{{ provider.display_name }}</span>
                         </div>
-                      </TableCell>
-                      <TableCell class="py-3">
-                        <div class="flex gap-0.5">
-                          <Zap
-                            v-if="provider.supports_streaming"
-                            class="w-3.5 h-3.5 text-muted-foreground"
-                            title="流式输出"
-                          />
-                          <Eye
-                            v-if="provider.supports_vision"
-                            class="w-3.5 h-3.5 text-muted-foreground"
-                            title="视觉理解"
-                          />
-                          <Wrench
-                            v-if="provider.supports_function_calling"
-                            class="w-3.5 h-3.5 text-muted-foreground"
-                            title="工具调用"
-                          />
-                        </div>
-                      </TableCell>
-                      <TableCell class="py-3">
-                        <div class="text-xs font-mono space-y-0.5">
-                          <!-- Token 计费：输入/输出 -->
-                          <div v-if="(provider.input_price_per_1m || 0) > 0 || (provider.output_price_per_1m || 0) > 0">
-                            <span class="text-muted-foreground">输入/输出:</span>
-                            <span class="ml-1">${{ (provider.input_price_per_1m || 0).toFixed(1) }}/${{ (provider.output_price_per_1m || 0).toFixed(1) }}</span>
-                            <!-- 阶梯标记 -->
-                            <span
-                              v-if="(provider.tier_count || 1) > 1"
-                              class="ml-1 text-muted-foreground"
-                              title="阶梯计费"
-                            >[阶梯]</span>
-                          </div>
-                          <!-- 缓存价格 -->
-                          <div
-                            v-if="(provider.cache_creation_price_per_1m || 0) > 0 || (provider.cache_read_price_per_1m || 0) > 0"
-                            class="text-muted-foreground"
-                          >
-                            <span>缓存:</span>
-                            <span class="ml-1">${{ (provider.cache_creation_price_per_1m || 0).toFixed(2) }}/${{ (provider.cache_read_price_per_1m || 0).toFixed(2) }}</span>
-                          </div>
-                          <!-- 1h 缓存价格 -->
-                          <div
-                            v-if="(provider.cache_1h_creation_price_per_1m || 0) > 0"
-                            class="text-muted-foreground"
-                          >
-                            <span>1h 缓存:</span>
-                            <span class="ml-1">${{ (provider.cache_1h_creation_price_per_1m || 0).toFixed(2) }}</span>
-                          </div>
-                          <!-- 按次计费 -->
-                          <div v-if="(provider.price_per_request || 0) > 0">
-                            <span class="text-muted-foreground">按次:</span>
-                            <span class="ml-1">${{ (provider.price_per_request || 0).toFixed(3) }}/次</span>
-                          </div>
-                          <!-- 无定价 -->
-                          <span
-                            v-if="!(provider.input_price_per_1m || 0) && !(provider.output_price_per_1m || 0) && !(provider.price_per_request || 0)"
-                            class="text-muted-foreground"
-                          >-</span>
-                        </div>
-                      </TableCell>
-                      <TableCell class="py-3 text-center">
-                        <div class="flex items-center justify-center gap-1">
+                        <div class="flex items-center gap-1 shrink-0">
                           <Button
                             variant="ghost"
                             size="icon"
                             class="h-7 w-7"
-                            title="编辑此关联"
                             @click="$emit('editProvider', provider)"
                           >
                             <Edit class="w-3.5 h-3.5" />
@@ -575,7 +610,6 @@
                             variant="ghost"
                             size="icon"
                             class="h-7 w-7"
-                            :title="provider.is_active ? '停用此关联' : '启用此关联'"
                             @click="$emit('toggleProviderStatus', provider)"
                           >
                             <Power class="w-3.5 h-3.5" />
@@ -584,82 +618,35 @@
                             variant="ghost"
                             size="icon"
                             class="h-7 w-7"
-                            title="删除此关联"
                             @click="$emit('deleteProvider', provider)"
                           >
                             <Trash2 class="w-3.5 h-3.5" />
                           </Button>
                         </div>
-                      </TableCell>
-                    </TableRow>
-                  </TableBody>
-                  </Table>
-
-                  <!-- 移动端卡片列表 -->
-                  <div class="sm:hidden divide-y divide-border/40">
-                  <div
-                    v-for="provider in providers"
-                    :key="provider.id"
-                    class="p-4 space-y-3"
-                  >
-                    <div class="flex items-start justify-between gap-3">
-                      <div class="flex items-center gap-2 min-w-0">
-                        <span
-                          class="w-2 h-2 rounded-full shrink-0"
-                          :class="provider.is_active ? 'bg-green-500' : 'bg-gray-300'"
-                        />
-                        <span class="font-medium truncate">{{ provider.display_name }}</span>
                       </div>
-                      <div class="flex items-center gap-1 shrink-0">
-                        <Button
-                          variant="ghost"
-                          size="icon"
-                          class="h-7 w-7"
-                          @click="$emit('editProvider', provider)"
+                      <div class="flex items-center gap-3 text-xs">
+                        <div class="flex gap-1">
+                          <Zap
+                            v-if="provider.supports_streaming"
+                            class="w-3.5 h-3.5 text-muted-foreground"
+                          />
+                          <Eye
+                            v-if="provider.supports_vision"
+                            class="w-3.5 h-3.5 text-muted-foreground"
+                          />
+                          <Wrench
+                            v-if="provider.supports_function_calling"
+                            class="w-3.5 h-3.5 text-muted-foreground"
+                          />
+                        </div>
+                        <div
+                          v-if="(provider.input_price_per_1m || 0) > 0 || (provider.output_price_per_1m || 0) > 0"
+                          class="text-muted-foreground font-mono"
                         >
-                          <Edit class="w-3.5 h-3.5" />
-                        </Button>
-                        <Button
-                          variant="ghost"
-                          size="icon"
-                          class="h-7 w-7"
-                          @click="$emit('toggleProviderStatus', provider)"
-                        >
-                          <Power class="w-3.5 h-3.5" />
-                        </Button>
-                        <Button
-                          variant="ghost"
-                          size="icon"
-                          class="h-7 w-7"
-                          @click="$emit('deleteProvider', provider)"
-                        >
-                          <Trash2 class="w-3.5 h-3.5" />
-                        </Button>
+                          ${{ (provider.input_price_per_1m || 0).toFixed(1) }}/${{ (provider.output_price_per_1m || 0).toFixed(1) }}
+                        </div>
                       </div>
                     </div>
-                    <div class="flex items-center gap-3 text-xs">
-                      <div class="flex gap-1">
-                        <Zap
-                          v-if="provider.supports_streaming"
-                          class="w-3.5 h-3.5 text-muted-foreground"
-                        />
-                        <Eye
-                          v-if="provider.supports_vision"
-                          class="w-3.5 h-3.5 text-muted-foreground"
-                        />
-                        <Wrench
-                          v-if="provider.supports_function_calling"
-                          class="w-3.5 h-3.5 text-muted-foreground"
-                        />
-                      </div>
-                      <div
-                        v-if="(provider.input_price_per_1m || 0) > 0 || (provider.output_price_per_1m || 0) > 0"
-                        class="text-muted-foreground font-mono"
-                      >
-                        ${{ (provider.input_price_per_1m || 0).toFixed(1) }}/${{ (provider.output_price_per_1m || 0).toFixed(1) }}
-                      </div>
-                    </div>
-                  </div>
                   </div>
                 </template>
 
@@ -684,236 +671,6 @@
                 </div>
               </Card>
             </div>
-
-            <!-- Tab 3: 别名 -->
-            <div v-show="detailTab === 'aliases'">
-              <Card class="overflow-hidden">
-                <!-- 标题栏 -->
-                <div class="px-4 py-3 border-b border-border/60">
-                  <div class="flex items-center justify-between gap-4">
-                    <div>
-                      <h4 class="text-sm font-semibold">
-                        别名与映射
-                      </h4>
-                    </div>
-                    <div class="flex items-center gap-2">
-                      <Button
-                        variant="ghost"
-                        size="icon"
-                        class="h-8 w-8"
-                        title="添加别名/映射"
-                        @click="$emit('addAlias')"
-                      >
-                        <Plus class="w-3.5 h-3.5" />
-                      </Button>
-                      <Button
-                        variant="ghost"
-                        size="icon"
-                        class="h-8 w-8"
-                        title="刷新"
-                        @click="$emit('refreshAliases')"
-                      >
-                        <RefreshCw
-                          class="w-3.5 h-3.5"
-                          :class="loadingAliases ? 'animate-spin' : ''"
-                        />
-                      </Button>
-                    </div>
-                  </div>
-                </div>
-
-                <!-- 表格内容 -->
-                <div
-                  v-if="loadingAliases"
-                  class="flex items-center justify-center py-12"
-                >
-                  <Loader2 class="w-6 h-6 animate-spin text-primary" />
-                </div>
-
-                <template v-else-if="aliases.length > 0">
-                  <!-- 桌面端表格 -->
-                  <Table class="hidden sm:table">
-                  <TableHeader>
-                    <TableRow class="border-b border-border/60 hover:bg-transparent">
-                      <TableHead class="h-10 font-semibold">
-                        别名
-                      </TableHead>
-                      <TableHead class="w-[80px] h-10 font-semibold">
-                        类型
-                      </TableHead>
-                      <TableHead class="w-[100px] h-10 font-semibold">
-                        作用域
-                      </TableHead>
-                      <TableHead class="w-[100px] h-10 font-semibold text-center">
-                        操作
-                      </TableHead>
-                    </TableRow>
-                  </TableHeader>
-                  <TableBody>
-                    <TableRow
-                      v-for="alias in aliases"
-                      :key="alias.id"
-                      class="border-b border-border/40 hover:bg-muted/30 transition-colors"
-                    >
-                      <TableCell class="py-3">
-                        <div class="flex items-center gap-2">
-                          <span
-                            class="w-2 h-2 rounded-full shrink-0"
-                            :class="alias.is_active ? 'bg-green-500' : 'bg-gray-300'"
-                            :title="alias.is_active ? '活跃' : '停用'"
-                          />
-                          <code class="text-sm font-medium bg-muted px-1.5 py-0.5 rounded">{{ alias.alias }}</code>
-                        </div>
-                      </TableCell>
-                      <TableCell class="py-3">
-                        <Badge
-                          variant="secondary"
-                          class="text-xs"
-                        >
-                          {{ alias.mapping_type === 'mapping' ? '映射' : '别名' }}
-                        </Badge>
-                      </TableCell>
-                      <TableCell class="py-3">
-                        <Badge
-                          v-if="alias.provider_id"
-                          variant="outline"
-                          class="text-xs truncate max-w-[90px]"
-                          :title="alias.provider_name || 'Provider'"
-                        >
-                          {{ alias.provider_name || 'Provider' }}
-                        </Badge>
-                        <Badge
-                          v-else
-                          variant="default"
-                          class="text-xs"
-                        >
-                          全局
-                        </Badge>
-                      </TableCell>
-                      <TableCell class="py-3 text-center">
-                        <div class="flex items-center justify-center gap-0.5">
-                          <Button
-                            variant="ghost"
-                            size="icon"
-                            class="h-7 w-7"
-                            title="编辑"
-                            @click="$emit('editAlias', alias)"
-                          >
-                            <Edit class="w-3.5 h-3.5" />
-                          </Button>
-                          <Button
-                            variant="ghost"
-                            size="icon"
-                            class="h-7 w-7"
-                            :title="alias.is_active ? '停用' : '启用'"
-                            @click="$emit('toggleAliasStatus', alias)"
-                          >
-                            <Power class="w-3.5 h-3.5" />
-                          </Button>
-                          <Button
-                            variant="ghost"
-                            size="icon"
-                            class="h-7 w-7"
-                            title="删除"
-                            @click="$emit('deleteAlias', alias)"
-                          >
-                            <Trash2 class="w-3.5 h-3.5" />
-                          </Button>
-                        </div>
-                      </TableCell>
-                    </TableRow>
-                  </TableBody>
-                  </Table>
-
-                  <!-- 移动端卡片列表 -->
-                  <div class="sm:hidden divide-y divide-border/40">
-                  <div
-                    v-for="alias in aliases"
-                    :key="alias.id"
-                    class="p-4 space-y-2"
-                  >
-                    <div class="flex items-start justify-between gap-3">
-                      <div class="flex items-center gap-2 min-w-0 flex-1">
-                        <span
-                          class="w-2 h-2 rounded-full shrink-0"
-                          :class="alias.is_active ? 'bg-green-500' : 'bg-gray-300'"
-                        />
-                        <code class="text-sm font-medium bg-muted px-1.5 py-0.5 rounded truncate">{{ alias.alias }}</code>
-                      </div>
-                      <div class="flex items-center gap-1 shrink-0">
-                        <Button
-                          variant="ghost"
-                          size="icon"
-                          class="h-7 w-7"
-                          @click="$emit('editAlias', alias)"
-                        >
-                          <Edit class="w-3.5 h-3.5" />
-                        </Button>
-                        <Button
-                          variant="ghost"
-                          size="icon"
-                          class="h-7 w-7"
-                          @click="$emit('toggleAliasStatus', alias)"
-                        >
-                          <Power class="w-3.5 h-3.5" />
-                        </Button>
-                        <Button
-                          variant="ghost"
-                          size="icon"
-                          class="h-7 w-7"
-                          @click="$emit('deleteAlias', alias)"
-                        >
-                          <Trash2 class="w-3.5 h-3.5" />
-                        </Button>
-                      </div>
-                    </div>
-                    <div class="flex items-center gap-2">
-                      <Badge
-                        variant="secondary"
-                        class="text-xs"
-                      >
-                        {{ alias.mapping_type === 'mapping' ? '映射' : '别名' }}
-                      </Badge>
-                      <Badge
-                        v-if="alias.provider_id"
-                        variant="outline"
-                        class="text-xs truncate max-w-[120px]"
-                      >
-                        {{ alias.provider_name || 'Provider' }}
-                      </Badge>
-                      <Badge
-                        v-else
-                        variant="default"
-                        class="text-xs"
-                      >
-                        全局
-                      </Badge>
-                    </div>
-                  </div>
-                  </div>
-                </template>
-
-                <div
-                  v-else
-                  class="text-center py-12"
-                >
-                  <!-- 空状态 -->
-                  <Tag class="w-12 h-12 mx-auto text-muted-foreground/30 mb-3" />
-                  <p class="text-sm text-muted-foreground">
-                    暂无别名或映射
-                  </p>
-                  <Button
-                    size="sm"
-                    variant="outline"
-                    class="mt-4"
-                    @click="$emit('addAlias')"
-                  >
-                    <Plus class="w-4 h-4 mr-1" />
-                    添加别名/映射
-                  </Button>
-                </div>
-              </Card>
-            </div>
           </div>
         </Card>
       </div>
@@ -931,7 +688,6 @@ import {
   Zap,
   Image,
   Building2,
-  Tag,
   Plus,
   Edit,
   Trash2,
@@ -939,8 +695,10 @@ import {
   Loader2,
   RefreshCw,
   Copy,
-  Layers
+  Layers,
+  BarChart3
 } from 'lucide-vue-next'
+import { useEscapeKey } from '@/composables/useEscapeKey'
 import { useToast } from '@/composables/useToast'
 import Card from '@/components/ui/card.vue'
 import Badge from '@/components/ui/badge.vue'
@@ -955,13 +713,11 @@ import TableCell from '@/components/ui/table-cell.vue'
 
 // 使用外部类型定义
 import type { GlobalModelResponse } from '@/api/global-models'
-import type { ModelAlias } from '@/api/endpoints/aliases'
 import type { TieredPricingConfig, PricingTier } from '@/api/endpoints/types'
 import type { CapabilityDefinition } from '@/api/endpoints'
 
 const props = withDefaults(defineProps<Props>(), {
   loadingProviders: false,
-  loadingAliases: false,
   hasBlockingDialogOpen: false,
 })
 const emit = defineEmits<{
@@ -973,11 +729,6 @@ const emit = defineEmits<{
   'deleteProvider': [provider: any]
   'toggleProviderStatus': [provider: any]
   'refreshProviders': []
-  'addAlias': []
-  'editAlias': [alias: ModelAlias]
-  'toggleAliasStatus': [alias: ModelAlias]
-  'deleteAlias': [alias: ModelAlias]
-  'refreshAliases': []
 }>()
 const { success: showSuccess, error: showError } = useToast()
 
@@ -985,9 +736,7 @@ interface Props {
   model: GlobalModelResponse | null
   open: boolean
   providers: any[]
-  aliases: ModelAlias[]
   loadingProviders?: boolean
-  loadingAliases?: boolean
   hasBlockingDialogOpen?: boolean
   capabilities?: CapabilityDefinition[]
 }
@@ -1085,6 +834,16 @@ watch(() => props.open, (newOpen) => {
     detailTab.value = 'basic'
   }
 })
+
+// 添加 ESC 键监听
+useEscapeKey(() => {
+  if (props.open) {
+    handleClose()
+  }
+}, {
+  disableOnInput: true,
+  once: false
+})
 </script>
 
 <style scoped>

frontend/src/features/models/components/index.ts

@@ -1,4 +1,3 @@
 export { default as GlobalModelFormDialog } from './GlobalModelFormDialog.vue'
-export { default as AliasDialog } from './AliasDialog.vue'
 export { default as ModelDetailDrawer } from './ModelDetailDrawer.vue'
 export { default as TieredPricingEditor } from './TieredPricingEditor.vue'

frontend/src/features/providers/components/BatchAssignModelsDialog.vue

@@ -31,29 +31,46 @@
 
         <!-- 左右对比布局 -->
         <div class="flex gap-2 items-stretch">
-          <!-- 左侧：可添加的模型 -->
+          <!-- 左侧：可添加的模型（分组折叠） -->
           <div class="flex-1 space-y-2">
-            <div class="flex items-center justify-between">
-              <div class="flex items-center gap-2">
-                <p class="text-sm font-medium">
-                  可添加
-                </p>
-                <Button
-                  v-if="availableModels.length > 0"
-                  variant="ghost"
-                  size="sm"
-                  class="h-6 px-2 text-xs"
-                  @click="toggleSelectAllLeft"
-                >
-                  {{ isAllLeftSelected ? '取消全选' : '全选' }}
-                </Button>
+            <div class="flex items-center justify-between gap-2">
+              <p class="text-sm font-medium shrink-0">
+                可添加
+              </p>
+              <div class="flex-1 relative">
+                <Search class="absolute left-2 top-1/2 -translate-y-1/2 w-3.5 h-3.5 text-muted-foreground" />
+                <Input
+                  v-model="searchQuery"
+                  placeholder="搜索模型..."
+                  class="pl-7 h-7 text-xs"
+                />
               </div>
-              <Badge
-                variant="secondary"
-                class="text-xs"
+              <button
+                v-if="upstreamModelsLoaded"
+                type="button"
+                class="p-1.5 hover:bg-muted rounded-md transition-colors shrink-0"
+                title="刷新上游模型"
+                :disabled="fetchingUpstreamModels"
+                @click="fetchUpstreamModels(true)"
               >
-                {{ availableModels.length }} 个
-              </Badge>
+                <RefreshCw
+                  class="w-3.5 h-3.5"
+                  :class="{ 'animate-spin': fetchingUpstreamModels }"
+                />
+              </button>
+              <button
+                v-else-if="!fetchingUpstreamModels"
+                type="button"
+                class="p-1.5 hover:bg-muted rounded-md transition-colors shrink-0"
+                title="从提供商获取模型"
+                @click="fetchUpstreamModels"
+              >
+                <Zap class="w-3.5 h-3.5" />
+              </button>
+              <Loader2
+                v-else
+                class="w-3.5 h-3.5 animate-spin text-muted-foreground shrink-0"
+              />
             </div>
             <div class="border rounded-lg h-80 overflow-y-auto">
               <div
@@ -63,7 +80,7 @@
                 <Loader2 class="w-6 h-6 animate-spin text-primary" />
               </div>
               <div
-                v-else-if="availableModels.length === 0"
+                v-else-if="totalAvailableCount === 0 && !upstreamModelsLoaded"
                 class="flex flex-col items-center justify-center h-full text-muted-foreground"
               >
                 <Layers class="w-10 h-10 mb-2 opacity-30" />
@@ -73,37 +90,142 @@
               </div>
               <div
                 v-else
-                class="p-2 space-y-1"
+                class="p-2 space-y-2"
               >
+                <!-- 全局模型折叠组 -->
                 <div
-                  v-for="model in availableModels"
-                  :key="model.id"
-                  class="flex items-center gap-2 p-2 rounded-lg border transition-colors"
-                  :class="selectedLeftIds.includes(model.id)
-                    ? 'border-primary bg-primary/10'
-                    : 'hover:bg-muted/50 cursor-pointer'"
-                  @click="toggleLeftSelection(model.id)"
+                  v-if="availableGlobalModels.length > 0 || !upstreamModelsLoaded"
+                  class="border rounded-lg overflow-hidden"
                 >
-                  <Checkbox
-                    :checked="selectedLeftIds.includes(model.id)"
-                    @update:checked="toggleLeftSelection(model.id)"
-                    @click.stop
-                  />
-                  <div class="flex-1 min-w-0">
-                    <p class="font-medium text-sm truncate">
-                      {{ model.display_name }}
-                    </p>
-                    <p class="text-xs text-muted-foreground truncate font-mono">
-                      {{ model.name }}
-                    </p>
+                  <div class="flex items-center gap-2 px-3 py-2 bg-muted/30">
+                    <button
+                      type="button"
+                      class="flex items-center gap-2 flex-1 hover:bg-muted/50 -mx-1 px-1 rounded transition-colors"
+                      @click="toggleGroupCollapse('global')"
+                    >
+                      <ChevronDown
+                        class="w-4 h-4 transition-transform shrink-0"
+                        :class="collapsedGroups.has('global') ? '-rotate-90' : ''"
+                      />
+                      <span class="text-xs font-medium">
+                        全局模型
+                      </span>
+                      <span class="text-xs text-muted-foreground">
+                        ({{ availableGlobalModels.length }})
+                      </span>
+                    </button>
+                    <button
+                      v-if="availableGlobalModels.length > 0"
+                      type="button"
+                      class="text-xs text-primary hover:underline shrink-0"
+                      @click.stop="selectAllGlobalModels"
+                    >
+                      {{ isAllGlobalModelsSelected ? '取消' : '全选' }}
+                    </button>
                   </div>
-                  <Badge
-                    :variant="model.is_active ? 'outline' : 'secondary'"
-                    :class="model.is_active ? 'text-green-600 border-green-500/60' : ''"
-                    class="text-xs shrink-0"
+                  <div
+                    v-show="!collapsedGroups.has('global')"
+                    class="p-2 space-y-1 border-t"
                   >
-                    {{ model.is_active ? '活跃' : '停用' }}
-                  </Badge>
+                    <div
+                      v-if="availableGlobalModels.length === 0"
+                      class="py-4 text-center text-xs text-muted-foreground"
+                    >
+                      所有全局模型均已关联
+                    </div>
+                    <div
+                      v-for="model in availableGlobalModels"
+                      v-else
+                      :key="model.id"
+                      class="flex items-center gap-2 p-2 rounded-lg border transition-colors cursor-pointer"
+                      :class="selectedGlobalModelIds.includes(model.id)
+                        ? 'border-primary bg-primary/10'
+                        : 'hover:bg-muted/50'"
+                      @click="toggleGlobalModelSelection(model.id)"
+                    >
+                      <Checkbox
+                        :checked="selectedGlobalModelIds.includes(model.id)"
+                        @update:checked="toggleGlobalModelSelection(model.id)"
+                        @click.stop
+                      />
+                      <div class="flex-1 min-w-0">
+                        <p class="font-medium text-sm truncate">
+                          {{ model.display_name }}
+                        </p>
+                        <p class="text-xs text-muted-foreground truncate font-mono">
+                          {{ model.name }}
+                        </p>
+                      </div>
+                      <Badge
+                        :variant="model.is_active ? 'outline' : 'secondary'"
+                        :class="model.is_active ? 'text-green-600 border-green-500/60' : ''"
+                        class="text-xs shrink-0"
+                      >
+                        {{ model.is_active ? '活跃' : '停用' }}
+                      </Badge>
+                    </div>
+                  </div>
+                </div>
+
+                <!-- 从提供商获取的模型折叠组 -->
+                <div
+                  v-for="group in upstreamModelGroups"
+                  :key="group.api_format"
+                  class="border rounded-lg overflow-hidden"
+                >
+                  <div class="flex items-center gap-2 px-3 py-2 bg-muted/30">
+                    <button
+                      type="button"
+                      class="flex items-center gap-2 flex-1 hover:bg-muted/50 -mx-1 px-1 rounded transition-colors"
+                      @click="toggleGroupCollapse(group.api_format)"
+                    >
+                      <ChevronDown
+                        class="w-4 h-4 transition-transform shrink-0"
+                        :class="collapsedGroups.has(group.api_format) ? '-rotate-90' : ''"
+                      />
+                      <span class="text-xs font-medium">
+                        {{ API_FORMAT_LABELS[group.api_format] || group.api_format }}
+                      </span>
+                      <span class="text-xs text-muted-foreground">
+                        ({{ group.models.length }})
+                      </span>
+                    </button>
+                    <button
+                      type="button"
+                      class="text-xs text-primary hover:underline shrink-0"
+                      @click.stop="selectAllUpstreamModels(group.api_format)"
+                    >
+                      {{ isUpstreamGroupAllSelected(group.api_format) ? '取消' : '全选' }}
+                    </button>
+                  </div>
+                  <div
+                    v-show="!collapsedGroups.has(group.api_format)"
+                    class="p-2 space-y-1 border-t"
+                  >
+                    <div
+                      v-for="model in group.models"
+                      :key="model.id"
+                      class="flex items-center gap-2 p-2 rounded-lg border transition-colors cursor-pointer"
+                      :class="selectedUpstreamModelIds.includes(model.id)
+                        ? 'border-primary bg-primary/10'
+                        : 'hover:bg-muted/50'"
+                      @click="toggleUpstreamModelSelection(model.id)"
+                    >
+                      <Checkbox
+                        :checked="selectedUpstreamModelIds.includes(model.id)"
+                        @update:checked="toggleUpstreamModelSelection(model.id)"
+                        @click.stop
+                      />
+                      <div class="flex-1 min-w-0">
+                        <p class="font-medium text-sm truncate">
+                          {{ model.id }}
+                        </p>
+                        <p class="text-xs text-muted-foreground truncate font-mono">
+                          {{ model.owned_by || model.id }}
+                        </p>
+                      </div>
+                    </div>
+                  </div>
                 </div>
               </div>
             </div>
@@ -115,8 +237,8 @@
               variant="outline"
               size="sm"
               class="w-9 h-8"
-              :class="selectedLeftIds.length > 0 && !submittingAdd ? 'border-primary' : ''"
-              :disabled="selectedLeftIds.length === 0 || submittingAdd"
+              :class="totalSelectedCount > 0 && !submittingAdd ? 'border-primary' : ''"
+              :disabled="totalSelectedCount === 0 || submittingAdd"
               title="添加选中"
               @click="batchAddSelected"
             >
@@ -127,7 +249,7 @@
               <ChevronRight
                 v-else
                 class="w-6 h-6 stroke-[3]"
-                :class="selectedLeftIds.length > 0 && !submittingAdd ? 'text-primary' : ''"
+                :class="totalSelectedCount > 0 && !submittingAdd ? 'text-primary' : ''"
               />
             </Button>
             <Button
@@ -154,26 +276,18 @@
           <!-- 右侧：已添加的模型 -->
           <div class="flex-1 space-y-2">
             <div class="flex items-center justify-between">
-              <div class="flex items-center gap-2">
-                <p class="text-sm font-medium">
-                  已添加
-                </p>
-                <Button
-                  v-if="existingModels.length > 0"
-                  variant="ghost"
-                  size="sm"
-                  class="h-6 px-2 text-xs"
-                  @click="toggleSelectAllRight"
-                >
-                  {{ isAllRightSelected ? '取消全选' : '全选' }}
-                </Button>
-              </div>
-              <Badge
-                variant="secondary"
-                class="text-xs"
+              <p class="text-sm font-medium">
+                已添加
+              </p>
+              <Button
+                v-if="existingModels.length > 0"
+                variant="ghost"
+                size="sm"
+                class="h-6 px-2 text-xs"
+                @click="toggleSelectAllRight"
               >
-                {{ existingModels.length }} 个
-              </Badge>
+                {{ isAllRightSelected ? '取消' : '全选' }}
+              </Button>
             </div>
             <div class="border rounded-lg h-80 overflow-y-auto">
               <div
@@ -238,11 +352,12 @@
 
 <script setup lang="ts">
 import { ref, computed, watch } from 'vue'
-import { Layers, Loader2, ChevronRight, ChevronLeft } from 'lucide-vue-next'
+import { Layers, Loader2, ChevronRight, ChevronLeft, ChevronDown, Zap, RefreshCw, Search } from 'lucide-vue-next'
 import Dialog from '@/components/ui/dialog/Dialog.vue'
 import Button from '@/components/ui/button.vue'
 import Badge from '@/components/ui/badge.vue'
 import Checkbox from '@/components/ui/checkbox.vue'
+import Input from '@/components/ui/input.vue'
 import { useToast } from '@/composables/useToast'
 import { parseApiError } from '@/utils/errorParser'
 import {
@@ -253,8 +368,13 @@ import {
   getProviderModels,
   batchAssignModelsToProvider,
   deleteModel,
+  importModelsFromUpstream,
+  API_FORMAT_LABELS,
   type Model
 } from '@/api/endpoints'
+import { useUpstreamModelsCache, type UpstreamModel } from '../composables/useUpstreamModelsCache'
+
+const { fetchModels: fetchCachedModels, clearCache, getCachedModels } = useUpstreamModelsCache()
 
 const props = defineProps<{
   open: boolean
@@ -274,17 +394,27 @@ const { error: showError, success } = useToast()
 const loadingGlobalModels = ref(false)
 const submittingAdd = ref(false)
 const submittingRemove = ref(false)
+const fetchingUpstreamModels = ref(false)
+const upstreamModelsLoaded = ref(false)
 
 // 数据
 const allGlobalModels = ref<GlobalModelResponse[]>([])
 const existingModels = ref<Model[]>([])
+const upstreamModels = ref<UpstreamModel[]>([])
 
 // 选择状态
-const selectedLeftIds = ref<string[]>([])
+const selectedGlobalModelIds = ref<string[]>([])
+const selectedUpstreamModelIds = ref<string[]>([])
 const selectedRightIds = ref<string[]>([])
 
-// 计算可添加的模型（排除已关联的）
-const availableModels = computed(() => {
+// 折叠状态
+const collapsedGroups = ref<Set<string>>(new Set())
+
+// 搜索状态
+const searchQuery = ref('')
+
+// 计算可添加的全局模型（排除已关联的）
+const availableGlobalModelsBase = computed(() => {
   const existingGlobalModelIds = new Set(
     existingModels.value
       .filter(m => m.global_model_id)
@@ -293,31 +423,123 @@ const availableModels = computed(() => {
   return allGlobalModels.value.filter(m => !existingGlobalModelIds.has(m.id))
 })
 
-// 全选状态
-const isAllLeftSelected = computed(() =>
-  availableModels.value.length > 0 &&
-  selectedLeftIds.value.length === availableModels.value.length
-)
+// 搜索过滤后的全局模型
+const availableGlobalModels = computed(() => {
+  if (!searchQuery.value.trim()) return availableGlobalModelsBase.value
+  const query = searchQuery.value.toLowerCase()
+  return availableGlobalModelsBase.value.filter(m =>
+    m.name.toLowerCase().includes(query) ||
+    m.display_name.toLowerCase().includes(query)
+  )
+})
 
+// 计算可添加的上游模型（排除已关联的）
+const availableUpstreamModelsBase = computed(() => {
+  const existingModelNames = new Set(
+    existingModels.value.map(m => m.provider_model_name)
+  )
+  return upstreamModels.value.filter(m => !existingModelNames.has(m.id))
+})
+
+// 搜索过滤后的上游模型
+const availableUpstreamModels = computed(() => {
+  if (!searchQuery.value.trim()) return availableUpstreamModelsBase.value
+  const query = searchQuery.value.toLowerCase()
+  return availableUpstreamModelsBase.value.filter(m =>
+    m.id.toLowerCase().includes(query) ||
+    (m.owned_by && m.owned_by.toLowerCase().includes(query))
+  )
+})
+
+// 按 API 格式分组的上游模型
+const upstreamModelGroups = computed(() => {
+  const groups: Record<string, UpstreamModel[]> = {}
+
+  for (const model of availableUpstreamModels.value) {
+    const format = model.api_format || 'unknown'
+    if (!groups[format]) {
+      groups[format] = []
+    }
+    groups[format].push(model)
+  }
+
+  // 按 API_FORMAT_LABELS 的顺序排序
+  const order = Object.keys(API_FORMAT_LABELS)
+  return Object.entries(groups)
+    .map(([api_format, models]) => ({ api_format, models }))
+    .sort((a, b) => {
+      const aIndex = order.indexOf(a.api_format)
+      const bIndex = order.indexOf(b.api_format)
+      if (aIndex === -1 && bIndex === -1) return a.api_format.localeCompare(b.api_format)
+      if (aIndex === -1) return 1
+      if (bIndex === -1) return -1
+      return aIndex - bIndex
+    })
+})
+
+// 总可添加数量
+const totalAvailableCount = computed(() => {
+  return availableGlobalModels.value.length + availableUpstreamModels.value.length
+})
+
+// 总选中数量
+const totalSelectedCount = computed(() => {
+  return selectedGlobalModelIds.value.length + selectedUpstreamModelIds.value.length
+})
+
+// 全选状态
 const isAllRightSelected = computed(() =>
   existingModels.value.length > 0 &&
   selectedRightIds.value.length === existingModels.value.length
 )
 
+// 全局模型是否全选
+const isAllGlobalModelsSelected = computed(() => {
+  if (availableGlobalModels.value.length === 0) return false
+  return availableGlobalModels.value.every(m => selectedGlobalModelIds.value.includes(m.id))
+})
+
+// 检查某个上游组是否全选
+function isUpstreamGroupAllSelected(apiFormat: string): boolean {
+  const group = upstreamModelGroups.value.find(g => g.api_format === apiFormat)
+  if (!group || group.models.length === 0) return false
+  return group.models.every(m => selectedUpstreamModelIds.value.includes(m.id))
+}
+
 // 监听打开状态
 watch(() => props.open, async (isOpen) => {
   if (isOpen && props.providerId) {
     await loadData()
   } else {
     // 重置状态
-    selectedLeftIds.value = []
+    selectedGlobalModelIds.value = []
+    selectedUpstreamModelIds.value = []
     selectedRightIds.value = []
+    upstreamModels.value = []
+    upstreamModelsLoaded.value = false
+    collapsedGroups.value = new Set()
+    searchQuery.value = ''
   }
 })
 
 // 加载数据
 async function loadData() {
   await Promise.all([loadGlobalModels(), loadExistingModels()])
+  // 默认折叠全局模型组
+  collapsedGroups.value = new Set(['global'])
+
+  // 检查缓存，如果有缓存数据则直接使用
+  const cachedModels = getCachedModels(props.providerId)
+  if (cachedModels) {
+    upstreamModels.value = cachedModels
+    upstreamModelsLoaded.value = true
+    // 折叠所有上游模型组
+    for (const model of cachedModels) {
+      if (model.api_format) {
+        collapsedGroups.value.add(model.api_format)
+      }
+    }
+  }
 }
 
 // 加载全局模型列表
@@ -342,13 +564,91 @@ async function loadExistingModels() {
   }
 }
 
-// 切换左侧选择
-function toggleLeftSelection(id: string) {
-  const index = selectedLeftIds.value.indexOf(id)
-  if (index === -1) {
-    selectedLeftIds.value.push(id)
+// 从提供商获取模型
+async function fetchUpstreamModels(forceRefresh = false) {
+  if (forceRefresh) {
+    clearCache(props.providerId)
+  }
+
+  try {
+    fetchingUpstreamModels.value = true
+    const result = await fetchCachedModels(props.providerId, forceRefresh)
+    if (result) {
+      if (result.error) {
+        showError(result.error, '错误')
+      } else {
+        upstreamModels.value = result.models
+        upstreamModelsLoaded.value = true
+        // 折叠所有上游模型组
+        const allGroups = new Set(collapsedGroups.value)
+        for (const model of result.models) {
+          if (model.api_format) {
+            allGroups.add(model.api_format)
+          }
+        }
+        collapsedGroups.value = allGroups
+      }
+    }
+  } finally {
+    fetchingUpstreamModels.value = false
+  }
+}
+
+// 切换折叠状态
+function toggleGroupCollapse(group: string) {
+  if (collapsedGroups.value.has(group)) {
+    collapsedGroups.value.delete(group)
   } else {
-    selectedLeftIds.value.splice(index, 1)
+    collapsedGroups.value.add(group)
+  }
+  // 触发响应式更新
+  collapsedGroups.value = new Set(collapsedGroups.value)
+}
+
+// 切换全局模型选择
+function toggleGlobalModelSelection(id: string) {
+  const index = selectedGlobalModelIds.value.indexOf(id)
+  if (index === -1) {
+    selectedGlobalModelIds.value.push(id)
+  } else {
+    selectedGlobalModelIds.value.splice(index, 1)
+  }
+}
+
+// 切换上游模型选择
+function toggleUpstreamModelSelection(id: string) {
+  const index = selectedUpstreamModelIds.value.indexOf(id)
+  if (index === -1) {
+    selectedUpstreamModelIds.value.push(id)
+  } else {
+    selectedUpstreamModelIds.value.splice(index, 1)
+  }
+}
+
+// 全选全局模型
+function selectAllGlobalModels() {
+  const allIds = availableGlobalModels.value.map(m => m.id)
+  const allSelected = allIds.every(id => selectedGlobalModelIds.value.includes(id))
+  if (allSelected) {
+    selectedGlobalModelIds.value = selectedGlobalModelIds.value.filter(id => !allIds.includes(id))
+  } else {
+    const newIds = allIds.filter(id => !selectedGlobalModelIds.value.includes(id))
+    selectedGlobalModelIds.value.push(...newIds)
+  }
+}
+
+// 全选某个 API 格式的上游模型
+function selectAllUpstreamModels(apiFormat: string) {
+  const group = upstreamModelGroups.value.find(g => g.api_format === apiFormat)
+  if (!group) return
+
+  const allIds = group.models.map(m => m.id)
+  const allSelected = allIds.every(id => selectedUpstreamModelIds.value.includes(id))
+  if (allSelected) {
+    selectedUpstreamModelIds.value = selectedUpstreamModelIds.value.filter(id => !allIds.includes(id))
+  } else {
+    const newIds = allIds.filter(id => !selectedUpstreamModelIds.value.includes(id))
+    selectedUpstreamModelIds.value.push(...newIds)
   }
 }
 
@@ -362,15 +662,6 @@ function toggleRightSelection(id: string) {
   }
 }
 
-// 全选/取消全选左侧
-function toggleSelectAllLeft() {
-  if (isAllLeftSelected.value) {
-    selectedLeftIds.value = []
-  } else {
-    selectedLeftIds.value = availableModels.value.map(m => m.id)
-  }
-}
-
 // 全选/取消全选右侧
 function toggleSelectAllRight() {
   if (isAllRightSelected.value) {
@@ -382,22 +673,41 @@ function toggleSelectAllRight() {
 
 // 批量添加选中的模型
 async function batchAddSelected() {
-  if (selectedLeftIds.value.length === 0) return
+  if (totalSelectedCount.value === 0) return
 
   try {
     submittingAdd.value = true
-    const result = await batchAssignModelsToProvider(props.providerId, selectedLeftIds.value)
+    let totalSuccess = 0
+    const allErrors: string[] = []
 
-    if (result.success.length > 0) {
-      success(`成功添加 ${result.success.length} 个模型`)
+    // 处理全局模型
+    if (selectedGlobalModelIds.value.length > 0) {
+      const result = await batchAssignModelsToProvider(props.providerId, selectedGlobalModelIds.value)
+      totalSuccess += result.success.length
+      if (result.errors.length > 0) {
+        allErrors.push(...result.errors.map(e => e.error))
+      }
     }
 
-    if (result.errors.length > 0) {
-      const errorMessages = result.errors.map(e => e.error).join(', ')
-      showError(`部分模型添加失败: ${errorMessages}`, '警告')
+    // 处理上游模型（调用 import-from-upstream API）
+    if (selectedUpstreamModelIds.value.length > 0) {
+      const result = await importModelsFromUpstream(props.providerId, selectedUpstreamModelIds.value)
+      totalSuccess += result.success.length
+      if (result.errors.length > 0) {
+        allErrors.push(...result.errors.map(e => e.error))
+      }
     }
 
-    selectedLeftIds.value = []
+    if (totalSuccess > 0) {
+      success(`成功添加 ${totalSuccess} 个模型`)
+    }
+
+    if (allErrors.length > 0) {
+      showError(`部分模型添加失败: ${allErrors.slice(0, 3).join(', ')}${allErrors.length > 3 ? '...' : ''}`, '警告')
+    }
+
+    selectedGlobalModelIds.value = []
+    selectedUpstreamModelIds.value = []
     await loadExistingModels()
     emit('changed')
   } catch (err: any) {

frontend/src/features/providers/components/EndpointFormDialog.vue

@@ -9,7 +9,7 @@
   >
     <form
       class="space-y-6"
-      @submit.prevent="handleSubmit"
+      @submit.prevent="handleSubmit()"
     >
       <!-- API 配置 -->
       <div class="space-y-4">
@@ -132,6 +132,79 @@
           </div>
         </div>
       </div>
+
+      <!-- 代理配置 -->
+      <div class="space-y-4">
+        <div class="flex items-center justify-between">
+          <h3 class="text-sm font-medium">
+            代理配置
+          </h3>
+          <div class="flex items-center gap-2">
+            <Switch v-model="proxyEnabled" />
+            <span class="text-sm text-muted-foreground">启用代理</span>
+          </div>
+        </div>
+
+        <div
+          v-if="proxyEnabled"
+          class="space-y-4 rounded-lg border p-4"
+        >
+          <div class="space-y-2">
+            <Label for="proxy_url">代理 URL *</Label>
+            <Input
+              id="proxy_url"
+              v-model="form.proxy_url"
+              placeholder="http://host:port 或 socks5://host:port"
+              required
+              :class="proxyUrlError ? 'border-red-500' : ''"
+            />
+            <p
+              v-if="proxyUrlError"
+              class="text-xs text-red-500"
+            >
+              {{ proxyUrlError }}
+            </p>
+            <p
+              v-else
+              class="text-xs text-muted-foreground"
+            >
+              支持 HTTP、HTTPS、SOCKS5 代理
+            </p>
+          </div>
+
+          <div class="grid grid-cols-2 gap-4">
+            <div class="space-y-2">
+              <Label for="proxy_user">用户名（可选）</Label>
+              <Input
+                :id="`proxy_user_${formId}`"
+                :name="`proxy_user_${formId}`"
+                v-model="form.proxy_username"
+                placeholder="代理认证用户名"
+                autocomplete="off"
+                data-form-type="other"
+                data-lpignore="true"
+                data-1p-ignore="true"
+              />
+            </div>
+
+            <div class="space-y-2">
+              <Label :for="`proxy_pass_${formId}`">密码（可选）</Label>
+              <Input
+                :id="`proxy_pass_${formId}`"
+                :name="`proxy_pass_${formId}`"
+                v-model="form.proxy_password"
+                type="text"
+                :placeholder="passwordPlaceholder"
+                autocomplete="off"
+                data-form-type="other"
+                data-lpignore="true"
+                data-1p-ignore="true"
+                :style="{ '-webkit-text-security': 'disc', 'text-security': 'disc' }"
+              />
+            </div>
+          </div>
+        </div>
+      </div>
     </form>
 
     <template #footer>
@@ -145,12 +218,24 @@
       </Button>
       <Button
         :disabled="loading || !form.base_url || (!isEditMode && !form.api_format)"
-        @click="handleSubmit"
+        @click="handleSubmit()"
       >
         {{ loading ? (isEditMode ? '保存中...' : '创建中...') : (isEditMode ? '保存修改' : '创建') }}
       </Button>
     </template>
   </Dialog>
+
+  <!-- 确认清空凭据对话框 -->
+  <AlertDialog
+    v-model="showClearCredentialsDialog"
+    title="清空代理凭据"
+    description="代理 URL 为空，但用户名和密码仍有值。是否清空这些凭据并继续保存？"
+    type="warning"
+    confirm-text="清空并保存"
+    cancel-text="返回编辑"
+    @confirm="confirmClearCredentials"
+    @cancel="showClearCredentialsDialog = false"
+  />
 </template>
 
 <script setup lang="ts">
@@ -165,7 +250,9 @@ import {
   SelectValue,
   SelectContent,
   SelectItem,
+  Switch,
 } from '@/components/ui'
+import AlertDialog from '@/components/common/AlertDialog.vue'
 import { Link, SquarePen } from 'lucide-vue-next'
 import { useToast } from '@/composables/useToast'
 import { useFormDialog } from '@/composables/useFormDialog'
@@ -194,6 +281,11 @@ const emit = defineEmits<{
 const { success, error: showError } = useToast()
 const loading = ref(false)
 const selectOpen = ref(false)
+const proxyEnabled = ref(false)
+const showClearCredentialsDialog = ref(false)  // 确认清空凭据对话框
+
+// 生成随机 ID 防止浏览器自动填充
+const formId = Math.random().toString(36).substring(2, 10)
 
 // 内部状态
 const internalOpen = computed(() => props.modelValue)
@@ -207,7 +299,11 @@ const form = ref({
   max_retries: 3,
   max_concurrent: undefined as number | undefined,
   rate_limit: undefined as number | undefined,
-  is_active: true
+  is_active: true,
+  // 代理配置
+  proxy_url: '',
+  proxy_username: '',
+  proxy_password: '',
 })
 
 // API 格式列表
@@ -237,6 +333,53 @@ const defaultPathPlaceholder = computed(() => {
   return `留空使用默认路径：${defaultPath.value}`
 })
 
+// 检查是否有已保存的密码（后端返回 *** 表示有密码）
+const hasExistingPassword = computed(() => {
+  if (!props.endpoint?.proxy) return false
+  const proxy = props.endpoint.proxy as { password?: string }
+  return proxy?.password === MASKED_PASSWORD
+})
+
+// 密码输入框的 placeholder
+const passwordPlaceholder = computed(() => {
+  if (hasExistingPassword.value) {
+    return '已保存密码，留空保持不变'
+  }
+  return '代理认证密码'
+})
+
+// 代理 URL 验证
+const proxyUrlError = computed(() => {
+  // 只有启用代理且填写了 URL 时才验证
+  if (!proxyEnabled.value || !form.value.proxy_url) {
+    return ''
+  }
+  const url = form.value.proxy_url.trim()
+
+  // 检查禁止的特殊字符
+  if (/[\n\r]/.test(url)) {
+    return '代理 URL 包含非法字符'
+  }
+
+  // 验证协议（不支持 SOCKS4）
+  if (!/^(http|https|socks5):\/\//i.test(url)) {
+    return '代理 URL 必须以 http://, https:// 或 socks5:// 开头'
+  }
+  try {
+    const parsed = new URL(url)
+    if (!parsed.host) {
+      return '代理 URL 必须包含有效的 host'
+    }
+    // 禁止 URL 中内嵌认证信息
+    if (parsed.username || parsed.password) {
+      return '请勿在 URL 中包含用户名和密码，请使用独立的认证字段'
+    }
+  } catch {
+    return '代理 URL 格式无效'
+  }
+  return ''
+})
+
 // 组件挂载时加载API格式
 onMounted(() => {
   loadApiFormats()
@@ -252,14 +395,23 @@ function resetForm() {
     max_retries: 3,
     max_concurrent: undefined,
     rate_limit: undefined,
-    is_active: true
+    is_active: true,
+    proxy_url: '',
+    proxy_username: '',
+    proxy_password: '',
   }
+  proxyEnabled.value = false
 }
 
+// 原始密码占位符（后端返回的脱敏标记）
+const MASKED_PASSWORD = '***'
+
 // 加载端点数据（编辑模式）
 function loadEndpointData() {
   if (!props.endpoint) return
 
+  const proxy = props.endpoint.proxy as { url?: string; username?: string; password?: string; enabled?: boolean } | null
+
   form.value = {
     api_format: props.endpoint.api_format,
     base_url: props.endpoint.base_url,
@@ -268,8 +420,15 @@ function loadEndpointData() {
     max_retries: props.endpoint.max_retries,
     max_concurrent: props.endpoint.max_concurrent || undefined,
     rate_limit: props.endpoint.rate_limit || undefined,
-    is_active: props.endpoint.is_active
+    is_active: props.endpoint.is_active,
+    proxy_url: proxy?.url || '',
+    proxy_username: proxy?.username || '',
+    // 如果密码是脱敏标记，显示为空（让用户知道有密码但看不到）
+    proxy_password: proxy?.password === MASKED_PASSWORD ? '' : (proxy?.password || ''),
   }
+
+  // 根据 enabled 字段或 url 存在判断是否启用代理
+  proxyEnabled.value = proxy?.enabled ?? !!proxy?.url
 }
 
 // 使用 useFormDialog 统一处理对话框逻辑
@@ -282,12 +441,47 @@ const { isEditMode, handleDialogUpdate, handleCancel } = useFormDialog({
   resetForm,
 })
 
+// 构建代理配置
+// - 有 URL 时始终保存配置，通过 enabled 字段控制是否启用
+// - 无 URL 时返回 null
+function buildProxyConfig(): { url: string; username?: string; password?: string; enabled: boolean } | null {
+  if (!form.value.proxy_url) {
+    // 没填 URL，无代理配置
+    return null
+  }
+  return {
+    url: form.value.proxy_url,
+    username: form.value.proxy_username || undefined,
+    password: form.value.proxy_password || undefined,
+    enabled: proxyEnabled.value,  // 开关状态决定是否启用
+  }
+}
+
 // 提交表单
-const handleSubmit = async () => {
+const handleSubmit = async (skipCredentialCheck = false) => {
   if (!props.provider && !props.endpoint) return
 
+  // 只在开关开启且填写了 URL 时验证
+  if (proxyEnabled.value && form.value.proxy_url && proxyUrlError.value) {
+    showError(proxyUrlError.value, '代理配置错误')
+    return
+  }
+
+  // 检查：开关开启但没有 URL，却有用户名或密码
+  const hasOrphanedCredentials = proxyEnabled.value
+    && !form.value.proxy_url
+    && (form.value.proxy_username || form.value.proxy_password)
+
+  if (hasOrphanedCredentials && !skipCredentialCheck) {
+    // 弹出确认对话框
+    showClearCredentialsDialog.value = true
+    return
+  }
+
   loading.value = true
   try {
+    const proxyConfig = buildProxyConfig()
+
     if (isEditMode.value && props.endpoint) {
       // 更新端点
       await updateEndpoint(props.endpoint.id, {
@@ -297,7 +491,8 @@ const handleSubmit = async () => {
         max_retries: form.value.max_retries,
         max_concurrent: form.value.max_concurrent,
         rate_limit: form.value.rate_limit,
-        is_active: form.value.is_active
+        is_active: form.value.is_active,
+        proxy: proxyConfig,
       })
 
       success('端点已更新', '保存成功')
@@ -313,7 +508,8 @@ const handleSubmit = async () => {
         max_retries: form.value.max_retries,
         max_concurrent: form.value.max_concurrent,
         rate_limit: form.value.rate_limit,
-        is_active: form.value.is_active
+        is_active: form.value.is_active,
+        proxy: proxyConfig,
       })
 
       success('端点创建成功', '成功')
@@ -329,4 +525,12 @@ const handleSubmit = async () => {
     loading.value = false
   }
 }
+
+// 确认清空凭据并继续保存
+const confirmClearCredentials = () => {
+  form.value.proxy_username = ''
+  form.value.proxy_password = ''
+  showClearCredentialsDialog.value = false
+  handleSubmit(true)  // 跳过凭据检查，直接提交
+}
 </script>

frontend/src/features/providers/components/ModelAliasDialog.vue

@@ -0,0 +1,337 @@
+<template>
+  <Dialog
+    :model-value="open"
+    title="管理模型名称映射"
+    description="配置 Provider 对此模型使用的名称变体，系统会按优先级顺序选择"
+    :icon="Tag"
+    size="lg"
+    @update:model-value="handleClose"
+  >
+    <div class="space-y-4">
+      <!-- 模型信息 -->
+      <div class="rounded-lg border bg-muted/30 p-3">
+        <p class="font-medium">
+          {{ model?.global_model_display_name || model?.provider_model_name }}
+        </p>
+        <p class="text-sm text-muted-foreground font-mono">
+          主名称: {{ model?.provider_model_name }}
+        </p>
+      </div>
+
+      <!-- 映射列表 -->
+      <div class="space-y-3">
+        <div class="flex items-center justify-between">
+          <Label class="text-sm font-medium">名称映射</Label>
+          <Button
+            type="button"
+            variant="outline"
+            size="sm"
+            @click="addAlias"
+          >
+            <Plus class="w-4 h-4 mr-1" />
+            添加
+          </Button>
+        </div>
+
+        <!-- 提示信息 -->
+        <div
+          v-if="aliases.length > 0"
+          class="flex items-center gap-2 px-3 py-2 text-xs text-muted-foreground bg-muted/30 rounded-md"
+        >
+          <Info class="w-3.5 h-3.5 shrink-0" />
+          <span>拖拽调整顺序，点击序号可编辑（相同数字为同级，负载均衡）</span>
+        </div>
+
+        <div
+          v-if="aliases.length > 0"
+          class="space-y-2"
+        >
+          <div
+            v-for="(alias, index) in aliases"
+            :key="index"
+            class="group flex items-center gap-3 px-3 py-2.5 rounded-lg border transition-all duration-200"
+            :class="[
+              draggedIndex === index
+                ? 'border-primary/50 bg-primary/5 shadow-md scale-[1.01]'
+                : dragOverIndex === index
+                  ? 'border-primary/30 bg-primary/5'
+                  : 'border-border/50 bg-background hover:border-border hover:bg-muted/30'
+            ]"
+            draggable="true"
+            @dragstart="handleDragStart(index, $event)"
+            @dragend="handleDragEnd"
+            @dragover.prevent="handleDragOver(index)"
+            @dragleave="handleDragLeave"
+            @drop="handleDrop(index)"
+          >
+            <!-- 拖拽手柄 -->
+            <div class="cursor-grab active:cursor-grabbing p-1 rounded hover:bg-muted text-muted-foreground/40 group-hover:text-muted-foreground transition-colors shrink-0">
+              <GripVertical class="w-4 h-4" />
+            </div>
+
+            <!-- 可编辑优先级 -->
+            <div class="shrink-0">
+              <input
+                v-if="editingPriorityIndex === index"
+                type="number"
+                min="1"
+                :value="alias.priority"
+                class="w-8 h-6 rounded-md bg-background border border-primary text-xs font-medium text-center focus:outline-none [appearance:textfield] [&::-webkit-outer-spin-button]:appearance-none [&::-webkit-inner-spin-button]:appearance-none"
+                autofocus
+                @blur="finishEditPriority(index, $event)"
+                @keydown.enter="($event.target as HTMLInputElement).blur()"
+                @keydown.escape="cancelEditPriority"
+              >
+              <div
+                v-else
+                class="w-6 h-6 rounded-md bg-muted/50 flex items-center justify-center text-xs font-medium text-muted-foreground cursor-pointer hover:bg-primary/10 hover:text-primary transition-colors"
+                title="点击编辑优先级，相同数字为同级（负载均衡）"
+                @click.stop="startEditPriority(index)"
+              >
+                {{ alias.priority }}
+              </div>
+            </div>
+
+            <!-- 映射输入框 -->
+            <Input
+              v-model="alias.name"
+              placeholder="映射名称，如 Claude-Sonnet-4.5"
+              class="flex-1"
+            />
+
+            <!-- 删除按钮 -->
+            <Button
+              type="button"
+              variant="ghost"
+              size="icon"
+              class="shrink-0 text-destructive hover:text-destructive h-8 w-8"
+              @click="removeAlias(index)"
+            >
+              <X class="w-4 h-4" />
+            </Button>
+          </div>
+        </div>
+
+        <div
+          v-else
+          class="text-center py-6 text-muted-foreground border rounded-lg border-dashed"
+        >
+          <Tag class="w-8 h-8 mx-auto mb-2 opacity-50" />
+          <p class="text-sm">
+            未配置映射
+          </p>
+          <p class="text-xs mt-1">
+            将只使用主模型名称
+          </p>
+        </div>
+      </div>
+    </div>
+
+    <template #footer>
+      <Button
+        variant="outline"
+        @click="handleClose(false)"
+      >
+        取消
+      </Button>
+      <Button
+        :disabled="submitting"
+        @click="handleSubmit"
+      >
+        <Loader2
+          v-if="submitting"
+          class="w-4 h-4 mr-2 animate-spin"
+        />
+        保存
+      </Button>
+    </template>
+  </Dialog>
+</template>
+
+<script setup lang="ts">
+import { ref, watch } from 'vue'
+import { Tag, Plus, X, Loader2, GripVertical, Info } from 'lucide-vue-next'
+import { Dialog, Button, Input, Label } from '@/components/ui'
+import { useToast } from '@/composables/useToast'
+import { updateModel } from '@/api/endpoints/models'
+import type { Model, ProviderModelAlias } from '@/api/endpoints'
+
+interface Props {
+  open: boolean
+  providerId: string
+  model: Model | null
+}
+
+const props = defineProps<Props>()
+
+const emit = defineEmits<{
+  'update:open': [value: boolean]
+  'saved': []
+}>()
+
+const { error: showError, success: showSuccess } = useToast()
+
+const submitting = ref(false)
+const aliases = ref<ProviderModelAlias[]>([])
+
+// 拖拽状态
+const draggedIndex = ref<number | null>(null)
+const dragOverIndex = ref<number | null>(null)
+
+// 优先级编辑状态
+const editingPriorityIndex = ref<number | null>(null)
+
+// 监听 open 变化
+watch(() => props.open, (newOpen) => {
+  if (newOpen && props.model) {
+    // 加载现有映射配置
+    if (props.model.provider_model_mappings && Array.isArray(props.model.provider_model_mappings)) {
+      aliases.value = JSON.parse(JSON.stringify(props.model.provider_model_mappings))
+    } else {
+      aliases.value = []
+    }
+    // 重置状态
+    editingPriorityIndex.value = null
+    draggedIndex.value = null
+    dragOverIndex.value = null
+  }
+})
+
+// 添加映射
+function addAlias() {
+  // 新映射优先级为当前最大优先级 + 1，或者默认为 1
+  const maxPriority = aliases.value.length > 0
+    ? Math.max(...aliases.value.map(a => a.priority))
+    : 0
+  aliases.value.push({ name: '', priority: maxPriority + 1 })
+}
+
+// 移除映射
+function removeAlias(index: number) {
+  aliases.value.splice(index, 1)
+}
+
+// ===== 拖拽排序 =====
+function handleDragStart(index: number, event: DragEvent) {
+  draggedIndex.value = index
+  if (event.dataTransfer) {
+    event.dataTransfer.effectAllowed = 'move'
+  }
+}
+
+function handleDragEnd() {
+  draggedIndex.value = null
+  dragOverIndex.value = null
+}
+
+function handleDragOver(index: number) {
+  if (draggedIndex.value !== null && draggedIndex.value !== index) {
+    dragOverIndex.value = index
+  }
+}
+
+function handleDragLeave() {
+  dragOverIndex.value = null
+}
+
+function handleDrop(targetIndex: number) {
+  const dragIndex = draggedIndex.value
+  if (dragIndex === null || dragIndex === targetIndex) {
+    dragOverIndex.value = null
+    return
+  }
+
+  const items = [...aliases.value]
+  const draggedItem = items[dragIndex]
+
+  // 记录每个映射的原始优先级（在修改前）
+  const originalPriorityMap = new Map<number, number>()
+  items.forEach((alias, idx) => {
+    originalPriorityMap.set(idx, alias.priority)
+  })
+
+  // 重排数组
+  items.splice(dragIndex, 1)
+  items.splice(targetIndex, 0, draggedItem)
+
+  // 按新顺序为每个组分配新的优先级
+  // 同组的映射保持相同的优先级（被拖动的映射单独成组）
+  const groupNewPriority = new Map<number, number>() // 原优先级 -> 新优先级
+  let currentPriority = 1
+
+  // 找到被拖动项在原数组中的索引对应的原始优先级
+  const draggedOriginalPriority = originalPriorityMap.get(dragIndex)!
+
+  items.forEach((alias, newIdx) => {
+    // 找到这个映射在原数组中的索引
+    const originalIdx = aliases.value.findIndex(a => a === alias)
+    const originalPriority = originalIdx >= 0 ? originalPriorityMap.get(originalIdx)! : alias.priority
+
+    if (alias === draggedItem) {
+      // 被拖动的映射是独立的新组，获得当前优先级
+      alias.priority = currentPriority
+      currentPriority++
+    } else {
+      if (groupNewPriority.has(originalPriority)) {
+        // 这个组已经分配过优先级，使用相同的值
+        alias.priority = groupNewPriority.get(originalPriority)!
+      } else {
+        // 这个组第一次出现，分配新优先级
+        groupNewPriority.set(originalPriority, currentPriority)
+        alias.priority = currentPriority
+        currentPriority++
+      }
+    }
+  })
+
+  aliases.value = items
+  draggedIndex.value = null
+  dragOverIndex.value = null
+}
+
+// ===== 优先级编辑 =====
+function startEditPriority(index: number) {
+  editingPriorityIndex.value = index
+}
+
+function finishEditPriority(index: number, event: FocusEvent) {
+  const input = event.target as HTMLInputElement
+  const newPriority = parseInt(input.value) || 1
+  aliases.value[index].priority = Math.max(1, newPriority)
+  editingPriorityIndex.value = null
+}
+
+function cancelEditPriority() {
+  editingPriorityIndex.value = null
+}
+
+// 关闭对话框
+function handleClose(value: boolean) {
+  if (!submitting.value) {
+    emit('update:open', value)
+  }
+}
+
+// 提交保存
+async function handleSubmit() {
+  if (submitting.value || !props.model) return
+
+  submitting.value = true
+  try {
+    // 过滤掉空的映射
+    const validAliases = aliases.value.filter(a => a.name.trim())
+
+    await updateModel(props.providerId, props.model.id, {
+      provider_model_mappings: validAliases.length > 0 ? validAliases : null
+    })
+
+    showSuccess('映射配置已保存')
+    emit('update:open', false)
+    emit('saved')
+  } catch (err: any) {
+    showError(err.response?.data?.detail || '保存失败', '错误')
+  } finally {
+    submitting.value = false
+  }
+}
+</script>

frontend/src/features/providers/components/ModelMappingDialog.vue

@@ -0,0 +1,777 @@
+<template>
+  <Dialog
+    :model-value="open"
+    :title="editingGroup ? '编辑模型映射' : '添加模型映射'"
+    :description="editingGroup ? '修改映射配置' : '为模型添加新的名称映射'"
+    :icon="Tag"
+    size="4xl"
+    @update:model-value="$emit('update:open', $event)"
+  >
+    <div class="space-y-4">
+      <!-- 第一行：目标模型 | 作用域 -->
+      <div class="flex gap-4">
+        <!-- 目标模型 -->
+        <div class="flex-1 space-y-1.5">
+          <Label class="text-xs">目标模型</Label>
+          <Select
+            v-model:open="modelSelectOpen"
+            :model-value="formData.modelId"
+            :disabled="!!editingGroup"
+            @update:model-value="formData.modelId = $event"
+          >
+            <SelectTrigger class="h-9">
+              <SelectValue placeholder="请选择模型" />
+            </SelectTrigger>
+            <SelectContent>
+              <SelectItem
+                v-for="model in models"
+                :key="model.id"
+                :value="model.id"
+              >
+                {{ model.global_model_display_name || model.provider_model_name }}
+              </SelectItem>
+            </SelectContent>
+          </Select>
+        </div>
+
+        <!-- 作用域 -->
+        <div class="flex-1 space-y-1.5">
+          <Label class="text-xs">作用域 <span class="text-muted-foreground font-normal">(不选则适用全部)</span></Label>
+          <div
+            v-if="providerApiFormats.length > 0"
+            class="flex flex-wrap gap-1.5 p-2 rounded-md border bg-muted/30 min-h-[36px]"
+          >
+            <button
+              v-for="format in providerApiFormats"
+              :key="format"
+              type="button"
+              class="px-2.5 py-0.5 rounded text-xs font-medium transition-colors"
+              :class="[
+                formData.apiFormats.includes(format)
+                  ? 'bg-primary text-primary-foreground'
+                  : 'bg-background border border-border hover:bg-muted'
+              ]"
+              @click="toggleApiFormat(format)"
+            >
+              {{ API_FORMAT_LABELS[format] || format }}
+            </button>
+          </div>
+          <div
+            v-else
+            class="h-9 flex items-center text-xs text-muted-foreground"
+          >
+            无可用格式
+          </div>
+        </div>
+      </div>
+
+      <!-- 第二行：两栏布局 -->
+      <div class="flex gap-4 items-stretch">
+        <!-- 左侧：上游模型列表 -->
+        <div class="flex-1 space-y-2">
+          <div class="flex items-center justify-between gap-2">
+            <span class="text-sm font-medium shrink-0">
+              上游模型
+            </span>
+            <div class="flex-1 relative">
+              <Search class="absolute left-2 top-1/2 -translate-y-1/2 w-3.5 h-3.5 text-muted-foreground" />
+              <Input
+                v-model="upstreamModelSearch"
+                placeholder="搜索模型..."
+                class="pl-7 h-7 text-xs"
+              />
+            </div>
+            <button
+              v-if="upstreamModelsLoaded"
+              type="button"
+              class="p-1.5 hover:bg-muted rounded-md transition-colors shrink-0"
+              title="刷新列表"
+              :disabled="refreshingUpstreamModels"
+              @click="refreshUpstreamModels"
+            >
+              <RefreshCw
+                class="w-3.5 h-3.5"
+                :class="{ 'animate-spin': refreshingUpstreamModels }"
+              />
+            </button>
+            <button
+              v-else-if="!fetchingUpstreamModels"
+              type="button"
+              class="p-1.5 hover:bg-muted rounded-md transition-colors shrink-0"
+              title="获取上游模型列表"
+              @click="fetchUpstreamModels"
+            >
+              <Zap class="w-3.5 h-3.5" />
+            </button>
+            <Loader2
+              v-else
+              class="w-3.5 h-3.5 animate-spin text-muted-foreground shrink-0"
+            />
+          </div>
+          <div class="border rounded-lg h-80 overflow-y-auto">
+            <template v-if="upstreamModelsLoaded">
+              <div
+                v-if="groupedAvailableUpstreamModels.length === 0"
+                class="flex flex-col items-center justify-center h-full text-muted-foreground"
+              >
+                <Zap class="w-10 h-10 mb-2 opacity-30" />
+                <p class="text-sm">
+                  {{ upstreamModelSearch ? '没有匹配的模型' : '所有模型已添加' }}
+                </p>
+              </div>
+              <div
+                v-else
+                class="p-2 space-y-2"
+              >
+                <!-- 按分组显示（可折叠） -->
+                <div
+                  v-for="group in groupedAvailableUpstreamModels"
+                  :key="group.api_format"
+                  class="border rounded-lg overflow-hidden"
+                >
+                  <div class="flex items-center gap-2 px-3 py-2 bg-muted/30">
+                    <button
+                      type="button"
+                      class="flex items-center gap-2 flex-1 hover:bg-muted/50 -mx-1 px-1 rounded transition-colors"
+                      @click="toggleGroupCollapse(group.api_format)"
+                    >
+                      <ChevronDown
+                        class="w-4 h-4 transition-transform shrink-0"
+                        :class="collapsedGroups.has(group.api_format) ? '-rotate-90' : ''"
+                      />
+                      <span class="text-xs font-medium">
+                        {{ API_FORMAT_LABELS[group.api_format] || group.api_format }}
+                      </span>
+                      <span class="text-xs text-muted-foreground">
+                        ({{ group.models.length }})
+                      </span>
+                    </button>
+                  </div>
+                  <div
+                    v-show="!collapsedGroups.has(group.api_format)"
+                    class="p-2 space-y-1 border-t"
+                  >
+                    <div
+                      v-for="model in group.models"
+                      :key="model.id"
+                      class="flex items-center gap-2 p-2 rounded-lg border transition-colors hover:bg-muted/30"
+                      :title="model.id"
+                    >
+                      <div class="flex-1 min-w-0">
+                        <p class="font-medium text-sm truncate">
+                          {{ model.id }}
+                        </p>
+                        <p class="text-xs text-muted-foreground truncate font-mono">
+                          {{ model.owned_by || model.id }}
+                        </p>
+                      </div>
+                      <button
+                        type="button"
+                        class="p-1 hover:bg-primary/10 rounded transition-colors shrink-0"
+                        title="添加到映射"
+                        @click="addUpstreamModel(model.id)"
+                      >
+                        <ChevronRight class="w-4 h-4 text-muted-foreground hover:text-primary" />
+                      </button>
+                    </div>
+                  </div>
+                </div>
+              </div>
+            </template>
+
+            <!-- 未加载状态 -->
+            <div
+              v-else
+              class="flex flex-col items-center justify-center h-full text-muted-foreground"
+            >
+              <Zap class="w-10 h-10 mb-2 opacity-30" />
+              <p class="text-sm">
+                点击右上角按钮
+              </p>
+              <p class="text-xs mt-1">
+                从上游获取可用模型
+              </p>
+            </div>
+          </div>
+        </div>
+
+        <!-- 右侧：映射名称列表 -->
+        <div class="flex-1 space-y-2">
+          <div class="flex items-center justify-between">
+            <p class="text-sm font-medium">
+              映射名称
+            </p>
+            <button
+              type="button"
+              class="p-1.5 hover:bg-muted rounded-md transition-colors"
+              title="手动添加"
+              @click="addAliasItem"
+            >
+              <Plus class="w-3.5 h-3.5" />
+            </button>
+          </div>
+          <div class="border rounded-lg h-80 overflow-y-auto">
+            <div
+              v-if="formData.aliases.length === 0"
+              class="flex flex-col items-center justify-center h-full text-muted-foreground"
+            >
+              <Tag class="w-10 h-10 mb-2 opacity-30" />
+              <p class="text-sm">
+                从左侧选择模型
+              </p>
+              <p class="text-xs mt-1">
+                或点击上方"手动添加"
+              </p>
+            </div>
+            <div
+              v-else
+              class="p-2 space-y-1"
+            >
+              <div
+                v-for="(alias, index) in formData.aliases"
+                :key="`alias-${index}`"
+                class="group flex items-center gap-2 p-2 rounded-lg border transition-colors hover:bg-muted/30"
+                :class="[
+                  draggedIndex === index ? 'bg-primary/5' : '',
+                  dragOverIndex === index ? 'bg-primary/10 border-primary' : ''
+                ]"
+                draggable="true"
+                @dragstart="handleDragStart(index, $event)"
+                @dragend="handleDragEnd"
+                @dragover.prevent="handleDragOver(index)"
+                @dragleave="handleDragLeave"
+                @drop="handleDrop(index)"
+              >
+                <!-- 删除按钮 -->
+                <button
+                  type="button"
+                  class="p-1 hover:bg-destructive/10 rounded transition-colors shrink-0"
+                  title="移除"
+                  @click="removeAliasItem(index)"
+                >
+                  <ChevronLeft class="w-4 h-4 text-muted-foreground hover:text-destructive" />
+                </button>
+
+                <!-- 优先级 -->
+                <div class="shrink-0">
+                  <input
+                    v-if="editingPriorityIndex === index"
+                    type="number"
+                    min="1"
+                    :value="alias.priority"
+                    class="w-7 h-6 rounded bg-background border border-primary text-xs text-center focus:outline-none [appearance:textfield] [&::-webkit-outer-spin-button]:appearance-none [&::-webkit-inner-spin-button]:appearance-none"
+                    autofocus
+                    @blur="finishEditPriority(index, $event)"
+                    @keydown.enter="($event.target as HTMLInputElement).blur()"
+                    @keydown.escape="cancelEditPriority"
+                  >
+                  <div
+                    v-else
+                    class="w-6 h-6 rounded bg-muted/50 flex items-center justify-center text-xs text-muted-foreground cursor-pointer hover:bg-primary/10 hover:text-primary"
+                    title="点击编辑优先级"
+                    @click.stop="startEditPriority(index)"
+                  >
+                    {{ alias.priority }}
+                  </div>
+                </div>
+
+                <!-- 名称显示/编辑 -->
+                <div class="flex-1 min-w-0">
+                  <Input
+                    v-if="alias.isEditing"
+                    v-model="alias.name"
+                    placeholder="输入映射名称"
+                    class="h-7 text-xs"
+                    autofocus
+                    @blur="alias.isEditing = false"
+                    @keydown.enter="alias.isEditing = false"
+                  />
+                  <p
+                    v-else
+                    class="font-medium text-sm truncate cursor-pointer hover:text-primary"
+                    title="点击编辑"
+                    @click="alias.isEditing = true"
+                  >
+                    {{ alias.name || '点击输入名称' }}
+                  </p>
+                </div>
+
+                <!-- 拖拽手柄 -->
+                <div class="cursor-grab active:cursor-grabbing text-muted-foreground/30 group-hover:text-muted-foreground shrink-0">
+                  <GripVertical class="w-4 h-4" />
+                </div>
+              </div>
+            </div>
+            <!-- 拖拽提示 -->
+            <div
+              v-if="formData.aliases.length > 1"
+              class="px-3 py-1.5 bg-muted/30 border-t text-xs text-muted-foreground text-center"
+            >
+              拖拽调整优先级顺序
+            </div>
+          </div>
+        </div>
+      </div>
+    </div>
+
+    <template #footer>
+      <Button
+        variant="outline"
+        @click="$emit('update:open', false)"
+      >
+        取消
+      </Button>
+      <Button
+        :disabled="submitting || !formData.modelId || formData.aliases.length === 0 || !hasValidAliases"
+        @click="handleSubmit"
+      >
+        <Loader2
+          v-if="submitting"
+          class="w-4 h-4 mr-2 animate-spin"
+        />
+        {{ editingGroup ? '保存' : '添加' }}
+      </Button>
+    </template>
+  </Dialog>
+</template>
+
+<script setup lang="ts">
+import { ref, computed, watch } from 'vue'
+import { Tag, Loader2, GripVertical, Zap, Search, RefreshCw, ChevronDown, ChevronRight, ChevronLeft, Plus } from 'lucide-vue-next'
+import {
+  Button,
+  Input,
+  Label,
+  Dialog,
+  Select,
+  SelectContent,
+  SelectItem,
+  SelectTrigger,
+  SelectValue,
+} from '@/components/ui'
+import { useToast } from '@/composables/useToast'
+import {
+  API_FORMAT_LABELS,
+  type Model,
+  type ProviderModelAlias
+} from '@/api/endpoints'
+import { updateModel } from '@/api/endpoints/models'
+import { useUpstreamModelsCache, type UpstreamModel } from '../composables/useUpstreamModelsCache'
+
+interface FormAlias {
+  name: string
+  priority: number
+  isEditing?: boolean
+}
+
+export interface AliasGroup {
+  model: Model
+  apiFormatsKey: string
+  apiFormats: string[]
+  aliases: ProviderModelAlias[]
+}
+
+const props = defineProps<{
+  open: boolean
+  providerId: string
+  providerApiFormats: string[]
+  models: Model[]
+  editingGroup?: AliasGroup | null
+}>()
+
+const emit = defineEmits<{
+  'update:open': [value: boolean]
+  'saved': []
+}>()
+
+const { error: showError, success: showSuccess } = useToast()
+const { fetchModels: fetchCachedModels, clearCache, getCachedModels } = useUpstreamModelsCache()
+
+// 状态
+const submitting = ref(false)
+const modelSelectOpen = ref(false)
+
+// 拖拽状态
+const draggedIndex = ref<number | null>(null)
+const dragOverIndex = ref<number | null>(null)
+
+// 优先级编辑状态
+const editingPriorityIndex = ref<number | null>(null)
+
+// 快速添加（上游模型）状态
+const fetchingUpstreamModels = ref(false)
+const refreshingUpstreamModels = ref(false)
+const upstreamModelsLoaded = ref(false)
+const upstreamModels = ref<UpstreamModel[]>([])
+const upstreamModelSearch = ref('')
+
+// 分组折叠状态
+const collapsedGroups = ref<Set<string>>(new Set())
+
+// 表单数据
+const formData = ref<{
+  modelId: string
+  apiFormats: string[]
+  aliases: FormAlias[]
+}>({
+  modelId: '',
+  apiFormats: [],
+  aliases: []
+})
+
+// 检查是否有有效的映射
+const hasValidAliases = computed(() => {
+  return formData.value.aliases.some(a => a.name.trim())
+})
+
+// 过滤和排序后的上游模型列表
+const filteredUpstreamModels = computed(() => {
+  const searchText = upstreamModelSearch.value.toLowerCase().trim()
+  let result = [...upstreamModels.value]
+
+  result.sort((a, b) => a.id.localeCompare(b.id))
+
+  if (searchText) {
+    const keywords = searchText.split(/\s+/).filter(k => k.length > 0)
+    result = result.filter(m => {
+      const searchableText = `${m.id} ${m.owned_by || ''} ${m.api_format || ''}`.toLowerCase()
+      return keywords.every(keyword => searchableText.includes(keyword))
+    })
+  }
+
+  return result
+})
+
+// 按 API 格式分组的上游模型列表
+interface UpstreamModelGroup {
+  api_format: string
+  models: Array<{ id: string; owned_by?: string; api_format?: string }>
+}
+
+const groupedAvailableUpstreamModels = computed<UpstreamModelGroup[]>(() => {
+  const addedNames = new Set(formData.value.aliases.map(a => a.name.trim()))
+  const availableModels = filteredUpstreamModels.value.filter(m => !addedNames.has(m.id))
+
+  const groups = new Map<string, UpstreamModelGroup>()
+
+  for (const model of availableModels) {
+    const format = model.api_format || 'UNKNOWN'
+    if (!groups.has(format)) {
+      groups.set(format, { api_format: format, models: [] })
+    }
+    groups.get(format)!.models.push(model)
+  }
+
+  const order = Object.keys(API_FORMAT_LABELS)
+  return Array.from(groups.values()).sort((a, b) => {
+    const aIndex = order.indexOf(a.api_format)
+    const bIndex = order.indexOf(b.api_format)
+    if (aIndex === -1 && bIndex === -1) return a.api_format.localeCompare(b.api_format)
+    if (aIndex === -1) return 1
+    if (bIndex === -1) return -1
+    return aIndex - bIndex
+  })
+})
+
+// 监听打开状态
+watch(() => props.open, (isOpen) => {
+  if (isOpen) {
+    initForm()
+  }
+})
+
+// 初始化表单
+function initForm() {
+  if (props.editingGroup) {
+    formData.value = {
+      modelId: props.editingGroup.model.id,
+      apiFormats: [...props.editingGroup.apiFormats],
+      aliases: props.editingGroup.aliases.map(a => ({ name: a.name, priority: a.priority }))
+    }
+  } else {
+    formData.value = {
+      modelId: '',
+      apiFormats: [],
+      aliases: []
+    }
+  }
+  // 重置状态
+  editingPriorityIndex.value = null
+  draggedIndex.value = null
+  dragOverIndex.value = null
+  upstreamModelSearch.value = ''
+  collapsedGroups.value = new Set()
+
+  // 检查缓存，如果有缓存数据则直接使用
+  const cachedModels = getCachedModels(props.providerId)
+  if (cachedModels) {
+    upstreamModels.value = cachedModels
+    upstreamModelsLoaded.value = true
+    // 默认折叠所有分组
+    for (const model of cachedModels) {
+      if (model.api_format) {
+        collapsedGroups.value.add(model.api_format)
+      }
+    }
+  } else {
+    upstreamModelsLoaded.value = false
+    upstreamModels.value = []
+  }
+}
+
+// 切换 API 格式
+function toggleApiFormat(format: string) {
+  const index = formData.value.apiFormats.indexOf(format)
+  if (index >= 0) {
+    formData.value.apiFormats.splice(index, 1)
+  } else {
+    formData.value.apiFormats.push(format)
+  }
+}
+
+// 切换分组折叠状态
+function toggleGroupCollapse(apiFormat: string) {
+  if (collapsedGroups.value.has(apiFormat)) {
+    collapsedGroups.value.delete(apiFormat)
+  } else {
+    collapsedGroups.value.add(apiFormat)
+  }
+}
+
+// 添加映射项
+function addAliasItem() {
+  const maxPriority = formData.value.aliases.length > 0
+    ? Math.max(...formData.value.aliases.map(a => a.priority))
+    : 0
+  formData.value.aliases.push({ name: '', priority: maxPriority + 1, isEditing: true })
+}
+
+// 删除映射项
+function removeAliasItem(index: number) {
+  formData.value.aliases.splice(index, 1)
+}
+
+// ===== 拖拽排序 =====
+function handleDragStart(index: number, event: DragEvent) {
+  draggedIndex.value = index
+  if (event.dataTransfer) {
+    event.dataTransfer.effectAllowed = 'move'
+  }
+}
+
+function handleDragEnd() {
+  draggedIndex.value = null
+  dragOverIndex.value = null
+}
+
+function handleDragOver(index: number) {
+  if (draggedIndex.value !== null && draggedIndex.value !== index) {
+    dragOverIndex.value = index
+  }
+}
+
+function handleDragLeave() {
+  dragOverIndex.value = null
+}
+
+function handleDrop(targetIndex: number) {
+  const dragIndex = draggedIndex.value
+  if (dragIndex === null || dragIndex === targetIndex) {
+    dragOverIndex.value = null
+    return
+  }
+
+  const items = [...formData.value.aliases]
+  const draggedItem = items[dragIndex]
+
+  const originalPriorityMap = new Map<number, number>()
+  items.forEach((alias, idx) => {
+    originalPriorityMap.set(idx, alias.priority)
+  })
+
+  items.splice(dragIndex, 1)
+  items.splice(targetIndex, 0, draggedItem)
+
+  const groupNewPriority = new Map<number, number>()
+  let currentPriority = 1
+
+  items.forEach((alias) => {
+    const originalIdx = formData.value.aliases.findIndex(a => a === alias)
+    const originalPriority = originalIdx >= 0 ? originalPriorityMap.get(originalIdx)! : alias.priority
+
+    if (alias === draggedItem) {
+      alias.priority = currentPriority
+      currentPriority++
+    } else {
+      if (groupNewPriority.has(originalPriority)) {
+        alias.priority = groupNewPriority.get(originalPriority)!
+      } else {
+        groupNewPriority.set(originalPriority, currentPriority)
+        alias.priority = currentPriority
+        currentPriority++
+      }
+    }
+  })
+
+  formData.value.aliases = items
+  draggedIndex.value = null
+  dragOverIndex.value = null
+}
+
+// ===== 优先级编辑 =====
+function startEditPriority(index: number) {
+  editingPriorityIndex.value = index
+}
+
+function finishEditPriority(index: number, event: FocusEvent) {
+  const input = event.target as HTMLInputElement
+  const newPriority = parseInt(input.value) || 1
+  formData.value.aliases[index].priority = Math.max(1, newPriority)
+  editingPriorityIndex.value = null
+}
+
+function cancelEditPriority() {
+  editingPriorityIndex.value = null
+}
+
+// ===== 快速添加（上游模型）=====
+async function fetchUpstreamModels() {
+  if (!props.providerId) return
+
+  upstreamModelSearch.value = ''
+  fetchingUpstreamModels.value = true
+
+  try {
+    const result = await fetchCachedModels(props.providerId)
+    if (result) {
+      if (result.error) {
+        showError(result.error, '错误')
+      } else {
+        upstreamModels.value = result.models
+        upstreamModelsLoaded.value = true
+        // 默认折叠所有分组
+        for (const model of result.models) {
+          if (model.api_format) {
+            collapsedGroups.value.add(model.api_format)
+          }
+        }
+      }
+    }
+  } finally {
+    fetchingUpstreamModels.value = false
+  }
+}
+
+function addUpstreamModel(modelId: string) {
+  if (formData.value.aliases.some(a => a.name === modelId)) {
+    return
+  }
+
+  const maxPriority = formData.value.aliases.length > 0
+    ? Math.max(...formData.value.aliases.map(a => a.priority))
+    : 0
+
+  formData.value.aliases.push({ name: modelId, priority: maxPriority + 1 })
+}
+
+async function refreshUpstreamModels() {
+  if (!props.providerId || refreshingUpstreamModels.value) return
+
+  refreshingUpstreamModels.value = true
+  clearCache(props.providerId)
+
+  try {
+    const result = await fetchCachedModels(props.providerId, true)
+    if (result) {
+      if (result.error) {
+        showError(result.error, '错误')
+      } else {
+        upstreamModels.value = result.models
+      }
+    }
+  } finally {
+    refreshingUpstreamModels.value = false
+  }
+}
+
+// 生成作用域唯一键
+function getApiFormatsKey(formats: string[] | undefined): string {
+  if (!formats || formats.length === 0) return ''
+  return [...formats].sort().join(',')
+}
+
+// 提交表单
+async function handleSubmit() {
+  if (submitting.value) return
+  if (!formData.value.modelId || formData.value.aliases.length === 0) return
+
+  const validAliases = formData.value.aliases.filter(a => a.name.trim())
+  if (validAliases.length === 0) {
+    showError('请至少添加一个有效的映射名称', '错误')
+    return
+  }
+
+  submitting.value = true
+  try {
+    const targetModel = props.models.find(m => m.id === formData.value.modelId)
+    if (!targetModel) {
+      showError('模型不存在', '错误')
+      return
+    }
+
+    const currentAliases = targetModel.provider_model_mappings || []
+    let newAliases: ProviderModelAlias[]
+
+    const buildAlias = (a: FormAlias): ProviderModelAlias => ({
+      name: a.name.trim(),
+      priority: a.priority,
+      ...(formData.value.apiFormats.length > 0 ? { api_formats: formData.value.apiFormats } : {})
+    })
+
+    if (props.editingGroup) {
+      const oldApiFormatsKey = props.editingGroup.apiFormatsKey
+      const oldAliasNames = new Set(props.editingGroup.aliases.map(a => a.name))
+
+      const filteredAliases = currentAliases.filter((a: ProviderModelAlias) => {
+        const currentKey = getApiFormatsKey(a.api_formats)
+        return !(currentKey === oldApiFormatsKey && oldAliasNames.has(a.name))
+      })
+
+      const existingNames = new Set(filteredAliases.map((a: ProviderModelAlias) => a.name))
+      const duplicates = validAliases.filter(a => existingNames.has(a.name.trim()))
+      if (duplicates.length > 0) {
+        showError(`以下映射名称已存在：${duplicates.map(d => d.name).join(', ')}`, '错误')
+        return
+      }
+
+      newAliases = [
+        ...filteredAliases,
+        ...validAliases.map(buildAlias)
+      ]
+    } else {
+      const existingNames = new Set(currentAliases.map((a: ProviderModelAlias) => a.name))
+      const duplicates = validAliases.filter(a => existingNames.has(a.name.trim()))
+      if (duplicates.length > 0) {
+        showError(`以下映射名称已存在：${duplicates.map(d => d.name).join(', ')}`, '错误')
+        return
+      }
+      newAliases = [
+        ...currentAliases,
+        ...validAliases.map(buildAlias)
+      ]
+    }
+
+    await updateModel(props.providerId, targetModel.id, {
+      provider_model_mappings: newAliases
+    })
+
+    showSuccess(props.editingGroup ? '映射组已更新' : '映射已添加')
+    emit('update:open', false)
+    emit('saved')
+  } catch (err: any) {
+    showError(err.response?.data?.detail || '操作失败', '错误')
+  } finally {
+    submitting.value = false
+  }
+}
+</script>

frontend/src/features/providers/components/PriorityManagementDialog.vue

@@ -312,8 +312,41 @@
 
     <template #footer>
       <div class="flex items-center justify-between w-full">
-        <div class="text-xs text-muted-foreground">
-          当前模式: <span class="font-medium">{{ activeMainTab === 'provider' ? '提供商优先' : 'Key 优先' }}</span>
+        <div class="flex items-center gap-4">
+          <div class="text-xs text-muted-foreground">
+            当前模式: <span class="font-medium">{{ activeMainTab === 'provider' ? '提供商优先' : 'Key 优先' }}</span>
+          </div>
+          <div class="flex items-center gap-2 pl-4 border-l border-border">
+            <span class="text-xs text-muted-foreground">调度:</span>
+            <div class="flex gap-0.5 p-0.5 bg-muted/40 rounded-md">
+              <button
+                type="button"
+                class="px-2 py-1 text-xs font-medium rounded transition-all"
+                :class="[
+                  schedulingMode === 'fixed_order'
+                    ? 'bg-primary text-primary-foreground shadow-sm'
+                    : 'text-muted-foreground hover:text-foreground hover:bg-muted/50'
+                ]"
+                title="严格按优先级顺序，不考虑缓存"
+                @click="schedulingMode = 'fixed_order'"
+              >
+                固定顺序
+              </button>
+              <button
+                type="button"
+                class="px-2 py-1 text-xs font-medium rounded transition-all"
+                :class="[
+                  schedulingMode === 'cache_affinity'
+                    ? 'bg-primary text-primary-foreground shadow-sm'
+                    : 'text-muted-foreground hover:text-foreground hover:bg-muted/50'
+                ]"
+                title="优先使用已缓存的Provider，利用Prompt Cache"
+                @click="schedulingMode = 'cache_affinity'"
+              >
+                缓存亲和
+              </button>
+            </div>
+          </div>
         </div>
         <div class="flex gap-2">
           <Button
@@ -410,6 +443,9 @@ const saving = ref(false)
 // Key 优先级编辑状态
 const editingKeyPriority = ref<Record<string, string | null>>({})  // format -> keyId
 
+// 调度模式状态
+const schedulingMode = ref<'fixed_order' | 'cache_affinity'>('cache_affinity')
+
 // 可用的 API 格式
 const availableFormats = computed(() => {
   return Object.keys(keysByFormat.value).sort()
@@ -433,11 +469,18 @@ watch(internalOpen, async (open) => {
 // 加载当前的优先级模式配置
 async function loadCurrentPriorityMode() {
   try {
-    const response = await adminApi.getSystemConfig('provider_priority_mode')
-    const currentMode = response.value || 'provider'
+    const [priorityResponse, schedulingResponse] = await Promise.all([
+      adminApi.getSystemConfig('provider_priority_mode'),
+      adminApi.getSystemConfig('scheduling_mode')
+    ])
+    const currentMode = priorityResponse.value || 'provider'
     activeMainTab.value = currentMode === 'global_key' ? 'key' : 'provider'
+
+    const currentSchedulingMode = schedulingResponse.value || 'cache_affinity'
+    schedulingMode.value = currentSchedulingMode === 'fixed_order' ? 'fixed_order' : 'cache_affinity'
   } catch {
     activeMainTab.value = 'provider'
+    schedulingMode.value = 'cache_affinity'
   }
 }
 
@@ -611,11 +654,19 @@ async function save() {
 
     const newMode = activeMainTab.value === 'key' ? 'global_key' : 'provider'
 
-    await adminApi.updateSystemConfig(
-      'provider_priority_mode',
-      newMode,
-      'Provider/Key 优先级策略：provider(提供商优先模式) 或 global_key(全局Key优先模式)'
-    )
+    // 保存优先级模式和调度模式
+    await Promise.all([
+      adminApi.updateSystemConfig(
+        'provider_priority_mode',
+        newMode,
+        'Provider/Key 优先级策略：provider(提供商优先模式) 或 global_key(全局Key优先模式)'
+      ),
+      adminApi.updateSystemConfig(
+        'scheduling_mode',
+        schedulingMode.value,
+        '调度模式：fixed_order(固定顺序模式) 或 cache_affinity(缓存亲和模式)'
+      )
+    ])
 
     const providerUpdates = sortedProviders.value.map((provider, index) =>
       updateProvider(provider.id, { provider_priority: index + 1 })

frontend/src/features/providers/components/ProviderDetailDrawer.vue

@@ -528,10 +528,10 @@
                 @batch-assign="handleBatchAssign"
               />
 
-              <!-- 模型映射 -->
-              <MappingsTab
+              <!-- 模型名称映射 -->
+              <ModelAliasesTab
                 v-if="provider"
-                :key="`mappings-${provider.id}`"
+                :key="`aliases-${provider.id}`"
                 :provider="provider"
                 @refresh="handleRelatedDataRefresh"
               />
@@ -655,6 +655,7 @@ import {
   GripVertical,
   Copy
 } from 'lucide-vue-next'
+import { useEscapeKey } from '@/composables/useEscapeKey'
 import Button from '@/components/ui/button.vue'
 import Badge from '@/components/ui/badge.vue'
 import Card from '@/components/ui/card.vue'
@@ -663,8 +664,8 @@ import { getProvider, getProviderEndpoints } from '@/api/endpoints'
 import {
   KeyFormDialog,
   KeyAllowedModelsDialog,
-  MappingsTab,
   ModelsTab,
+  ModelAliasesTab,
   BatchAssignModelsDialog
 } from '@/features/providers/components'
 import EndpointFormDialog from '@/features/providers/components/EndpointFormDialog.vue'
@@ -1296,6 +1297,16 @@ async function loadEndpoints() {
     showError(err.response?.data?.detail || '加载端点失败', '错误')
   }
 }
+
+// 添加 ESC 键监听
+useEscapeKey(() => {
+  if (props.open) {
+    handleClose()
+  }
+}, {
+  disableOnInput: true,
+  once: false
+})
 </script>
 
 <style scoped>

frontend/src/features/providers/components/index.ts

@@ -7,6 +7,7 @@ export { default as ProviderModelFormDialog } from './ProviderModelFormDialog.vu
 export { default as ProviderDetailDrawer } from './ProviderDetailDrawer.vue'
 export { default as EndpointHealthTimeline } from './EndpointHealthTimeline.vue'
 export { default as BatchAssignModelsDialog } from './BatchAssignModelsDialog.vue'
+export { default as ModelAliasDialog } from './ModelAliasDialog.vue'
 
-export { default as MappingsTab } from './provider-tabs/MappingsTab.vue'
 export { default as ModelsTab } from './provider-tabs/ModelsTab.vue'
+export { default as ModelAliasesTab } from './provider-tabs/ModelAliasesTab.vue'

frontend/src/features/providers/components/provider-tabs/MappingsTab.vue

@@ -1,310 +0,0 @@
-<template>
-  <Card class="overflow-hidden">
-    <!-- 标题头部 -->
-    <div class="p-4 border-b border-border/60">
-      <div class="flex items-center justify-between">
-        <div class="flex items-center gap-2">
-          <h3 class="text-sm font-semibold leading-none">
-            别名与映射管理
-          </h3>
-        </div>
-        <Button
-          v-if="!hideAddButton"
-          variant="outline"
-          size="sm"
-          class="h-8"
-          @click="openCreateDialog"
-        >
-          <Plus class="w-3.5 h-3.5 mr-1.5" />
-          创建别名/映射
-        </Button>
-      </div>
-    </div>
-
-    <!-- 加载状态 -->
-    <div
-      v-if="loading"
-      class="flex items-center justify-center py-12"
-    >
-      <div class="animate-spin rounded-full h-8 w-8 border-b-2 border-primary" />
-    </div>
-
-    <!-- 别名列表 -->
-    <div
-      v-else-if="mappings.length > 0"
-      class="overflow-x-auto"
-    >
-      <table class="w-full text-sm">
-        <thead class="bg-muted/50 text-xs uppercase tracking-wide text-muted-foreground">
-          <tr>
-            <th class="text-left px-4 py-3 font-semibold">
-              名称
-            </th>
-            <th class="text-left px-4 py-3 font-semibold w-24">
-              类型
-            </th>
-            <th class="text-left px-4 py-3 font-semibold">
-              指向模型
-            </th>
-            <th
-              v-if="!hideAddButton"
-              class="px-4 py-3 font-semibold w-28 text-center"
-            >
-              操作
-            </th>
-          </tr>
-        </thead>
-        <tbody>
-          <tr
-            v-for="mapping in mappings"
-            :key="mapping.id"
-            class="border-b border-border/40 last:border-b-0 hover:bg-muted/30 transition-colors"
-          >
-            <td class="px-4 py-3">
-              <div class="flex items-center gap-2">
-                <!-- 状态指示灯 -->
-                <span
-                  class="w-2 h-2 rounded-full shrink-0"
-                  :class="mapping.is_active ? 'bg-green-500' : 'bg-gray-300'"
-                  :title="mapping.is_active ? '活跃' : '停用'"
-                />
-                <span class="font-mono">{{ mapping.alias }}</span>
-              </div>
-            </td>
-            <td class="px-4 py-3">
-              <Badge
-                variant="secondary"
-                class="text-xs"
-              >
-                {{ mapping.mapping_type === 'mapping' ? '映射' : '别名' }}
-              </Badge>
-            </td>
-            <td class="px-4 py-3">
-              {{ mapping.global_model_display_name || mapping.global_model_name }}
-            </td>
-            <td
-              v-if="!hideAddButton"
-              class="px-4 py-3"
-            >
-              <div class="flex justify-center gap-1.5">
-                <Button
-                  variant="ghost"
-                  size="icon"
-                  class="h-8 w-8"
-                  title="编辑"
-                  @click="openEditDialog(mapping)"
-                >
-                  <Edit class="w-3.5 h-3.5" />
-                </Button>
-                <Button
-                  variant="ghost"
-                  size="icon"
-                  class="h-8 w-8"
-                  :disabled="togglingId === mapping.id"
-                  :title="mapping.is_active ? '点击停用' : '点击启用'"
-                  @click="toggleActive(mapping)"
-                >
-                  <Power class="w-3.5 h-3.5" />
-                </Button>
-                <Button
-                  variant="ghost"
-                  size="icon"
-                  class="h-8 w-8 text-destructive hover:text-destructive"
-                  title="删除"
-                  @click="confirmDelete(mapping)"
-                >
-                  <Trash2 class="w-3.5 h-3.5" />
-                </Button>
-              </div>
-            </td>
-          </tr>
-        </tbody>
-      </table>
-    </div>
-
-    <!-- 空状态 -->
-    <div
-      v-else
-      class="p-8 text-center text-muted-foreground"
-    >
-      <ArrowLeftRight class="w-12 h-12 mx-auto mb-3 opacity-50" />
-      <p class="text-sm">
-        暂无特定别名/映射
-      </p>
-      <p class="text-xs mt-1">
-        点击上方按钮添加
-      </p>
-    </div>
-  </Card>
-
-  <!-- 使用共享的 AliasDialog 组件 -->
-  <AliasDialog
-    :open="dialogOpen"
-    :editing-alias="editingAlias"
-    :global-models="availableModels"
-    :fixed-provider="fixedProviderOption"
-    :show-provider-select="true"
-    @update:open="handleDialogVisibility"
-    @submit="handleAliasSubmit"
-  />
-</template>
-
-<script setup lang="ts">
-import { ref, computed, onMounted } from 'vue'
-import { ArrowLeftRight, Plus, Edit, Trash2, Power } from 'lucide-vue-next'
-import Card from '@/components/ui/card.vue'
-import Badge from '@/components/ui/badge.vue'
-import Button from '@/components/ui/button.vue'
-import AliasDialog from '@/features/models/components/AliasDialog.vue'
-import { useToast } from '@/composables/useToast'
-import {
-  getAliases,
-  createAlias,
-  updateAlias,
-  deleteAlias,
-  type ModelAlias,
-  type CreateModelAliasRequest,
-  type UpdateModelAliasRequest,
-} from '@/api/endpoints/aliases'
-import { listGlobalModels, type GlobalModelResponse } from '@/api/global-models'
-
-const props = withDefaults(defineProps<{
-  provider: any
-  hideAddButton?: boolean
-}>(), {
-  hideAddButton: false
-})
-
-const emit = defineEmits<{
-  refresh: []
-}>()
-
-const { success, error: showError } = useToast()
-
-// 状态
-const loading = ref(false)
-const submitting = ref(false)
-const togglingId = ref<string | null>(null)
-const mappings = ref<ModelAlias[]>([])
-const availableModels = ref<GlobalModelResponse[]>([])
-const dialogOpen = ref(false)
-const editingAlias = ref<ModelAlias | null>(null)
-
-// 固定的 Provider 选项（传递给 AliasDialog）
-const fixedProviderOption = computed(() => ({
-  id: props.provider.id,
-  name: props.provider.name,
-  display_name: props.provider.display_name
-}))
-
-// 加载映射 (实际返回的是该 Provider 的别名列表)
-async function loadMappings() {
-  try {
-    loading.value = true
-    mappings.value = await getAliases({ provider_id: props.provider.id })
-  } catch (err: any) {
-    showError(err.response?.data?.detail || '加载失败', '错误')
-  } finally {
-    loading.value = false
-  }
-}
-
-// 加载可用的 GlobalModel 列表
-async function loadAvailableModels() {
-  try {
-    const response = await listGlobalModels({ limit: 1000, is_active: true })
-    availableModels.value = response.models || []
-  } catch (err: any) {
-    showError(err.response?.data?.detail || '加载模型列表失败', '错误')
-  }
-}
-
-// 打开创建对话框
-function openCreateDialog() {
-  editingAlias.value = null
-  dialogOpen.value = true
-}
-
-// 打开编辑对话框
-function openEditDialog(alias: ModelAlias) {
-  editingAlias.value = alias
-  dialogOpen.value = true
-}
-
-// 处理对话框可见性变化
-function handleDialogVisibility(value: boolean) {
-  dialogOpen.value = value
-  if (!value) {
-    editingAlias.value = null
-  }
-}
-
-// 处理别名提交（来自 AliasDialog 组件）
-async function handleAliasSubmit(data: CreateModelAliasRequest | UpdateModelAliasRequest, isEdit: boolean) {
-  submitting.value = true
-  try {
-    if (isEdit && editingAlias.value) {
-      // 更新
-      await updateAlias(editingAlias.value.id, data as UpdateModelAliasRequest)
-      success(data.mapping_type === 'mapping' ? '映射已更新' : '别名已更新')
-    } else {
-      // 创建 - 确保 provider_id 设置为当前 Provider
-      const createData = data as CreateModelAliasRequest
-      createData.provider_id = props.provider.id
-      await createAlias(createData)
-      success(data.mapping_type === 'mapping' ? '映射已创建' : '别名已创建')
-    }
-    dialogOpen.value = false
-    editingAlias.value = null
-    await loadMappings()
-    emit('refresh')
-  } catch (err: any) {
-    const detail = err.response?.data?.detail || err.message
-    let errorMessage = detail
-    if (detail === '映射已存在') {
-      errorMessage = '该名称已存在，请使用其他名称'
-    }
-    showError(errorMessage, isEdit ? '更新失败' : '创建失败')
-  } finally {
-    submitting.value = false
-  }
-}
-
-// 切换启用状态
-async function toggleActive(alias: ModelAlias) {
-  if (togglingId.value) return
-
-  togglingId.value = alias.id
-  try {
-    const newStatus = !alias.is_active
-    await updateAlias(alias.id, { is_active: newStatus })
-    alias.is_active = newStatus
-  } catch (err: any) {
-    showError(err.response?.data?.detail || '操作失败', '错误')
-  } finally {
-    togglingId.value = null
-  }
-}
-
-// 确认删除
-async function confirmDelete(alias: ModelAlias) {
-  const typeName = alias.mapping_type === 'mapping' ? '映射' : '别名'
-  if (!confirm(`确定要删除${typeName} "${alias.alias}" 吗？`)) {
-    return
-  }
-
-  try {
-    await deleteAlias(alias.id)
-    success(`${typeName}已删除`)
-    await loadMappings()
-    emit('refresh')
-  } catch (err: any) {
-    showError(err.response?.data?.detail || err.message, '删除失败')
-  }
-}
-
-onMounted(() => {
-  loadMappings()
-  loadAvailableModels()
-})
-</script>

frontend/src/features/providers/components/provider-tabs/ModelAliasesTab.vue

@@ -0,0 +1,352 @@
+<template>
+  <Card class="overflow-hidden">
+    <!-- 标题头部 -->
+    <div class="p-4 border-b border-border/60">
+      <div class="flex items-center justify-between">
+        <h3 class="text-sm font-semibold flex items-center gap-2">
+          模型名称映射
+        </h3>
+        <Button
+          variant="outline"
+          size="sm"
+          class="h-8"
+          @click="openAddDialog"
+        >
+          <Plus class="w-3.5 h-3.5 mr-1.5" />
+          添加映射
+        </Button>
+      </div>
+    </div>
+
+    <!-- 加载状态 -->
+    <div
+      v-if="loading"
+      class="flex items-center justify-center py-12"
+    >
+      <div class="animate-spin rounded-full h-8 w-8 border-b-2 border-primary" />
+    </div>
+
+    <!-- 分组映射列表 -->
+    <div
+      v-else-if="aliasGroups.length > 0"
+      class="divide-y divide-border/40"
+    >
+      <div
+        v-for="group in aliasGroups"
+        :key="`${group.model.id}-${group.apiFormatsKey}`"
+        class="transition-colors"
+      >
+        <!-- 分组头部（可点击展开） -->
+        <div
+          class="flex items-center justify-between px-4 py-3 hover:bg-muted/20 cursor-pointer"
+          @click="toggleAliasGroupExpand(`${group.model.id}-${group.apiFormatsKey}`)"
+        >
+          <div class="flex items-center gap-2 flex-1 min-w-0">
+            <!-- 展开/收起图标 -->
+            <ChevronRight
+              class="w-4 h-4 text-muted-foreground shrink-0 transition-transform"
+              :class="{ 'rotate-90': expandedAliasGroups.has(`${group.model.id}-${group.apiFormatsKey}`) }"
+            />
+            <!-- 模型名称 -->
+            <span class="font-semibold text-sm truncate">
+              {{ group.model.global_model_display_name || group.model.provider_model_name }}
+            </span>
+            <!-- 作用域标签 -->
+            <div class="flex items-center gap-1 shrink-0">
+              <Badge
+                v-if="group.apiFormats.length === 0"
+                variant="outline"
+                class="text-xs"
+              >
+                全部
+              </Badge>
+              <Badge
+                v-for="format in group.apiFormats"
+                v-else
+                :key="format"
+                variant="outline"
+                class="text-xs"
+              >
+                {{ API_FORMAT_LABELS[format] || format }}
+              </Badge>
+            </div>
+            <!-- 映射数量 -->
+            <span class="text-xs text-muted-foreground shrink-0">
+              ({{ group.aliases.length }} 个映射)
+            </span>
+          </div>
+          <!-- 操作按钮 -->
+          <div
+            class="flex items-center gap-1.5 ml-4 shrink-0"
+            @click.stop
+          >
+            <Button
+              variant="ghost"
+              size="icon"
+              class="h-8 w-8"
+              title="编辑映射组"
+              @click="editGroup(group)"
+            >
+              <Edit class="w-3.5 h-3.5" />
+            </Button>
+            <Button
+              variant="ghost"
+              size="icon"
+              class="h-8 w-8 text-destructive hover:text-destructive"
+              title="删除映射组"
+              @click="deleteGroup(group)"
+            >
+              <Trash2 class="w-3.5 h-3.5" />
+            </Button>
+          </div>
+        </div>
+
+        <!-- 展开的映射列表 -->
+        <div
+          v-show="expandedAliasGroups.has(`${group.model.id}-${group.apiFormatsKey}`)"
+          class="bg-muted/30 border-t border-border/30"
+        >
+          <div class="px-4 py-2 space-y-1">
+            <div
+              v-for="mapping in group.aliases"
+              :key="mapping.name"
+              class="flex items-center gap-2 py-1"
+            >
+              <!-- 优先级标签 -->
+              <span class="inline-flex items-center justify-center w-5 h-5 rounded bg-background border text-xs font-medium shrink-0">
+                {{ mapping.priority }}
+              </span>
+              <!-- 映射名称 -->
+              <span class="font-mono text-sm truncate">
+                {{ mapping.name }}
+              </span>
+            </div>
+          </div>
+        </div>
+      </div>
+    </div>
+
+    <!-- 空状态 -->
+    <div
+      v-else
+      class="p-8 text-center text-muted-foreground"
+    >
+      <Tag class="w-12 h-12 mx-auto mb-3 opacity-50" />
+      <p class="text-sm">
+        暂无模型映射
+      </p>
+      <p class="text-xs mt-1">
+        点击上方"添加映射"按钮为模型创建名称映射
+      </p>
+    </div>
+  </Card>
+
+  <!-- 添加/编辑映射对话框 -->
+  <ModelMappingDialog
+    v-model:open="dialogOpen"
+    :provider-id="provider.id"
+    :provider-api-formats="providerApiFormats"
+    :models="models"
+    :editing-group="editingGroup"
+    @saved="onDialogSaved"
+  />
+
+  <!-- 删除确认对话框 -->
+  <AlertDialog
+    v-model="deleteConfirmOpen"
+    title="删除映射组"
+    :description="deleteConfirmDescription"
+    confirm-text="删除"
+    cancel-text="取消"
+    type="danger"
+    @confirm="confirmDelete"
+    @cancel="deleteConfirmOpen = false"
+  />
+</template>
+
+<script setup lang="ts">
+import { ref, computed, onMounted, watch } from 'vue'
+import { Tag, Plus, Edit, Trash2, ChevronRight } from 'lucide-vue-next'
+import { Card, Button, Badge } from '@/components/ui'
+import AlertDialog from '@/components/common/AlertDialog.vue'
+import ModelMappingDialog, { type AliasGroup } from '../ModelMappingDialog.vue'
+import { useToast } from '@/composables/useToast'
+import {
+  getProviderModels,
+  API_FORMAT_LABELS,
+  type Model,
+  type ProviderModelAlias
+} from '@/api/endpoints'
+import { updateModel } from '@/api/endpoints/models'
+
+const props = defineProps<{
+  provider: any
+}>()
+
+const emit = defineEmits<{
+  'refresh': []
+}>()
+
+const { error: showError, success: showSuccess } = useToast()
+
+// 状态
+const loading = ref(false)
+const models = ref<Model[]>([])
+const dialogOpen = ref(false)
+const deleteConfirmOpen = ref(false)
+const editingGroup = ref<AliasGroup | null>(null)
+const deletingGroup = ref<AliasGroup | null>(null)
+
+// 列表展开状态
+const expandedAliasGroups = ref<Set<string>>(new Set())
+
+// 获取 Provider 支持的 API 格式
+const providerApiFormats = computed(() => {
+  const formats = props.provider?.api_formats
+  if (Array.isArray(formats) && formats.length > 0) {
+    const order = Object.keys(API_FORMAT_LABELS)
+    return [...formats].sort((a, b) => order.indexOf(a) - order.indexOf(b))
+  }
+  return []
+})
+
+// 生成作用域唯一键
+function getApiFormatsKey(formats: string[] | undefined): string {
+  if (!formats || formats.length === 0) return ''
+  return [...formats].sort().join(',')
+}
+
+// 按"模型+作用域"分组的映射列表
+const aliasGroups = computed<AliasGroup[]>(() => {
+  const groups: AliasGroup[] = []
+  const groupMap = new Map<string, AliasGroup>()
+
+  for (const model of models.value) {
+    if (!model.provider_model_mappings || !Array.isArray(model.provider_model_mappings)) continue
+
+    for (const alias of model.provider_model_mappings) {
+      const apiFormatsKey = getApiFormatsKey(alias.api_formats)
+      const groupKey = `${model.id}|${apiFormatsKey}`
+
+      if (!groupMap.has(groupKey)) {
+        const group: AliasGroup = {
+          model,
+          apiFormatsKey,
+          apiFormats: alias.api_formats || [],
+          aliases: []
+        }
+        groupMap.set(groupKey, group)
+        groups.push(group)
+      }
+      groupMap.get(groupKey)!.aliases.push(alias)
+    }
+  }
+
+  for (const group of groups) {
+    group.aliases.sort((a, b) => a.priority - b.priority)
+  }
+
+  return groups.sort((a, b) => {
+    const nameA = (a.model.global_model_display_name || a.model.provider_model_name || '').toLowerCase()
+    const nameB = (b.model.global_model_display_name || b.model.provider_model_name || '').toLowerCase()
+    if (nameA !== nameB) return nameA.localeCompare(nameB)
+    return a.apiFormatsKey.localeCompare(b.apiFormatsKey)
+  })
+})
+
+// 加载模型
+async function loadModels() {
+  try {
+    loading.value = true
+    models.value = await getProviderModels(props.provider.id)
+  } catch (err: any) {
+    showError(err.response?.data?.detail || '加载失败', '错误')
+  } finally {
+    loading.value = false
+  }
+}
+
+// 删除确认描述
+const deleteConfirmDescription = computed(() => {
+  if (!deletingGroup.value) return ''
+  const { model, aliases, apiFormats } = deletingGroup.value
+  const modelName = model.global_model_display_name || model.provider_model_name
+  const scopeText = apiFormats.length === 0 ? '全部' : apiFormats.map(f => API_FORMAT_LABELS[f] || f).join(', ')
+  const aliasNames = aliases.map(a => a.name).join(', ')
+  return `确定要删除模型「${modelName}」在作用域「${scopeText}」下的 ${aliases.length} 个映射吗？\n\n映射名称：${aliasNames}`
+})
+
+// 切换映射组展开状态
+function toggleAliasGroupExpand(groupKey: string) {
+  if (expandedAliasGroups.value.has(groupKey)) {
+    expandedAliasGroups.value.delete(groupKey)
+  } else {
+    expandedAliasGroups.value.add(groupKey)
+  }
+}
+
+// 打开添加对话框
+function openAddDialog() {
+  editingGroup.value = null
+  dialogOpen.value = true
+}
+
+// 编辑分组
+function editGroup(group: AliasGroup) {
+  editingGroup.value = group
+  dialogOpen.value = true
+}
+
+// 删除分组
+function deleteGroup(group: AliasGroup) {
+  deletingGroup.value = group
+  deleteConfirmOpen.value = true
+}
+
+// 确认删除
+async function confirmDelete() {
+  if (!deletingGroup.value) return
+
+  const { model, aliases, apiFormatsKey } = deletingGroup.value
+
+  try {
+    const currentAliases = model.provider_model_mappings || []
+    const aliasNamesToRemove = new Set(aliases.map(a => a.name))
+    const newAliases = currentAliases.filter((a: ProviderModelAlias) => {
+      const currentKey = getApiFormatsKey(a.api_formats)
+      return !(currentKey === apiFormatsKey && aliasNamesToRemove.has(a.name))
+    })
+
+    await updateModel(props.provider.id, model.id, {
+      provider_model_mappings: newAliases.length > 0 ? newAliases : null
+    })
+
+    showSuccess('映射组已删除')
+    deleteConfirmOpen.value = false
+    deletingGroup.value = null
+    await loadModels()
+    emit('refresh')
+  } catch (err: any) {
+    showError(err.response?.data?.detail || '删除失败', '错误')
+  }
+}
+
+// 对话框保存后回调
+async function onDialogSaved() {
+  await loadModels()
+  emit('refresh')
+}
+
+// 监听 provider 变化
+watch(() => props.provider?.id, (newId) => {
+  if (newId) {
+    loadModels()
+  }
+}, { immediate: true })
+
+onMounted(() => {
+  if (props.provider?.id) {
+    loadModels()
+  }
+})
+</script>

frontend/src/features/providers/composables/useUpstreamModelsCache.ts

@@ -0,0 +1,112 @@
+/**
+ * 上游模型缓存 - 共享缓存，避免重复请求
+ */
+import { ref } from 'vue'
+import { adminApi } from '@/api/admin'
+import type { UpstreamModel } from '@/api/endpoints/types'
+
+// 扩展类型，包含可能的额外字段
+export type { UpstreamModel }
+
+interface CacheEntry {
+  models: UpstreamModel[]
+  timestamp: number
+}
+
+type FetchResult = { models: UpstreamModel[]; error?: string }
+
+// 全局缓存（模块级别，所有组件共享）
+const cache = new Map<string, CacheEntry>()
+const CACHE_TTL = 5 * 60 * 1000 // 5分钟
+
+// 进行中的请求（用于去重并发请求）
+const pendingRequests = new Map<string, Promise<FetchResult>>()
+
+// 请求状态
+const loadingMap = ref<Map<string, boolean>>(new Map())
+
+export function useUpstreamModelsCache() {
+  /**
+   * 获取上游模型列表
+   * @param providerId 提供商ID
+   * @param forceRefresh 是否强制刷新
+   * @returns 模型列表或 null（如果请求失败）
+   */
+  async function fetchModels(
+    providerId: string,
+    forceRefresh = false
+  ): Promise<FetchResult> {
+    // 检查缓存
+    if (!forceRefresh) {
+      const cached = cache.get(providerId)
+      if (cached && Date.now() - cached.timestamp < CACHE_TTL) {
+        return { models: cached.models }
+      }
+    }
+
+    // 检查是否有进行中的请求（非强制刷新时复用）
+    if (!forceRefresh && pendingRequests.has(providerId)) {
+      return pendingRequests.get(providerId)!
+    }
+
+    // 创建新请求
+    const requestPromise = (async (): Promise<FetchResult> => {
+      try {
+        loadingMap.value.set(providerId, true)
+        const response = await adminApi.queryProviderModels(providerId)
+
+        if (response.success && response.data?.models) {
+          // 存入缓存
+          cache.set(providerId, {
+            models: response.data.models,
+            timestamp: Date.now()
+          })
+          return { models: response.data.models }
+        } else {
+          return { models: [], error: response.data?.error || '获取上游模型失败' }
+        }
+      } catch (err: any) {
+        return { models: [], error: err.response?.data?.detail || '获取上游模型失败' }
+      } finally {
+        loadingMap.value.set(providerId, false)
+        pendingRequests.delete(providerId)
+      }
+    })()
+
+    pendingRequests.set(providerId, requestPromise)
+    return requestPromise
+  }
+
+  /**
+   * 获取缓存的模型（不发起请求）
+   */
+  function getCachedModels(providerId: string): UpstreamModel[] | null {
+    const cached = cache.get(providerId)
+    if (cached && Date.now() - cached.timestamp < CACHE_TTL) {
+      return cached.models
+    }
+    return null
+  }
+
+  /**
+   * 清除指定提供商的缓存
+   */
+  function clearCache(providerId: string) {
+    cache.delete(providerId)
+  }
+
+  /**
+   * 检查是否正在加载
+   */
+  function isLoading(providerId: string): boolean {
+    return loadingMap.value.get(providerId) || false
+  }
+
+  return {
+    fetchModels,
+    getCachedModels,
+    clearCache,
+    isLoading,
+    loadingMap
+  }
+}

frontend/src/features/usage/components/HorizontalRequestTimeline.vue

@@ -479,10 +479,25 @@ const groupedTimeline = computed<NodeGroup[]>(() => {
   return groups
 })
 
-// 计算链路总耗时（从第一个节点开始到最后一个节点结束）
+// 计算链路总耗时（使用成功候选的 latency_ms 字段）
+// 优先使用 latency_ms，因为它与 Usage.response_time_ms 使用相同的时间基准
+// 避免 finished_at - started_at 带来的额外延迟（数据库操作时间）
 const totalTraceLatency = computed(() => {
   if (!timeline.value || timeline.value.length === 0) return 0
 
+  // 查找成功的候选，使用其 latency_ms
+  const successCandidate = timeline.value.find(c => c.status === 'success')
+  if (successCandidate?.latency_ms != null) {
+    return successCandidate.latency_ms
+  }
+
+  // 如果没有成功的候选，查找失败但有 latency_ms 的候选
+  const failedWithLatency = timeline.value.find(c => c.status === 'failed' && c.latency_ms != null)
+  if (failedWithLatency?.latency_ms != null) {
+    return failedWithLatency.latency_ms
+  }
+
+  // 回退：使用 finished_at - started_at 计算
   let earliestStart: number | null = null
   let latestEnd: number | null = null
 

frontend/src/features/usage/components/RequestDetailDrawer.vue

@@ -25,7 +25,7 @@
                 </h3>
                 <div class="flex items-center gap-1 text-sm font-mono text-muted-foreground bg-muted px-2 py-0.5 rounded">
                   <span>{{ detail?.model || '-' }}</span>
-                  <template v-if="detail?.target_model">
+                  <template v-if="detail?.target_model && detail.target_model !== detail.model">
                     <svg
                       xmlns="http://www.w3.org/2000/svg"
                       viewBox="0 0 20 20"
@@ -472,6 +472,7 @@
 <script setup lang="ts">
 import { ref, watch, computed } from 'vue'
 import Button from '@/components/ui/button.vue'
+import { useEscapeKey } from '@/composables/useEscapeKey'
 import Card from '@/components/ui/card.vue'
 import Badge from '@/components/ui/badge.vue'
 import Separator from '@/components/ui/separator.vue'
@@ -897,6 +898,16 @@ const providerHeadersWithDiff = computed(() => {
 
   return result
 })
+
+// 添加 ESC 键监听
+useEscapeKey(() => {
+  if (props.isOpen) {
+    handleClose()
+  }
+}, {
+  disableOnInput: true,
+  once: false
+})
 </script>
 
 <style scoped>

frontend/src/features/usage/components/UsageRecordsTable.vue

@@ -136,11 +136,20 @@
       <!-- 分隔线 -->
       <div class="hidden sm:block h-4 w-px bg-border" />
 
-      <!-- 刷新按钮 -->
-      <RefreshButton
-        :loading="loading"
-        @click="$emit('refresh')"
-      />
+      <!-- 自动刷新按钮 -->
+      <Button
+        variant="ghost"
+        size="icon"
+        class="h-8 w-8"
+        :class="autoRefresh ? 'text-primary' : ''"
+        :title="autoRefresh ? '点击关闭自动刷新' : '点击开启自动刷新（每10秒刷新）'"
+        @click="$emit('update:autoRefresh', !autoRefresh)"
+      >
+        <RefreshCcw
+          class="w-3.5 h-3.5"
+          :class="autoRefresh ? 'animate-spin' : ''"
+        />
+      </Button>
     </template>
 
     <Table>
@@ -177,8 +186,9 @@
             费用
           </TableHead>
           <TableHead class="h-12 font-semibold w-[70px] text-right">
-            <div class="inline-block max-w-[2rem] leading-tight">
-              响应时间
+            <div class="flex flex-col items-end text-xs gap-0.5">
+              <span>首字</span>
+              <span class="text-muted-foreground font-normal">总耗时</span>
             </div>
           </TableHead>
         </TableRow>
@@ -356,15 +366,28 @@
             </div>
           </TableCell>
           <TableCell class="text-right py-4 w-[70px]">
-            <span
+            <div
               v-if="record.status === 'pending' || record.status === 'streaming'"
-              class="text-primary tabular-nums"
+              class="flex flex-col items-end text-xs gap-0.5"
             >
-              {{ getElapsedTime(record) }}
-            </span>
-            <span v-else-if="record.response_time_ms">
-              {{ (record.response_time_ms / 1000).toFixed(2) }}s
-            </span>
+              <span class="text-primary tabular-nums">
+                {{ getElapsedTime(record) }}
+              </span>
+            </div>
+            <div
+              v-else-if="record.response_time_ms != null"
+              class="flex flex-col items-end text-xs gap-0.5"
+            >
+              <span
+                v-if="record.first_byte_time_ms != null"
+                class="tabular-nums"
+              >{{ (record.first_byte_time_ms / 1000).toFixed(2) }}s</span>
+              <span
+                v-else
+                class="text-muted-foreground"
+              >-</span>
+              <span class="text-muted-foreground tabular-nums">{{ (record.response_time_ms / 1000).toFixed(2) }}s</span>
+            </div>
             <span
               v-else
               class="text-muted-foreground"
@@ -394,6 +417,7 @@ import { ref, computed, onUnmounted, watch } from 'vue'
 import {
   TableCard,
   Badge,
+  Button,
   Select,
   SelectTrigger,
   SelectValue,
@@ -406,8 +430,8 @@ import {
   TableHead,
   TableCell,
   Pagination,
-  RefreshButton,
 } from '@/components/ui'
+import { RefreshCcw } from 'lucide-vue-next'
 import { formatTokens, formatCurrency } from '@/utils/format'
 import { formatDateTime } from '../composables'
 import { useRowClick } from '@/composables/useRowClick'
@@ -439,6 +463,8 @@ const props = defineProps<{
   pageSize: number
   totalRecords: number
   pageSizeOptions: number[]
+  // 自动刷新
+  autoRefresh: boolean
 }>()
 
 const emit = defineEmits<{
@@ -449,6 +475,7 @@ const emit = defineEmits<{
   'update:filterStatus': [value: string]
   'update:currentPage': [value: number]
   'update:pageSize': [value: number]
+  'update:autoRefresh': [value: boolean]
   'refresh': []
   'showDetail': [id: string]
 }>()
@@ -543,13 +570,14 @@ function formatApiFormat(format: string): string {
 }
 
 // 获取实际使用的模型（优先 target_model，其次 model_version）
+// 只有当实际模型与请求模型不同时才返回，用于显示映射箭头
 function getActualModel(record: UsageRecord): string | null {
   // 优先显示模型映射
-  if (record.target_model) {
+  if (record.target_model && record.target_model !== record.model) {
     return record.target_model
   }
   // 其次显示 Provider 返回的实际版本（如 Gemini 的 modelVersion）
-  if (record.request_metadata?.model_version) {
+  if (record.request_metadata?.model_version && record.request_metadata.model_version !== record.model) {
     return record.request_metadata.model_version
   }
   return null

frontend/src/features/usage/types.ts

@@ -78,6 +78,7 @@ export interface UsageRecord {
   cost: number
   actual_cost?: number
   response_time_ms?: number
+  first_byte_time_ms?: number  // 首字时间 (TTFB)
   is_stream: boolean
   status_code?: number
   error_message?: string

frontend/src/layouts/MainLayout.vue

@@ -313,7 +313,6 @@ import {
   Gauge,
   Layers,
   FolderTree,
-  Tag,
   Box,
   LogOut,
   SunMoon,
@@ -411,7 +410,6 @@ const navigation = computed(() => {
         { name: '用户管理', href: '/admin/users', icon: Users },
         { name: '提供商', href: '/admin/providers', icon: FolderTree },
         { name: '模型管理', href: '/admin/models', icon: Layers },
-        { name: '别名映射', href: '/admin/aliases', icon: Tag },
         { name: '独立密钥', href: '/admin/keys', icon: Key },
         { name: '使用记录', href: '/admin/usage', icon: BarChart3 },
       ]

frontend/src/mocks/data.ts

@@ -611,41 +611,42 @@ export const MOCK_GLOBAL_MODELS: GlobalModelResponse[] = [
     id: 'gm-001',
     name: 'claude-haiku-4-5-20251001',
     display_name: 'claude-haiku-4-5',
-    description: 'Anthropic 最快速的 Claude 4 系列模型',
     is_active: true,
     default_tiered_pricing: {
       tiers: [{ up_to: null, input_price_per_1m: 1.00, output_price_per_1m: 5.00, cache_creation_price_per_1m: 1.25, cache_read_price_per_1m: 0.1 }]
     },
-    default_supports_vision: true,
-    default_supports_function_calling: true,
-    default_supports_streaming: true,
-    default_supports_extended_thinking: true,
+    config: {
+      streaming: true,
+      vision: true,
+      function_calling: true,
+      extended_thinking: true,
+      description: 'Anthropic 最快速的 Claude 4 系列模型'
+    },
     provider_count: 3,
-    alias_count: 2,
     created_at: '2024-01-01T00:00:00Z'
   },
   {
     id: 'gm-002',
     name: 'claude-opus-4-5-20251101',
     display_name: 'claude-opus-4-5',
-    description: 'Anthropic 最强大的模型',
     is_active: true,
     default_tiered_pricing: {
       tiers: [{ up_to: null, input_price_per_1m: 5.00, output_price_per_1m: 25.00, cache_creation_price_per_1m: 6.25, cache_read_price_per_1m: 0.5 }]
     },
-    default_supports_vision: true,
-    default_supports_function_calling: true,
-    default_supports_streaming: true,
-    default_supports_extended_thinking: true,
+    config: {
+      streaming: true,
+      vision: true,
+      function_calling: true,
+      extended_thinking: true,
+      description: 'Anthropic 最强大的模型'
+    },
     provider_count: 2,
-    alias_count: 1,
     created_at: '2024-01-01T00:00:00Z'
   },
   {
     id: 'gm-003',
     name: 'claude-sonnet-4-5-20250929',
     display_name: 'claude-sonnet-4-5',
-    description: 'Anthropic 平衡型模型，支持 1h 缓存和 CLI 1M 上下文',
     is_active: true,
     default_tiered_pricing: {
       tiers: [
@@ -677,116 +678,124 @@ export const MOCK_GLOBAL_MODELS: GlobalModelResponse[] = [
         }
       ]
     },
-    default_supports_vision: true,
-    default_supports_function_calling: true,
-    default_supports_streaming: true,
-    default_supports_extended_thinking: true,
+    config: {
+      streaming: true,
+      vision: true,
+      function_calling: true,
+      extended_thinking: true,
+      description: 'Anthropic 平衡型模型，支持 1h 缓存和 CLI 1M 上下文'
+    },
     supported_capabilities: ['cache_1h', 'cli_1m'],
     provider_count: 3,
-    alias_count: 2,
     created_at: '2024-01-01T00:00:00Z'
   },
   {
     id: 'gm-004',
     name: 'gemini-3-pro-image-preview',
     display_name: 'gemini-3-pro-image-preview',
-    description: 'Google Gemini 3 Pro 图像生成预览版',
     is_active: true,
     default_price_per_request: 0.300,
     default_tiered_pricing: {
       tiers: []
     },
-    default_supports_vision: true,
-    default_supports_function_calling: false,
-    default_supports_streaming: true,
-    default_supports_image_generation: true,
+    config: {
+      streaming: true,
+      vision: true,
+      function_calling: false,
+      image_generation: true,
+      description: 'Google Gemini 3 Pro 图像生成预览版'
+    },
     provider_count: 1,
-    alias_count: 0,
     created_at: '2024-01-01T00:00:00Z'
   },
   {
     id: 'gm-005',
     name: 'gemini-3-pro-preview',
     display_name: 'gemini-3-pro-preview',
-    description: 'Google Gemini 3 Pro 预览版',
     is_active: true,
     default_tiered_pricing: {
       tiers: [{ up_to: null, input_price_per_1m: 2.00, output_price_per_1m: 12.00 }]
     },
-    default_supports_vision: true,
-    default_supports_function_calling: true,
-    default_supports_streaming: true,
-    default_supports_extended_thinking: true,
+    config: {
+      streaming: true,
+      vision: true,
+      function_calling: true,
+      extended_thinking: true,
+      description: 'Google Gemini 3 Pro 预览版'
+    },
     provider_count: 1,
-    alias_count: 0,
     created_at: '2024-01-01T00:00:00Z'
   },
   {
     id: 'gm-006',
     name: 'gpt-5.1',
     display_name: 'gpt-5.1',
-    description: 'OpenAI GPT-5.1 模型',
     is_active: true,
     default_tiered_pricing: {
       tiers: [{ up_to: null, input_price_per_1m: 1.25, output_price_per_1m: 10.00 }]
     },
-    default_supports_vision: true,
-    default_supports_function_calling: true,
-    default_supports_streaming: true,
-    default_supports_extended_thinking: true,
+    config: {
+      streaming: true,
+      vision: true,
+      function_calling: true,
+      extended_thinking: true,
+      description: 'OpenAI GPT-5.1 模型'
+    },
     provider_count: 2,
-    alias_count: 1,
     created_at: '2024-01-01T00:00:00Z'
   },
   {
     id: 'gm-007',
     name: 'gpt-5.1-codex',
     display_name: 'gpt-5.1-codex',
-    description: 'OpenAI GPT-5.1 Codex 代码专用模型',
     is_active: true,
     default_tiered_pricing: {
       tiers: [{ up_to: null, input_price_per_1m: 1.25, output_price_per_1m: 10.00 }]
     },
-    default_supports_vision: true,
-    default_supports_function_calling: true,
-    default_supports_streaming: true,
-    default_supports_extended_thinking: true,
+    config: {
+      streaming: true,
+      vision: true,
+      function_calling: true,
+      extended_thinking: true,
+      description: 'OpenAI GPT-5.1 Codex 代码专用模型'
+    },
     provider_count: 2,
-    alias_count: 0,
     created_at: '2024-01-01T00:00:00Z'
   },
   {
     id: 'gm-008',
     name: 'gpt-5.1-codex-max',
     display_name: 'gpt-5.1-codex-max',
-    description: 'OpenAI GPT-5.1 Codex Max 代码专用增强版',
     is_active: true,
     default_tiered_pricing: {
       tiers: [{ up_to: null, input_price_per_1m: 1.25, output_price_per_1m: 10.00 }]
     },
-    default_supports_vision: true,
-    default_supports_function_calling: true,
-    default_supports_streaming: true,
-    default_supports_extended_thinking: true,
+    config: {
+      streaming: true,
+      vision: true,
+      function_calling: true,
+      extended_thinking: true,
+      description: 'OpenAI GPT-5.1 Codex Max 代码专用增强版'
+    },
     provider_count: 2,
-    alias_count: 0,
     created_at: '2024-01-01T00:00:00Z'
   },
   {
     id: 'gm-009',
     name: 'gpt-5.1-codex-mini',
     display_name: 'gpt-5.1-codex-mini',
-    description: 'OpenAI GPT-5.1 Codex Mini 轻量代码模型',
     is_active: true,
     default_tiered_pricing: {
       tiers: [{ up_to: null, input_price_per_1m: 1.25, output_price_per_1m: 10.00 }]
     },
-    default_supports_vision: true,
-    default_supports_function_calling: true,
-    default_supports_streaming: true,
-    default_supports_extended_thinking: true,
+    config: {
+      streaming: true,
+      vision: true,
+      function_calling: true,
+      extended_thinking: true,
+      description: 'OpenAI GPT-5.1 Codex Mini 轻量代码模型'
+    },
     provider_count: 2,
-    alias_count: 0,
     created_at: '2024-01-01T00:00:00Z'
   }
 ]

frontend/src/mocks/handler.ts

@@ -403,7 +403,7 @@ function getUsageRecords() {
   return cachedUsageRecords
 }
 
-// Mock 别名数据
+// Mock 映射数据
 const MOCK_ALIASES = [
   { id: 'alias-001', source_model: 'claude-4-sonnet', target_global_model_id: 'gm-001', target_global_model_name: 'claude-sonnet-4-20250514', target_global_model_display_name: 'Claude Sonnet 4', provider_id: null, provider_name: null, scope: 'global', mapping_type: 'alias', is_active: true, created_at: '2024-01-01T00:00:00Z', updated_at: '2024-01-01T00:00:00Z' },
   { id: 'alias-002', source_model: 'claude-4-opus', target_global_model_id: 'gm-002', target_global_model_name: 'claude-opus-4-20250514', target_global_model_display_name: 'Claude Opus 4', provider_id: null, provider_name: null, scope: 'global', mapping_type: 'alias', is_active: true, created_at: '2024-01-01T00:00:00Z', updated_at: '2024-01-01T00:00:00Z' },
@@ -1000,17 +1000,11 @@ const mockHandlers: Record<string, (config: AxiosRequestConfig) => Promise<Axios
         id: m.id,
         name: m.name,
         display_name: m.display_name,
-        description: m.description,
-        icon_url: null,
         is_active: m.is_active,
         default_tiered_pricing: m.default_tiered_pricing,
-        default_price_per_request: null,
-        default_supports_vision: m.default_supports_vision,
-        default_supports_function_calling: m.default_supports_function_calling,
-        default_supports_streaming: m.default_supports_streaming,
-        default_supports_extended_thinking: m.default_supports_extended_thinking || false,
-        default_supports_image_generation: false,
-        supported_capabilities: null
+        default_price_per_request: m.default_price_per_request,
+        supported_capabilities: m.supported_capabilities,
+        config: m.config
       })),
       total: MOCK_GLOBAL_MODELS.length
     })
@@ -1688,7 +1682,7 @@ registerDynamicRoute('GET', '/api/admin/models/mappings/:mappingId', async (_con
   requireAdmin()
   const alias = MOCK_ALIASES.find(a => a.id === params.mappingId)
   if (!alias) {
-    throw { response: createMockResponse({ detail: '别名不存在' }, 404) }
+    throw { response: createMockResponse({ detail: '映射不存在' }, 404) }
   }
   return createMockResponse(alias)
 })
@@ -1699,7 +1693,7 @@ registerDynamicRoute('PATCH', '/api/admin/models/mappings/:mappingId', async (co
   requireAdmin()
   const alias = MOCK_ALIASES.find(a => a.id === params.mappingId)
   if (!alias) {
-    throw { response: createMockResponse({ detail: '别名不存在' }, 404) }
+    throw { response: createMockResponse({ detail: '映射不存在' }, 404) }
   }
   const body = JSON.parse(config.data || '{}')
   return createMockResponse({ ...alias, ...body, updated_at: new Date().toISOString() })
@@ -1711,7 +1705,7 @@ registerDynamicRoute('DELETE', '/api/admin/models/mappings/:mappingId', async (_
   requireAdmin()
   const alias = MOCK_ALIASES.find(a => a.id === params.mappingId)
   if (!alias) {
-    throw { response: createMockResponse({ detail: '别名不存在' }, 404) }
+    throw { response: createMockResponse({ detail: '映射不存在' }, 404) }
   }
   return createMockResponse({ message: '删除成功（演示模式）' })
 })

frontend/src/router/index.ts

@@ -91,11 +91,6 @@ const routes: RouteRecordRaw[] = [
         name: 'ModelManagement',
         component: () => importWithRetry(() => import('@/views/admin/ModelManagement.vue'))
       },
-      {
-        path: 'aliases',
-        name: 'AliasManagement',
-        component: () => importWithRetry(() => import('@/views/admin/AliasManagement.vue'))
-      },
       {
         path: 'health-monitor',
         name: 'HealthMonitor',

frontend/src/style.css

@@ -1169,4 +1169,26 @@ body[theme-mode='dark'] .literary-annotation {
   .scrollbar-hide::-webkit-scrollbar {
     display: none;
   }
-}
+
+  .scrollbar-thin {
+    scrollbar-width: thin;
+    scrollbar-color: hsl(var(--border)) transparent;
+  }
+
+  .scrollbar-thin::-webkit-scrollbar {
+    width: 6px;
+  }
+
+  .scrollbar-thin::-webkit-scrollbar-track {
+    background: transparent;
+  }
+
+  .scrollbar-thin::-webkit-scrollbar-thumb {
+    background-color: hsl(var(--border));
+    border-radius: 3px;
+  }
+
+  .scrollbar-thin::-webkit-scrollbar-thumb:hover {
+    background-color: hsl(var(--muted-foreground) / 0.5);
+  }
+}

frontend/src/views/admin/AliasManagement.vue

@@ -1,500 +0,0 @@
-<template>
-  <div class="flex flex-col">
-    <Card class="overflow-hidden">
-      <!-- 搜索和过滤区域 -->
-      <div class="px-4 sm:px-6 py-3 sm:py-3.5 border-b border-border/60">
-        <div class="flex flex-col sm:flex-row sm:items-center sm:justify-between gap-3 sm:gap-4">
-          <h3 class="text-sm sm:text-base font-semibold shrink-0">
-            别名管理
-          </h3>
-          <div class="flex flex-wrap items-center gap-2">
-            <!-- 搜索框 -->
-            <div class="relative">
-              <Search class="absolute left-2.5 top-1/2 -translate-y-1/2 h-3.5 w-3.5 text-muted-foreground z-10 pointer-events-none" />
-              <Input
-                id="alias-search"
-                v-model="aliasesSearch"
-                placeholder="搜索别名或关联模型"
-                class="w-32 sm:w-44 pl-8 pr-3 h-8 text-sm border-border/60 focus-visible:ring-1"
-              />
-            </div>
-
-            <div class="hidden sm:block h-4 w-px bg-border" />
-
-            <!-- 提供商过滤器 -->
-            <Select
-              v-model:open="aliasProviderSelectOpen"
-              :model-value="aliasProviderFilter"
-              @update:model-value="aliasProviderFilter = $event"
-            >
-              <SelectTrigger class="w-28 sm:w-40 h-8 text-xs border-border/60">
-                <SelectValue />
-              </SelectTrigger>
-              <SelectContent>
-                <SelectItem value="all">
-                  全部别名
-                </SelectItem>
-                <SelectItem value="global">
-                  仅全局别名
-                </SelectItem>
-                <SelectItem
-                  v-for="provider in providers"
-                  :key="provider.id"
-                  :value="provider.id"
-                >
-                  {{ provider.display_name }}
-                </SelectItem>
-              </SelectContent>
-            </Select>
-
-            <div class="hidden sm:block h-4 w-px bg-border" />
-
-            <!-- 操作按钮 -->
-            <Button
-              variant="ghost"
-              size="icon"
-              class="h-8 w-8"
-              title="新建别名"
-              @click="openCreateAliasDialog"
-            >
-              <Plus class="w-3.5 h-3.5" />
-            </Button>
-            <RefreshButton
-              :loading="loadingAliases"
-              @click="loadAliases"
-            />
-          </div>
-        </div>
-      </div>
-      <div
-        v-if="loadingAliases"
-        class="flex items-center justify-center py-12"
-      >
-        <Loader2 class="w-10 h-10 animate-spin text-primary" />
-      </div>
-      <div v-else>
-        <Table class="hidden xl:table text-sm">
-          <TableHeader>
-            <TableRow>
-              <TableHead class="w-[200px]">
-                别名
-              </TableHead>
-              <TableHead class="w-[280px]">
-                关联模型
-              </TableHead>
-              <TableHead class="w-[70px] text-center">
-                类型
-              </TableHead>
-              <TableHead class="w-[100px] text-center">
-                作用域
-              </TableHead>
-              <TableHead class="w-[70px] text-center">
-                状态
-              </TableHead>
-              <TableHead class="w-[100px] text-center">
-                操作
-              </TableHead>
-            </TableRow>
-          </TableHeader>
-          <TableBody>
-            <TableRow v-if="filteredAliases.length === 0">
-              <TableCell
-                colspan="6"
-                class="text-center py-8 text-muted-foreground"
-              >
-                {{ aliasProviderFilter === 'global' ? '暂无全局别名' : '暂无别名' }}
-              </TableCell>
-            </TableRow>
-            <TableRow
-              v-for="alias in paginatedAliases"
-              :key="alias.id"
-            >
-              <TableCell>
-                <span class="font-mono font-medium">{{ alias.alias }}</span>
-              </TableCell>
-              <TableCell>
-                <div class="flex flex-col gap-0.5">
-                  <span class="font-medium">{{ alias.global_model_display_name || alias.global_model_name }}</span>
-                  <span class="text-xs text-muted-foreground font-mono">{{ alias.global_model_name }}</span>
-                </div>
-              </TableCell>
-              <TableCell class="text-center">
-                <Badge
-                  variant="secondary"
-                  class="text-xs"
-                >
-                  {{ alias.mapping_type === 'mapping' ? '映射' : '别名' }}
-                </Badge>
-              </TableCell>
-              <TableCell class="text-center">
-                <Badge
-                  v-if="alias.provider_id"
-                  variant="outline"
-                  class="text-xs"
-                >
-                  {{ alias.provider_name || 'Provider 特定' }}
-                </Badge>
-                <Badge
-                  v-else
-                  variant="default"
-                  class="text-xs"
-                >
-                  全局
-                </Badge>
-              </TableCell>
-              <TableCell class="text-center">
-                <Badge
-                  :variant="alias.is_active ? 'default' : 'secondary'"
-                  class="text-xs"
-                >
-                  {{ alias.is_active ? '活跃' : '停用' }}
-                </Badge>
-              </TableCell>
-              <TableCell class="text-center">
-                <div class="flex items-center justify-center gap-1">
-                  <Button
-                    variant="ghost"
-                    size="icon"
-                    class="h-7 w-7"
-                    title="编辑别名"
-                    @click="openEditAliasDialog(alias)"
-                  >
-                    <Edit class="w-3.5 h-3.5" />
-                  </Button>
-                  <Button
-                    variant="ghost"
-                    size="icon"
-                    class="h-7 w-7"
-                    :title="alias.is_active ? '停用别名' : '启用别名'"
-                    @click="toggleAliasStatus(alias)"
-                  >
-                    <Power class="w-3.5 h-3.5" />
-                  </Button>
-                  <Button
-                    variant="ghost"
-                    size="icon"
-                    class="h-7 w-7"
-                    title="删除别名"
-                    @click="confirmDeleteAlias(alias)"
-                  >
-                    <Trash2 class="w-3.5 h-3.5" />
-                  </Button>
-                </div>
-              </TableCell>
-            </TableRow>
-          </TableBody>
-        </Table>
-
-        <!-- 移动端卡片列表 -->
-        <div
-          v-if="filteredAliases.length > 0"
-          class="xl:hidden divide-y divide-border/40"
-        >
-          <div
-            v-for="alias in paginatedAliases"
-            :key="alias.id"
-            class="p-4 space-y-2"
-          >
-            <div class="flex items-start justify-between gap-3">
-              <div class="flex-1 min-w-0">
-                <div class="flex items-center gap-2">
-                  <span class="font-mono font-medium truncate">{{ alias.alias }}</span>
-                  <Badge
-                    :variant="alias.is_active ? 'default' : 'secondary'"
-                    class="text-xs shrink-0"
-                  >
-                    {{ alias.is_active ? '活跃' : '停用' }}
-                  </Badge>
-                </div>
-                <div class="text-xs text-muted-foreground mt-1">
-                  <span class="font-medium">{{ alias.global_model_display_name || alias.global_model_name }}</span>
-                </div>
-              </div>
-              <div class="flex items-center gap-0.5 shrink-0">
-                <Button
-                  variant="ghost"
-                  size="icon"
-                  class="h-7 w-7"
-                  @click="openEditAliasDialog(alias)"
-                >
-                  <Edit class="w-3.5 h-3.5" />
-                </Button>
-                <Button
-                  variant="ghost"
-                  size="icon"
-                  class="h-7 w-7"
-                  @click="toggleAliasStatus(alias)"
-                >
-                  <Power class="w-3.5 h-3.5" />
-                </Button>
-                <Button
-                  variant="ghost"
-                  size="icon"
-                  class="h-7 w-7"
-                  @click="confirmDeleteAlias(alias)"
-                >
-                  <Trash2 class="w-3.5 h-3.5" />
-                </Button>
-              </div>
-            </div>
-            <div class="flex items-center gap-2">
-              <Badge
-                variant="secondary"
-                class="text-xs"
-              >
-                {{ alias.mapping_type === 'mapping' ? '映射' : '别名' }}
-              </Badge>
-              <Badge
-                v-if="alias.provider_id"
-                variant="outline"
-                class="text-xs"
-              >
-                {{ alias.provider_name || 'Provider 特定' }}
-              </Badge>
-              <Badge
-                v-else
-                variant="default"
-                class="text-xs"
-              >
-                全局
-              </Badge>
-            </div>
-          </div>
-        </div>
-
-        <!-- 分页 -->
-        <Pagination
-          v-if="!loadingAliases && filteredAliases.length > 0"
-          :current="aliasesCurrentPage"
-          :total="filteredAliases.length"
-          :page-size="aliasesPageSize"
-          @update:current="aliasesCurrentPage = $event"
-          @update:page-size="aliasesPageSize = $event"
-        />
-      </div>
-    </Card>
-
-    <!-- 创建/编辑别名对话框 -->
-    <AliasDialog
-      :open="createAliasDialogOpen"
-      :editing-alias="editingAlias"
-      :global-models="globalModels"
-      :providers="providers"
-      @update:open="handleAliasDialogUpdate"
-      @submit="handleAliasSubmit"
-    />
-  </div>
-</template>
-
-<script setup lang="ts">
-import { ref, computed, onMounted, watch } from 'vue'
-import {
-  Edit,
-  Loader2,
-  Plus,
-  Power,
-  Search,
-  Trash2
-} from 'lucide-vue-next'
-import {
-  Card,
-  Button,
-  Input,
-  Badge,
-  Table,
-  TableBody,
-  TableCell,
-  TableHead,
-  TableHeader,
-  TableRow,
-  Select,
-  SelectContent,
-  SelectItem,
-  SelectTrigger,
-  SelectValue,
-  RefreshButton,
-  Pagination
-} from '@/components/ui'
-import AliasDialog from '@/features/models/components/AliasDialog.vue'
-import { useToast } from '@/composables/useToast'
-import { useConfirm } from '@/composables/useConfirm'
-import {
-  getAliases,
-  createAlias,
-  updateAlias,
-  deleteAlias,
-  type ModelAlias,
-  type CreateModelAliasRequest,
-  type UpdateModelAliasRequest
-} from '@/api/endpoints/aliases'
-import { listGlobalModels, type GlobalModelResponse } from '@/api/global-models'
-import { getProvidersSummary } from '@/api/endpoints/providers'
-import { log } from '@/utils/logger'
-
-const { success, error: showError } = useToast()
-const { confirmDanger } = useConfirm()
-
-// 状态
-const loadingAliases = ref(false)
-const submitting = ref(false)
-const aliasesSearch = ref('')
-const aliasProviderFilter = ref<string>('all')
-const aliasProviderSelectOpen = ref(false)
-const createAliasDialogOpen = ref(false)
-const editingAliasId = ref<string | null>(null)
-
-// 数据
-const allAliases = ref<ModelAlias[]>([])
-const globalModels = ref<GlobalModelResponse[]>([])
-const providers = ref<any[]>([])
-
-// 分页
-const aliasesCurrentPage = ref(1)
-const aliasesPageSize = ref(20)
-
-// 编辑中的别名对象
-const editingAlias = computed(() => {
-  if (!editingAliasId.value) return null
-  return allAliases.value.find(a => a.id === editingAliasId.value) || null
-})
-
-// 筛选后的别名列表
-const filteredAliases = computed(() => {
-  let result = allAliases.value
-
-  // 按 Provider 筛选
-  if (aliasProviderFilter.value === 'global') {
-    result = result.filter(alias => !alias.provider_id)
-  } else if (aliasProviderFilter.value !== 'all') {
-    result = result.filter(alias => alias.provider_id === aliasProviderFilter.value)
-  }
-
-  // 按搜索关键词筛选
-  const keyword = aliasesSearch.value.trim().toLowerCase()
-  if (keyword) {
-    result = result.filter(alias =>
-      alias.alias.toLowerCase().includes(keyword) ||
-      alias.global_model_name?.toLowerCase().includes(keyword) ||
-      alias.global_model_display_name?.toLowerCase().includes(keyword)
-    )
-  }
-
-  return result
-})
-
-// 分页计算
-const paginatedAliases = computed(() => {
-  const start = (aliasesCurrentPage.value - 1) * aliasesPageSize.value
-  const end = start + aliasesPageSize.value
-  return filteredAliases.value.slice(start, end)
-})
-
-// 搜索或筛选变化时重置到第一页
-watch([aliasesSearch, aliasProviderFilter], () => {
-  aliasesCurrentPage.value = 1
-})
-
-async function loadAliases() {
-  loadingAliases.value = true
-  try {
-    allAliases.value = await getAliases({ limit: 1000 })
-  } catch (err: any) {
-    showError(err.response?.data?.detail || err.message, '加载别名失败')
-  } finally {
-    loadingAliases.value = false
-  }
-}
-
-async function loadGlobalModelsList() {
-  try {
-    const response = await listGlobalModels()
-    globalModels.value = response.models || []
-  } catch (err: any) {
-    log.error('加载模型失败:', err)
-  }
-}
-
-async function loadProviders() {
-  try {
-    providers.value = await getProvidersSummary()
-  } catch (err: any) {
-    showError(err.response?.data?.detail || err.message, '加载 Provider 列表失败')
-  }
-}
-
-function openCreateAliasDialog() {
-  editingAliasId.value = null
-  createAliasDialogOpen.value = true
-}
-
-function openEditAliasDialog(alias: ModelAlias) {
-  editingAliasId.value = alias.id
-  createAliasDialogOpen.value = true
-}
-
-function handleAliasDialogUpdate(value: boolean) {
-  createAliasDialogOpen.value = value
-  if (!value) {
-    editingAliasId.value = null
-  }
-}
-
-async function handleAliasSubmit(data: CreateModelAliasRequest | UpdateModelAliasRequest, isEdit: boolean) {
-  submitting.value = true
-  try {
-    if (isEdit && editingAliasId.value) {
-      await updateAlias(editingAliasId.value, data as UpdateModelAliasRequest)
-      success(data.mapping_type === 'mapping' ? '映射已更新' : '别名已更新')
-    } else {
-      await createAlias(data as CreateModelAliasRequest)
-      success(data.mapping_type === 'mapping' ? '映射已创建' : '别名已创建')
-    }
-    createAliasDialogOpen.value = false
-    editingAliasId.value = null
-    await loadAliases()
-  } catch (err: any) {
-    const detail = err.response?.data?.detail || err.message
-    let errorMessage = detail
-    if (detail === '映射已存在') {
-      errorMessage = '目标作用域已存在同名别名，请先删除冲突的映射或选择其他作用域'
-    }
-    showError(errorMessage, isEdit ? '更新失败' : '创建失败')
-  } finally {
-    submitting.value = false
-  }
-}
-
-async function confirmDeleteAlias(alias: ModelAlias) {
-  const confirmed = await confirmDanger(
-    `确定要删除别名 "${alias.alias}" 吗？`,
-    '删除别名'
-  )
-  if (!confirmed) return
-
-  try {
-    await deleteAlias(alias.id)
-    success('别名已删除')
-    await loadAliases()
-  } catch (err: any) {
-    showError(err.response?.data?.detail || err.message, '删除失败')
-  }
-}
-
-async function toggleAliasStatus(alias: ModelAlias) {
-  try {
-    await updateAlias(alias.id, { is_active: !alias.is_active })
-    alias.is_active = !alias.is_active
-    success(alias.is_active ? '别名已启用' : '别名已停用')
-  } catch (err: any) {
-    showError(err.response?.data?.detail || err.message, '操作失败')
-  }
-}
-
-onMounted(async () => {
-  await Promise.all([
-    loadAliases(),
-    loadGlobalModelsList(),
-    loadProviders()
-  ])
-})
-</script>

frontend/src/views/admin/ApiKeys.vue

@@ -751,15 +751,13 @@ const expiringSoonCount = computed(() => apiKeys.value.filter(key => isExpiringS
 const filteredApiKeys = computed(() => {
   let result = apiKeys.value
 
-  // 搜索筛选
+  // 搜索筛选（支持空格分隔的多关键词 AND 搜索）
   if (searchQuery.value) {
-    const query = searchQuery.value.toLowerCase()
-    result = result.filter(key =>
-      (key.name && key.name.toLowerCase().includes(query)) ||
-      (key.key_display && key.key_display.toLowerCase().includes(query)) ||
-      (key.username && key.username.toLowerCase().includes(query)) ||
-      (key.user_email && key.user_email.toLowerCase().includes(query))
-    )
+    const keywords = searchQuery.value.toLowerCase().split(/\s+/).filter(k => k.length > 0)
+    result = result.filter(key => {
+      const searchableText = `${key.name || ''} ${key.key_display || ''} ${key.username || ''} ${key.user_email || ''}`.toLowerCase()
+      return keywords.every(keyword => searchableText.includes(keyword))
+    })
   }
 
   // 状态筛选

frontend/src/views/admin/CacheMonitoring.vue

@@ -18,10 +18,10 @@ import SelectContent from '@/components/ui/select-content.vue'
 import SelectItem from '@/components/ui/select-item.vue'
 import SelectValue from '@/components/ui/select-value.vue'
 import ScatterChart from '@/components/charts/ScatterChart.vue'
-import { Trash2, Eraser, Search, X, BarChart3, ChevronDown, ChevronRight } from 'lucide-vue-next'
+import { Trash2, Eraser, Search, X, BarChart3, ChevronDown, ChevronRight, Database, ArrowRight } from 'lucide-vue-next'
 import { useToast } from '@/composables/useToast'
 import { useConfirm } from '@/composables/useConfirm'
-import { cacheApi, type CacheStats, type CacheConfig, type UserAffinity } from '@/api/cache'
+import { cacheApi, modelMappingCacheApi, type CacheStats, type CacheConfig, type UserAffinity, type ModelMappingCacheStats } from '@/api/cache'
 import type { TTLAnalysisUser } from '@/api/cache'
 import { formatNumber, formatTokens, formatCost, formatRemainingTime } from '@/utils/format'
 import {
@@ -47,6 +47,13 @@ const currentPage = ref(1)
 const pageSize = ref(20)
 const currentTime = ref(Math.floor(Date.now() / 1000))
 
+// ==================== 模型映射缓存 ====================
+
+const modelMappingStats = ref<ModelMappingCacheStats | null>(null)
+const modelMappingLoading = ref(false)
+const clearingModelMapping = ref(false)
+const clearingModelName = ref<string | null>(null)
+
 const { success: showSuccess, error: showError, info: showInfo } = useToast()
 const { confirm: showConfirm } = useConfirm()
 
@@ -135,32 +142,37 @@ async function resetAffinitySearch() {
   await fetchAffinityList()
 }
 
-async function clearUserCache(identifier: string, displayName?: string) {
-  const target = identifier?.trim()
-  if (!target) {
-    showError('无法识别标识符')
+async function clearSingleAffinity(item: UserAffinity) {
+  const affinityKey = item.affinity_key?.trim()
+  const endpointId = item.endpoint_id?.trim()
+  const modelId = item.global_model_id?.trim()
+  const apiFormat = item.api_format?.trim()
+
+  if (!affinityKey || !endpointId || !modelId || !apiFormat) {
+    showError('缓存记录信息不完整，无法删除')
     return
   }
 
-  const label = displayName || target
+  const label = item.user_api_key_name || affinityKey
+  const modelLabel = item.model_display_name || item.model_name || modelId
   const confirmed = await showConfirm({
     title: '确认清除',
-    message: `确定要清除 ${label} 的缓存吗？`,
+    message: `确定要清除 ${label} 在模型 ${modelLabel} 上的缓存亲和性吗？`,
     confirmText: '确认清除',
     variant: 'destructive'
   })
 
   if (!confirmed) return
 
-  clearingRowAffinityKey.value = target
+  clearingRowAffinityKey.value = affinityKey
   try {
-    await cacheApi.clearUserCache(target)
+    await cacheApi.clearSingleAffinity(affinityKey, endpointId, modelId, apiFormat)
     showSuccess('清除成功')
     await fetchCacheStats()
     await fetchAffinityList(tableKeyword.value.trim() || undefined)
   } catch (error) {
     showError('清除失败')
-    log.error('清除用户缓存失败', error)
+    log.error('清除单条缓存失败', error)
   } finally {
     clearingRowAffinityKey.value = null
   }
@@ -241,13 +253,107 @@ function stopCountdown() {
   }
 }
 
+// ==================== 模型映射缓存方法 ====================
+
+async function fetchModelMappingStats() {
+  modelMappingLoading.value = true
+  try {
+    modelMappingStats.value = await modelMappingCacheApi.getStats()
+  } catch (error) {
+    showError('获取模型映射缓存统计失败')
+    log.error('获取模型映射缓存统计失败', error)
+  } finally {
+    modelMappingLoading.value = false
+  }
+}
+
+async function clearAllModelMappingCache() {
+  const confirmed = await showConfirm({
+    title: '确认清除',
+    message: '确定要清除所有模型映射缓存吗？这会影响所有模型的名称解析。',
+    confirmText: '确认清除',
+    variant: 'destructive'
+  })
+
+  if (!confirmed) return
+
+  clearingModelMapping.value = true
+  try {
+    const result = await modelMappingCacheApi.clearAll()
+    showSuccess(`已清除 ${result.deleted_count} 个缓存键`)
+    await fetchModelMappingStats()
+  } catch (error) {
+    showError('清除模型映射缓存失败')
+    log.error('清除模型映射缓存失败', error)
+  } finally {
+    clearingModelMapping.value = false
+  }
+}
+
+async function clearModelMappingByName(modelName: string) {
+  clearingModelName.value = modelName
+  try {
+    await modelMappingCacheApi.clearByName(modelName)
+    showSuccess(`已清除 ${modelName} 的映射缓存`)
+    await fetchModelMappingStats()
+  } catch (error) {
+    showError('清除缓存失败')
+    log.error('清除模型映射缓存失败', error)
+  } finally {
+    clearingModelName.value = null
+  }
+}
+
+async function clearProviderModelMapping(providerId: string, globalModelId: string, displayName?: string) {
+  const confirmed = await showConfirm({
+    title: '确认清除',
+    message: `确定要清除 ${displayName || 'Provider 模型映射'} 的缓存吗？`,
+    confirmText: '确认清除',
+    variant: 'destructive'
+  })
+
+  if (!confirmed) return
+
+  try {
+    await modelMappingCacheApi.clearProviderModel(providerId, globalModelId)
+    showSuccess('已清除 Provider 模型映射缓存')
+    await fetchModelMappingStats()
+  } catch (error) {
+    showError('清除缓存失败')
+    log.error('清除 Provider 模型映射缓存失败', error)
+  }
+}
+
+function formatTTL(ttl: number | null): string {
+  if (ttl === null || ttl < 0) return '-'
+  if (ttl < 60) return `${ttl}s`
+  const minutes = Math.floor(ttl / 60)
+  const seconds = ttl % 60
+  if (seconds === 0) return `${minutes}m`
+  return `${minutes}m${seconds}s`
+}
+
+function getUnmappedStatusBadge(status: string): { variant: 'default' | 'secondary' | 'destructive' | 'outline', text: string } {
+  switch (status) {
+    case 'not_found':
+      return { variant: 'secondary', text: '未找到' }
+    case 'invalid':
+      return { variant: 'destructive', text: '无效' }
+    case 'error':
+      return { variant: 'destructive', text: '错误' }
+    default:
+      return { variant: 'outline', text: status }
+  }
+}
+
 // ==================== 刷新所有数据 ====================
 
 async function refreshData() {
   await Promise.all([
     fetchCacheStats(),
     fetchCacheConfig(),
-    fetchAffinityList()
+    fetchAffinityList(),
+    fetchModelMappingStats()
   ])
 }
 
@@ -272,6 +378,7 @@ onMounted(() => {
   fetchCacheStats()
   fetchCacheConfig()
   fetchAffinityList()
+  fetchModelMappingStats()
   startCountdown()
   refreshAnalysis()
 })
@@ -516,7 +623,7 @@ onBeforeUnmount(() => {
                 class="h-7 w-7 text-muted-foreground/70 hover:text-destructive"
                 :disabled="clearingRowAffinityKey === item.affinity_key"
                 title="清除缓存"
-                @click="clearUserCache(item.affinity_key, item.user_api_key_name || item.affinity_key)"
+                @click="clearSingleAffinity(item)"
               >
                 <Trash2 class="h-3.5 w-3.5" />
               </Button>
@@ -566,7 +673,7 @@ onBeforeUnmount(() => {
               variant="ghost"
               class="h-7 w-7 text-muted-foreground/70 hover:text-destructive shrink-0"
               :disabled="clearingRowAffinityKey === item.affinity_key"
-              @click="clearUserCache(item.affinity_key, item.user_api_key_name || item.affinity_key)"
+              @click="clearSingleAffinity(item)"
             >
               <Trash2 class="h-3.5 w-3.5" />
             </Button>
@@ -599,6 +706,344 @@ onBeforeUnmount(() => {
       />
     </Card>
 
+    <!-- 模型映射缓存管理 -->
+    <Card class="overflow-hidden">
+      <div class="px-4 sm:px-6 py-3 sm:py-3.5 border-b border-border/60">
+        <div class="flex flex-col sm:flex-row sm:items-center sm:justify-between gap-3 sm:gap-4">
+          <div class="flex items-center gap-3 shrink-0">
+            <Database class="h-5 w-5 text-muted-foreground hidden sm:block" />
+            <h3 class="text-sm sm:text-base font-semibold">
+              模型映射缓存
+            </h3>
+          </div>
+          <div class="flex flex-wrap items-center gap-2">
+            <Button
+              variant="ghost"
+              size="icon"
+              class="h-8 w-8 text-muted-foreground/70 hover:text-destructive"
+              title="清除全部映射缓存"
+              :disabled="clearingModelMapping || !modelMappingStats?.available"
+              @click="clearAllModelMappingCache"
+            >
+              <Eraser class="h-4 w-4" />
+            </Button>
+            <RefreshButton
+              :loading="modelMappingLoading"
+              @click="fetchModelMappingStats"
+            />
+          </div>
+        </div>
+      </div>
+
+      <!-- 映射缓存表格 -->
+      <Table
+        v-if="modelMappingStats?.available && modelMappingStats.mappings && modelMappingStats.mappings.length > 0"
+        class="hidden md:table"
+      >
+        <TableHeader>
+          <TableRow>
+            <TableHead class="w-[25%]">
+              全局模型
+            </TableHead>
+            <TableHead class="w-8 text-center" />
+            <TableHead class="w-[30%]">
+              映射模型
+            </TableHead>
+            <TableHead class="w-[25%]">
+              提供商
+            </TableHead>
+            <TableHead class="w-[10%] text-center">
+              剩余
+            </TableHead>
+            <TableHead class="w-[5%] text-right">
+              操作
+            </TableHead>
+          </TableRow>
+        </TableHeader>
+        <TableBody>
+          <TableRow
+            v-for="mapping in modelMappingStats.mappings"
+            :key="mapping.mapping_name"
+          >
+            <TableCell>
+              <div v-if="mapping.global_model_name">
+                <div class="text-sm font-medium">
+                  {{ mapping.global_model_display_name || mapping.global_model_name }}
+                </div>
+                <div
+                  v-if="mapping.global_model_display_name && mapping.global_model_display_name !== mapping.global_model_name"
+                  class="text-xs text-muted-foreground font-mono"
+                >
+                  {{ mapping.global_model_name }}
+                </div>
+              </div>
+              <span
+                v-else
+                class="text-sm text-muted-foreground"
+              >-</span>
+            </TableCell>
+            <TableCell class="text-center">
+              <ArrowRight class="h-4 w-4 text-muted-foreground" />
+            </TableCell>
+            <TableCell>
+              <span class="text-sm font-mono">{{ mapping.mapping_name }}</span>
+            </TableCell>
+            <TableCell>
+              <div
+                v-if="mapping.providers && mapping.providers.length > 0"
+                class="flex flex-wrap gap-1"
+              >
+                <Badge
+                  v-for="provider in mapping.providers.slice(0, 3)"
+                  :key="provider"
+                  variant="outline"
+                  class="text-xs"
+                >
+                  {{ provider }}
+                </Badge>
+                <Badge
+                  v-if="mapping.providers.length > 3"
+                  variant="outline"
+                  class="text-xs"
+                >
+                  +{{ mapping.providers.length - 3 }}
+                </Badge>
+              </div>
+              <span
+                v-else
+                class="text-sm text-muted-foreground"
+              >-</span>
+            </TableCell>
+            <TableCell class="text-center">
+              <span class="text-xs text-muted-foreground">{{ formatTTL(mapping.ttl) }}</span>
+            </TableCell>
+            <TableCell class="text-right">
+              <Button
+                size="icon"
+                variant="ghost"
+                class="h-6 w-6 text-muted-foreground/50 hover:text-destructive"
+                :disabled="clearingModelName === mapping.mapping_name"
+                title="清除缓存"
+                @click="clearModelMappingByName(mapping.mapping_name)"
+              >
+                <X class="h-3.5 w-3.5" />
+              </Button>
+            </TableCell>
+          </TableRow>
+        </TableBody>
+      </Table>
+
+      <!-- 移动端卡片列表 -->
+      <div
+        v-if="modelMappingStats?.available && modelMappingStats.mappings && modelMappingStats.mappings.length > 0"
+        class="md:hidden divide-y divide-border/40"
+      >
+        <div
+          v-for="mapping in modelMappingStats.mappings"
+          :key="`m-${mapping.mapping_name}`"
+          class="p-4 space-y-2"
+        >
+          <div class="flex items-center justify-between gap-2">
+            <span class="text-sm font-medium truncate">{{ mapping.global_model_display_name || mapping.global_model_name || '-' }}</span>
+            <Button
+              size="icon"
+              variant="ghost"
+              class="h-6 w-6 text-muted-foreground/50 hover:text-destructive shrink-0"
+              :disabled="clearingModelName === mapping.mapping_name"
+              @click="clearModelMappingByName(mapping.mapping_name)"
+            >
+              <X class="h-3.5 w-3.5" />
+            </Button>
+          </div>
+          <div class="flex items-center gap-2 text-xs text-muted-foreground">
+            <ArrowRight class="h-3.5 w-3.5 shrink-0" />
+            <span class="font-mono">{{ mapping.mapping_name }}</span>
+          </div>
+          <div
+            v-if="mapping.providers && mapping.providers.length > 0"
+            class="flex flex-wrap gap-1"
+          >
+            <Badge
+              v-for="provider in mapping.providers"
+              :key="provider"
+              variant="outline"
+              class="text-xs"
+            >
+              {{ provider }}
+            </Badge>
+          </div>
+        </div>
+      </div>
+
+      <!-- 未映射条目（NOT_FOUND 等） -->
+      <div
+        v-if="modelMappingStats?.available && modelMappingStats.unmapped && modelMappingStats.unmapped.length > 0"
+        class="px-6 py-4 border-t border-border/40"
+      >
+        <div class="text-xs text-muted-foreground mb-2">
+          未映射的缓存条目
+        </div>
+        <div class="flex flex-wrap gap-1.5">
+          <Badge
+            v-for="entry in modelMappingStats.unmapped"
+            :key="entry.mapping_name"
+            :variant="getUnmappedStatusBadge(entry.status).variant"
+            class="text-xs font-mono cursor-pointer"
+            :title="`${getUnmappedStatusBadge(entry.status).text} - 点击清除`"
+            @click="clearModelMappingByName(entry.mapping_name)"
+          >
+            {{ entry.mapping_name }}
+          </Badge>
+        </div>
+      </div>
+
+      <!-- Provider 模型映射缓存 -->
+      <div
+        v-if="modelMappingStats?.available && modelMappingStats.provider_model_mappings && modelMappingStats.provider_model_mappings.length > 0"
+        class="border-t border-border/40"
+      >
+        <div class="px-6 py-3 text-xs text-muted-foreground border-b border-border/30 bg-muted/20">
+          Provider 模型映射缓存
+        </div>
+        <!-- 桌面端表格 -->
+        <Table class="hidden md:table">
+          <TableHeader>
+            <TableRow>
+              <TableHead class="w-[15%]">
+                提供商
+              </TableHead>
+              <TableHead class="w-[25%]">
+                请求名称
+              </TableHead>
+              <TableHead class="w-8 text-center" />
+              <TableHead class="w-[25%]">
+                映射模型
+              </TableHead>
+              <TableHead class="w-[10%] text-center">
+                剩余
+              </TableHead>
+              <TableHead class="w-[10%] text-center">
+                次数
+              </TableHead>
+              <TableHead class="w-[7%] text-right">
+                操作
+              </TableHead>
+            </TableRow>
+          </TableHeader>
+          <TableBody>
+            <template
+              v-for="(mapping, index) in modelMappingStats.provider_model_mappings"
+              :key="index"
+            >
+              <TableRow
+                v-for="(alias, aliasIndex) in (mapping.aliases || [])"
+                :key="`${index}-${aliasIndex}`"
+              >
+                <TableCell>
+                  <Badge
+                    variant="outline"
+                    class="text-xs"
+                  >
+                    {{ mapping.provider_name }}
+                  </Badge>
+                </TableCell>
+                <TableCell>
+                  <span class="text-sm font-mono">{{ alias }}</span>
+                </TableCell>
+                <TableCell class="text-center">
+                  <ArrowRight class="h-4 w-4 text-muted-foreground" />
+                </TableCell>
+                <TableCell>
+                  <span class="text-sm font-mono font-medium">{{ mapping.provider_model_name }}</span>
+                </TableCell>
+                <TableCell class="text-center">
+                  <span class="text-xs text-muted-foreground">{{ formatTTL(mapping.ttl) }}</span>
+                </TableCell>
+                <TableCell class="text-center">
+                  <span class="text-sm">{{ mapping.hit_count || 0 }}</span>
+                </TableCell>
+                <TableCell class="text-right">
+                  <Button
+                    size="icon"
+                    variant="ghost"
+                    class="h-7 w-7 text-muted-foreground/70 hover:text-destructive"
+                    title="清除缓存"
+                    @click="clearProviderModelMapping(mapping.provider_id, mapping.global_model_id, `${mapping.provider_name} - ${alias}`)"
+                  >
+                    <Trash2 class="h-3.5 w-3.5" />
+                  </Button>
+                </TableCell>
+              </TableRow>
+            </template>
+          </TableBody>
+        </Table>
+        <!-- 移动端卡片 -->
+        <div class="md:hidden divide-y divide-border/40">
+          <template
+            v-for="(mapping, index) in modelMappingStats.provider_model_mappings"
+            :key="`m-pm-${index}`"
+          >
+            <div
+              v-for="(alias, aliasIndex) in (mapping.aliases || [])"
+              :key="`m-pm-${index}-${aliasIndex}`"
+              class="p-4 space-y-2"
+            >
+              <div class="flex items-center justify-between">
+                <Badge
+                  variant="outline"
+                  class="text-xs"
+                >
+                  {{ mapping.provider_name }}
+                </Badge>
+                <div class="flex items-center gap-2">
+                  <span class="text-xs text-muted-foreground">{{ formatTTL(mapping.ttl) }}</span>
+                  <span class="text-xs">{{ mapping.hit_count || 0 }}次</span>
+                  <Button
+                    size="icon"
+                    variant="ghost"
+                    class="h-6 w-6 text-muted-foreground/70 hover:text-destructive"
+                    title="清除缓存"
+                    @click="clearProviderModelMapping(mapping.provider_id, mapping.global_model_id, `${mapping.provider_name} - ${alias}`)"
+                  >
+                    <Trash2 class="h-3 w-3" />
+                  </Button>
+                </div>
+              </div>
+              <div class="flex items-center gap-2 text-sm">
+                <span class="font-mono">{{ alias }}</span>
+                <ArrowRight class="h-3.5 w-3.5 shrink-0 text-muted-foreground/60" />
+                <span class="font-mono font-medium">{{ mapping.provider_model_name }}</span>
+              </div>
+            </div>
+          </template>
+        </div>
+      </div>
+
+      <!-- 无缓存状态 -->
+      <div
+        v-else-if="modelMappingStats?.available && (!modelMappingStats.mappings || modelMappingStats.mappings.length === 0) && (!modelMappingStats.unmapped || modelMappingStats.unmapped.length === 0) && (!modelMappingStats.provider_model_mappings || modelMappingStats.provider_model_mappings.length === 0)"
+        class="px-6 py-8 text-center text-sm text-muted-foreground"
+      >
+        暂无模型解析缓存
+      </div>
+
+      <!-- Redis 未启用 -->
+      <div
+        v-else-if="modelMappingStats && !modelMappingStats.available"
+        class="px-6 py-8 text-center text-sm text-muted-foreground"
+      >
+        {{ modelMappingStats.message || 'Redis 未启用' }}
+      </div>
+
+      <!-- 加载中 -->
+      <div
+        v-else-if="modelMappingLoading"
+        class="px-6 py-8 text-center text-sm text-muted-foreground"
+      >
+        加载中...
+      </div>
+    </Card>
+
     <!-- TTL 分析区域 -->
     <Card class="overflow-hidden">
       <div class="px-4 sm:px-6 py-3 sm:py-3.5 border-b border-border/60">

frontend/src/views/admin/ModelManagement.vue

@@ -111,9 +111,6 @@
               <TableHead class="w-[80px] text-center">
                 提供商
               </TableHead>
-              <TableHead class="w-[70px] text-center">
-                别名/映射
-              </TableHead>
               <TableHead class="w-[80px] text-center">
                 调用次数
               </TableHead>
@@ -128,7 +125,7 @@
           <TableBody>
             <TableRow v-if="loading">
               <TableCell
-                colspan="8"
+                colspan="7"
                 class="text-center py-8"
               >
                 <Loader2 class="w-6 h-6 animate-spin mx-auto" />
@@ -136,7 +133,7 @@
             </TableRow>
             <TableRow v-else-if="filteredGlobalModels.length === 0">
               <TableCell
-                colspan="8"
+                colspan="7"
                 class="text-center py-8 text-muted-foreground"
               >
                 没有找到匹配的模型
@@ -171,27 +168,27 @@
                   <div class="space-y-1 w-fit">
                     <div class="flex flex-wrap gap-1">
                       <Zap
-                        v-if="model.default_supports_streaming"
+                        v-if="model.config?.streaming !== false"
                         class="w-4 h-4 text-muted-foreground"
                         title="流式输出"
                       />
                       <Image
-                        v-if="model.default_supports_image_generation"
+                        v-if="model.config?.image_generation === true"
                         class="w-4 h-4 text-muted-foreground"
                         title="图像生成"
                       />
                       <Eye
-                        v-if="model.default_supports_vision"
+                        v-if="model.config?.vision === true"
                         class="w-4 h-4 text-muted-foreground"
                         title="视觉理解"
                       />
                       <Wrench
-                        v-if="model.default_supports_function_calling"
+                        v-if="model.config?.function_calling === true"
                         class="w-4 h-4 text-muted-foreground"
                         title="工具调用"
                       />
                       <Brain
-                        v-if="model.default_supports_extended_thinking"
+                        v-if="model.config?.extended_thinking === true"
                         class="w-4 h-4 text-muted-foreground"
                         title="深度思考"
                       />
@@ -244,11 +241,6 @@
                     {{ model.provider_count || 0 }}
                   </Badge>
                 </TableCell>
-                <TableCell class="text-center">
-                  <Badge variant="secondary">
-                    {{ model.alias_count || 0 }}
-                  </Badge>
-                </TableCell>
                 <TableCell class="text-center">
                   <span class="text-sm font-mono">{{ formatUsageCount(model.usage_count || 0) }}</span>
                 </TableCell>
@@ -369,23 +361,23 @@
             <!-- 第二行：能力图标 -->
             <div class="flex flex-wrap gap-1.5">
               <Zap
-                v-if="model.default_supports_streaming"
+                v-if="model.config?.streaming !== false"
                 class="w-4 h-4 text-muted-foreground"
               />
               <Image
-                v-if="model.default_supports_image_generation"
+                v-if="model.config?.image_generation === true"
                 class="w-4 h-4 text-muted-foreground"
               />
               <Eye
-                v-if="model.default_supports_vision"
+                v-if="model.config?.vision === true"
                 class="w-4 h-4 text-muted-foreground"
               />
               <Wrench
-                v-if="model.default_supports_function_calling"
+                v-if="model.config?.function_calling === true"
                 class="w-4 h-4 text-muted-foreground"
               />
               <Brain
-                v-if="model.default_supports_extended_thinking"
+                v-if="model.config?.extended_thinking === true"
                 class="w-4 h-4 text-muted-foreground"
               />
             </div>
@@ -393,7 +385,6 @@
             <!-- 第三行：统计信息 -->
             <div class="flex flex-wrap items-center gap-3 text-xs text-muted-foreground">
               <span>提供商 {{ model.provider_count || 0 }}</span>
-              <span>别名 {{ model.alias_count || 0 }}</span>
               <span>调用 {{ formatUsageCount(model.usage_count || 0) }}</span>
               <span
                 v-if="getFirstTierPrice(model, 'input') || getFirstTierPrice(model, 'output')"
@@ -425,25 +416,12 @@
       @success="handleModelFormSuccess"
     />
 
-    <!-- 创建/编辑别名/映射对话框 -->
-    <AliasDialog
-      :open="createAliasDialogOpen"
-      :editing-alias="editingAlias"
-      :global-models="globalModels"
-      :providers="providers"
-      :fixed-target-model="isTargetModelFixed ? selectedModel : null"
-      @update:open="handleAliasDialogUpdate"
-      @submit="handleAliasSubmit"
-    />
-
     <!-- 模型详情抽屉 -->
     <ModelDetailDrawer
       :model="selectedModel"
       :open="!!selectedModel"
       :providers="selectedModelProviders"
-      :aliases="selectedModelAliases"
       :loading-providers="loadingModelProviders"
-      :loading-aliases="loadingModelAliases"
       :has-blocking-dialog-open="hasBlockingDialogOpen"
       :capabilities="capabilities"
       @update:open="handleDrawerOpenChange"
@@ -454,11 +432,6 @@
       @delete-provider="confirmDeleteProviderImplementation"
       @toggle-provider-status="toggleProviderStatus"
       @refresh-providers="refreshSelectedModelProviders"
-      @add-alias="openAddAliasDialog"
-      @edit-alias="openEditAliasDialog"
-      @toggle-alias-status="toggleAliasStatusFromDrawer"
-      @delete-alias="confirmDeleteAliasFromDrawer"
-      @refresh-aliases="refreshSelectedModelAliases"
     />
 
     <!-- 批量添加关联提供商对话框 -->
@@ -736,9 +709,7 @@ import {
 } from 'lucide-vue-next'
 import ModelDetailDrawer from '@/features/models/components/ModelDetailDrawer.vue'
 import GlobalModelFormDialog from '@/features/models/components/GlobalModelFormDialog.vue'
-import AliasDialog from '@/features/models/components/AliasDialog.vue'
 import ProviderModelFormDialog from '@/features/providers/components/ProviderModelFormDialog.vue'
-import type { CreateModelAliasRequest, UpdateModelAliasRequest } from '@/api/endpoints/aliases'
 import type { Model } from '@/api/endpoints'
 import { useToast } from '@/composables/useToast'
 import { useConfirm } from '@/composables/useConfirm'
@@ -768,13 +739,6 @@ import {
   type GlobalModelResponse,
 } from '@/api/global-models'
 import { log } from '@/utils/logger'
-import {
-  getAliases,
-  createAlias,
-  updateAlias,
-  deleteAlias,
-  type ModelAlias,
-} from '@/api/endpoints/aliases'
 import { getProvidersSummary } from '@/api/endpoints/providers'
 import { getAllCapabilities, type CapabilityDefinition } from '@/api/endpoints'
 
@@ -788,13 +752,9 @@ const searchQuery = ref('')
 const selectedModel = ref<GlobalModelResponse | null>(null)
 const createModelDialogOpen = ref(false)
 const editingModel = ref<GlobalModelResponse | null>(null)
-const createAliasDialogOpen = ref(false)
-const editingAliasId = ref<string | null>(null)
-const isTargetModelFixed = ref(false)  // 目标模型是否固定（从模型详情抽屉打开时为 true）
 
 // 数据
 const globalModels = ref<GlobalModelResponse[]>([])
-const allAliases = ref<ModelAlias[]>([])
 const providers = ref<any[]>([])
 const capabilities = ref<CapabilityDefinition[]>([])
 
@@ -804,9 +764,7 @@ const catalogPageSize = ref(20)
 
 // 选中模型的详细数据
 const selectedModelProviders = ref<any[]>([])
-const selectedModelAliases = ref<ModelAlias[]>([])
 const loadingModelProviders = ref(false)
-const loadingModelAliases = ref(false)
 
 // 批量添加关联提供商
 const batchAddProvidersDialogOpen = ref(false)
@@ -876,19 +834,10 @@ function hasTieredPricing(model: GlobalModelResponse): boolean {
 // 检测是否有对话框打开（防止误关闭抽屉）
 const hasBlockingDialogOpen = computed(() =>
   createModelDialogOpen.value ||
-  createAliasDialogOpen.value ||
   batchAddProvidersDialogOpen.value ||
   editProviderDialogOpen.value
 )
 
-// 编辑中的别名对象（用于传递给 AliasDialog）
-const editingAlias = computed(() => {
-  if (!editingAliasId.value) return null
-  return allAliases.value.find(a => a.id === editingAliasId.value) ||
-         selectedModelAliases.value.find(a => a.id === editingAliasId.value) ||
-         null
-})
-
 // 能力筛选
 const capabilityFilters = ref({
   streaming: false,
@@ -1053,30 +1002,30 @@ async function batchRemoveSelectedProviders() {
 const filteredGlobalModels = computed(() => {
   let result = globalModels.value
 
-  // 搜索
+  // 搜索（支持空格分隔的多关键词 AND 搜索）
   if (searchQuery.value) {
-    const query = searchQuery.value.toLowerCase()
-    result = result.filter(m =>
-      m.name.toLowerCase().includes(query) ||
-      m.display_name?.toLowerCase().includes(query)
-    )
+    const keywords = searchQuery.value.toLowerCase().split(/\s+/).filter(k => k.length > 0)
+    result = result.filter(m => {
+      const searchableText = `${m.name} ${m.display_name || ''}`.toLowerCase()
+      return keywords.every(keyword => searchableText.includes(keyword))
+    })
   }
 
   // 能力筛选
   if (capabilityFilters.value.streaming) {
-    result = result.filter(m => m.default_supports_streaming)
+    result = result.filter(m => m.config?.streaming !== false)
   }
   if (capabilityFilters.value.imageGeneration) {
-    result = result.filter(m => m.default_supports_image_generation)
+    result = result.filter(m => m.config?.image_generation === true)
   }
   if (capabilityFilters.value.vision) {
-    result = result.filter(m => m.default_supports_vision)
+    result = result.filter(m => m.config?.vision === true)
   }
   if (capabilityFilters.value.toolUse) {
-    result = result.filter(m => m.default_supports_function_calling)
+    result = result.filter(m => m.config?.function_calling === true)
   }
   if (capabilityFilters.value.extendedThinking) {
-    result = result.filter(m => m.default_supports_extended_thinking)
+    result = result.filter(m => m.config?.extended_thinking === true)
   }
 
   return result
@@ -1131,11 +1080,8 @@ async function selectModel(model: GlobalModelResponse) {
   selectedModel.value = model
   detailTab.value = 'basic'
 
-  // 加载该模型的关联提供商和别名
-  await Promise.all([
-    loadModelProviders(model.id),
-    loadModelAliases(model.id)
-  ])
+  // 加载该模型的关联提供商
+  await loadModelProviders(model.id)
 }
 
 // 加载指定模型的关联提供商
@@ -1187,27 +1133,6 @@ async function loadModelProviders(_globalModelId: string) {
   }
 }
 
-// 加载指定模型的别名
-async function loadModelAliases(globalModelId: string) {
-  loadingModelAliases.value = true
-  try {
-    const aliases = await getAliases({ limit: 1000 })
-    selectedModelAliases.value = aliases.filter(a => a.global_model_id === globalModelId)
-  } catch (err: any) {
-    log.error('加载别名失败:', err)
-    selectedModelAliases.value = []
-  } finally {
-    loadingModelAliases.value = false
-  }
-}
-
-// 刷新当前选中模型的别名
-async function refreshSelectedModelAliases() {
-  if (selectedModel.value) {
-    await loadModelAliases(selectedModel.value.id)
-  }
-}
-
 // 刷新当前选中模型的关联提供商
 async function refreshSelectedModelProviders() {
   if (selectedModel.value) {
@@ -1329,14 +1254,6 @@ async function confirmDeleteProviderImplementation(provider: any) {
   }
 }
 
-// 打开添加别名对话框（从模型详情抽屉）
-function openAddAliasDialog() {
-  if (!selectedModel.value) return
-  editingAliasId.value = null
-  isTargetModelFixed.value = true  // 目标模型固定为当前选中模型
-  createAliasDialogOpen.value = true
-}
-
 function openCreateModelDialog() {
   editingModel.value = null
   createModelDialogOpen.value = true
@@ -1391,106 +1308,6 @@ async function toggleModelStatus(model: GlobalModelResponse) {
   }
 }
 
-async function loadAliases() {
-  try {
-    allAliases.value = await getAliases({ limit: 1000 })
-  } catch (err: any) {
-    showError(err.response?.data?.detail || err.message, '加载别名失败')
-  }
-}
-
-function openEditAliasDialog(alias: ModelAlias) {
-  editingAliasId.value = alias.id
-  isTargetModelFixed.value = false
-  createAliasDialogOpen.value = true
-}
-
-// 处理别名对话框关闭事件
-function handleAliasDialogUpdate(value: boolean) {
-  createAliasDialogOpen.value = value
-  if (!value) {
-    editingAliasId.value = null
-    isTargetModelFixed.value = false
-  }
-}
-
-// 处理别名提交（来自 AliasDialog 组件）
-async function handleAliasSubmit(data: CreateModelAliasRequest | UpdateModelAliasRequest, isEdit: boolean) {
-  submitting.value = true
-  try {
-    if (isEdit && editingAliasId.value) {
-      // 更新
-      await updateAlias(editingAliasId.value, data as UpdateModelAliasRequest)
-      success(data.mapping_type === 'mapping' ? '映射已更新' : '别名已更新')
-    } else {
-      // 创建
-      await createAlias(data as CreateModelAliasRequest)
-      success(data.mapping_type === 'mapping' ? '映射已创建' : '别名已创建')
-    }
-    createAliasDialogOpen.value = false
-    editingAliasId.value = null
-    isTargetModelFixed.value = false
-
-    // 刷新数据
-    await loadAliases()
-    if (selectedModel.value) {
-      await loadModelAliases(selectedModel.value.id)
-    }
-    // 刷新外层模型列表以更新 alias_count
-    await loadGlobalModels()
-  } catch (err: any) {
-    const detail = err.response?.data?.detail || err.message
-    // 优化错误提示文案
-    let errorMessage = detail
-    if (detail === '映射已存在') {
-      errorMessage = '目标作用域已存在同名别名，请先删除冲突的映射或选择其他作用域'
-    }
-    showError(errorMessage, isEdit ? '更新失败' : '创建失败')
-  } finally {
-    submitting.value = false
-  }
-}
-
-async function confirmDeleteAlias(alias: ModelAlias) {
-  const confirmed = await confirmDanger(
-    `确定要删除别名 "${alias.alias}" 吗？`,
-    '删除别名'
-  )
-  if (!confirmed) return
-
-  try {
-    await deleteAlias(alias.id)
-    success('别名已删除')
-    await loadAliases()
-    // 刷新外层模型列表以更新 alias_count
-    await loadGlobalModels()
-  } catch (err: any) {
-    showError(err.response?.data?.detail || err.message, '删除失败')
-  }
-}
-
-async function toggleAliasStatus(alias: ModelAlias) {
-  try {
-    await updateAlias(alias.id, { is_active: !alias.is_active })
-    alias.is_active = !alias.is_active
-    success(alias.is_active ? '别名已启用' : '别名已停用')
-  } catch (err: any) {
-    showError(err.response?.data?.detail || err.message, '操作失败')
-  }
-}
-
-// 从抽屉中切换别名状态
-async function toggleAliasStatusFromDrawer(alias: ModelAlias) {
-  await toggleAliasStatus(alias)
-  await refreshSelectedModelAliases()
-}
-
-// 从抽屉中删除别名
-async function confirmDeleteAliasFromDrawer(alias: ModelAlias) {
-  await confirmDeleteAlias(alias)
-  await refreshSelectedModelAliases()
-}
-
 async function refreshData() {
   await loadGlobalModels()
 }

frontend/src/views/admin/ProviderManagement.vue

@@ -505,13 +505,13 @@ const priorityModeConfig = computed(() => {
 const filteredProviders = computed(() => {
   let result = [...providers.value]
 
-  // 搜索筛选
+  // 搜索筛选（支持空格分隔的多关键词 AND 搜索）
   if (searchQuery.value.trim()) {
-    const query = searchQuery.value.toLowerCase()
-    result = result.filter(p =>
-      p.display_name.toLowerCase().includes(query) ||
-      p.name.toLowerCase().includes(query)
-    )
+    const keywords = searchQuery.value.toLowerCase().split(/\s+/).filter(k => k.length > 0)
+    result = result.filter(p => {
+      const searchableText = `${p.display_name} ${p.name}`.toLowerCase()
+      return keywords.every(keyword => searchableText.includes(keyword))
+    })
   }
 
   // 排序

frontend/src/views/admin/SystemSettings.vue

@@ -7,6 +7,7 @@
       <template #actions>
         <Button
           :disabled="loading"
+          class="shadow-none hover:shadow-none"
           @click="saveSystemConfig"
         >
           {{ loading ? '保存中...' : '保存所有配置' }}
@@ -15,6 +16,94 @@
     </PageHeader>
 
     <div class="mt-6 space-y-6">
+      <!-- 配置导出/导入 -->
+      <CardSection
+        title="配置管理"
+        description="导出或导入提供商和模型配置，便于备份或迁移"
+      >
+        <div class="flex flex-wrap gap-4">
+          <div class="flex-1 min-w-[200px]">
+            <p class="text-sm text-muted-foreground mb-3">
+              导出当前所有提供商、端点、API Key 和模型配置到 JSON 文件
+            </p>
+            <Button
+              variant="outline"
+              :disabled="exportLoading"
+              @click="handleExportConfig"
+            >
+              <Download class="w-4 h-4 mr-2" />
+              {{ exportLoading ? '导出中...' : '导出配置' }}
+            </Button>
+          </div>
+          <div class="flex-1 min-w-[200px]">
+            <p class="text-sm text-muted-foreground mb-3">
+              从 JSON 文件导入配置，支持跳过、覆盖或报错三种冲突处理模式
+            </p>
+            <div class="flex items-center gap-2">
+              <input
+                ref="configFileInput"
+                type="file"
+                accept=".json"
+                class="hidden"
+                @change="handleConfigFileSelect"
+              >
+              <Button
+                variant="outline"
+                :disabled="importLoading"
+                @click="triggerConfigFileSelect"
+              >
+                <Upload class="w-4 h-4 mr-2" />
+                {{ importLoading ? '导入中...' : '导入配置' }}
+              </Button>
+            </div>
+          </div>
+        </div>
+      </CardSection>
+
+      <!-- 用户数据导出/导入 -->
+      <CardSection
+        title="用户数据管理"
+        description="导出或导入用户及其 API Keys 数据（不含管理员）"
+      >
+        <div class="flex flex-wrap gap-4">
+          <div class="flex-1 min-w-[200px]">
+            <p class="text-sm text-muted-foreground mb-3">
+              导出所有普通用户及其 API Keys 到 JSON 文件
+            </p>
+            <Button
+              variant="outline"
+              :disabled="exportUsersLoading"
+              @click="handleExportUsers"
+            >
+              <Download class="w-4 h-4 mr-2" />
+              {{ exportUsersLoading ? '导出中...' : '导出用户数据' }}
+            </Button>
+          </div>
+          <div class="flex-1 min-w-[200px]">
+            <p class="text-sm text-muted-foreground mb-3">
+              从 JSON 文件导入用户数据（需相同 ENCRYPTION_KEY）
+            </p>
+            <div class="flex items-center gap-2">
+              <input
+                ref="usersFileInput"
+                type="file"
+                accept=".json"
+                class="hidden"
+                @change="handleUsersFileSelect"
+              >
+              <Button
+                variant="outline"
+                :disabled="importUsersLoading"
+                @click="triggerUsersFileSelect"
+              >
+                <Upload class="w-4 h-4 mr-2" />
+                {{ importUsersLoading ? '导入中...' : '导入用户数据' }}
+              </Button>
+            </div>
+          </div>
+        </div>
+      </CardSection>
+
       <!-- 基础配置 -->
       <CardSection
         title="基础配置"
@@ -96,32 +185,13 @@
         </div>
       </CardSection>
 
-      <!-- API Key 管理配置 -->
+      <!-- 独立余额 Key 过期管理 -->
       <CardSection
-        title="API Key 管理"
-        description="API Key 相关配置"
+        title="独立余额 Key 过期管理"
+        description="独立余额 Key 的过期处理策略（普通用户 Key 不会过期）"
       >
         <div class="grid grid-cols-1 md:grid-cols-2 gap-6">
-          <div>
-            <Label
-              for="api-key-expire"
-              class="block text-sm font-medium"
-            >
-              API密钥过期天数
-            </Label>
-            <Input
-              id="api-key-expire"
-              v-model.number="systemConfig.api_key_expire_days"
-              type="number"
-              placeholder="0"
-              class="mt-1"
-            />
-            <p class="mt-1 text-xs text-muted-foreground">
-              0 表示永不过期
-            </p>
-          </div>
-
-          <div class="flex items-center h-full pt-6">
+          <div class="flex items-center h-full">
             <div class="flex items-center space-x-2">
               <Checkbox
                 id="auto-delete-expired-keys"
@@ -135,7 +205,7 @@
                   自动删除过期 Key
                 </Label>
                 <p class="text-xs text-muted-foreground">
-                  关闭时仅禁用过期 Key
+                  关闭时仅禁用过期 Key，不会物理删除
                 </p>
               </div>
             </div>
@@ -359,6 +429,25 @@
               避免单次操作过大影响性能
             </p>
           </div>
+
+          <div>
+            <Label
+              for="audit-log-retention-days"
+              class="block text-sm font-medium"
+            >
+              审计日志保留天数
+            </Label>
+            <Input
+              id="audit-log-retention-days"
+              v-model.number="systemConfig.audit_log_retention_days"
+              type="number"
+              placeholder="30"
+              class="mt-1"
+            />
+            <p class="mt-1 text-xs text-muted-foreground">
+              超过后删除审计日志记录
+            </p>
+          </div>
         </div>
 
         <!-- 清理策略说明 -->
@@ -371,15 +460,318 @@
             <p>2. <strong>压缩日志阶段</strong>: body 字段被压缩存储，节省空间</p>
             <p>3. <strong>统计阶段</strong>: 仅保留 tokens、成本等统计信息</p>
             <p>4. <strong>归档删除</strong>: 超过保留期限后完全删除记录</p>
+            <p>5. <strong>审计日志</strong>: 独立清理，记录用户登录、操作等安全事件</p>
           </div>
         </div>
       </CardSection>
+
     </div>
+
+    <!-- 导入配置对话框 -->
+    <Dialog
+      v-model:open="importDialogOpen"
+      title="导入配置"
+      description="选择冲突处理模式并确认导入"
+    >
+      <div class="space-y-4">
+        <div
+          v-if="importPreview"
+          class="p-3 bg-muted rounded-lg text-sm"
+        >
+          <p class="font-medium mb-2">
+            配置预览
+          </p>
+          <ul class="space-y-1 text-muted-foreground">
+            <li>全局模型: {{ importPreview.global_models?.length || 0 }} 个</li>
+            <li>提供商: {{ importPreview.providers?.length || 0 }} 个</li>
+            <li>
+              端点: {{ importPreview.providers?.reduce((sum: number, p: any) => sum + (p.endpoints?.length || 0), 0) }} 个
+            </li>
+            <li>
+              API Keys: {{ importPreview.providers?.reduce((sum: number, p: any) => sum + p.endpoints?.reduce((s: number, e: any) => s + (e.keys?.length || 0), 0), 0) }} 个
+            </li>
+          </ul>
+        </div>
+
+        <div>
+          <Label class="block text-sm font-medium mb-2">冲突处理模式</Label>
+          <Select
+            v-model="mergeMode"
+            v-model:open="mergeModeSelectOpen"
+          >
+            <SelectTrigger>
+              <SelectValue />
+            </SelectTrigger>
+            <SelectContent>
+              <SelectItem value="skip">
+                跳过 - 保留现有配置
+              </SelectItem>
+              <SelectItem value="overwrite">
+                覆盖 - 用导入配置替换
+              </SelectItem>
+              <SelectItem value="error">
+                报错 - 遇到冲突时中止
+              </SelectItem>
+            </SelectContent>
+          </Select>
+          <p class="mt-1 text-xs text-muted-foreground">
+            <template v-if="mergeMode === 'skip'">
+              已存在的配置将被保留，仅导入新配置
+            </template>
+            <template v-else-if="mergeMode === 'overwrite'">
+              已存在的配置将被导入的配置覆盖
+            </template>
+            <template v-else>
+              如果发现任何冲突，导入将中止并回滚
+            </template>
+          </p>
+        </div>
+
+        <p class="text-xs text-muted-foreground">
+          注意：相同的 API Keys 会自动跳过，不会创建重复记录。
+        </p>
+      </div>
+
+      <template #footer>
+        <Button
+          variant="outline"
+          @click="importDialogOpen = false; mergeModeSelectOpen = false"
+        >
+          取消
+        </Button>
+        <Button
+          :disabled="importLoading"
+          @click="confirmImport"
+        >
+          {{ importLoading ? '导入中...' : '确认导入' }}
+        </Button>
+      </template>
+    </Dialog>
+
+    <!-- 导入结果对话框 -->
+    <Dialog
+      v-model:open="importResultDialogOpen"
+      title="导入完成"
+    >
+      <div
+        v-if="importResult"
+        class="space-y-4"
+      >
+        <div class="grid grid-cols-2 gap-4 text-sm">
+          <div class="p-3 bg-muted rounded-lg">
+            <p class="font-medium">
+              全局模型
+            </p>
+            <p class="text-muted-foreground">
+              创建: {{ importResult.stats.global_models.created }},
+              更新: {{ importResult.stats.global_models.updated }},
+              跳过: {{ importResult.stats.global_models.skipped }}
+            </p>
+          </div>
+          <div class="p-3 bg-muted rounded-lg">
+            <p class="font-medium">
+              提供商
+            </p>
+            <p class="text-muted-foreground">
+              创建: {{ importResult.stats.providers.created }},
+              更新: {{ importResult.stats.providers.updated }},
+              跳过: {{ importResult.stats.providers.skipped }}
+            </p>
+          </div>
+          <div class="p-3 bg-muted rounded-lg">
+            <p class="font-medium">
+              端点
+            </p>
+            <p class="text-muted-foreground">
+              创建: {{ importResult.stats.endpoints.created }},
+              更新: {{ importResult.stats.endpoints.updated }},
+              跳过: {{ importResult.stats.endpoints.skipped }}
+            </p>
+          </div>
+          <div class="p-3 bg-muted rounded-lg">
+            <p class="font-medium">
+              API Keys
+            </p>
+            <p class="text-muted-foreground">
+              创建: {{ importResult.stats.keys.created }},
+              跳过: {{ importResult.stats.keys.skipped }}
+            </p>
+          </div>
+          <div class="p-3 bg-muted rounded-lg col-span-2">
+            <p class="font-medium">
+              模型配置
+            </p>
+            <p class="text-muted-foreground">
+              创建: {{ importResult.stats.models.created }},
+              更新: {{ importResult.stats.models.updated }},
+              跳过: {{ importResult.stats.models.skipped }}
+            </p>
+          </div>
+        </div>
+
+        <div
+          v-if="importResult.stats.errors.length > 0"
+          class="p-3 bg-destructive/10 rounded-lg"
+        >
+          <p class="font-medium text-destructive mb-2">
+            警告信息
+          </p>
+          <ul class="text-sm text-destructive space-y-1">
+            <li
+              v-for="(err, index) in importResult.stats.errors"
+              :key="index"
+            >
+              {{ err }}
+            </li>
+          </ul>
+        </div>
+      </div>
+
+      <template #footer>
+        <Button @click="importResultDialogOpen = false">
+          确定
+        </Button>
+      </template>
+    </Dialog>
+
+    <!-- 用户数据导入对话框 -->
+    <Dialog
+      v-model:open="importUsersDialogOpen"
+      title="导入用户数据"
+      description="选择冲突处理模式并确认导入"
+    >
+      <div class="space-y-4">
+        <div
+          v-if="importUsersPreview"
+          class="p-3 bg-muted rounded-lg text-sm"
+        >
+          <p class="font-medium mb-2">
+            数据预览
+          </p>
+          <ul class="space-y-1 text-muted-foreground">
+            <li>用户: {{ importUsersPreview.users?.length || 0 }} 个</li>
+            <li>
+              API Keys: {{ importUsersPreview.users?.reduce((sum: number, u: any) => sum + (u.api_keys?.length || 0), 0) }} 个
+            </li>
+          </ul>
+        </div>
+
+        <div>
+          <Label class="block text-sm font-medium mb-2">冲突处理模式</Label>
+          <Select
+            v-model="usersMergeMode"
+            v-model:open="usersMergeModeSelectOpen"
+          >
+            <SelectTrigger>
+              <SelectValue />
+            </SelectTrigger>
+            <SelectContent>
+              <SelectItem value="skip">
+                跳过 - 保留现有用户
+              </SelectItem>
+              <SelectItem value="overwrite">
+                覆盖 - 用导入数据替换
+              </SelectItem>
+              <SelectItem value="error">
+                报错 - 遇到冲突时中止
+              </SelectItem>
+            </SelectContent>
+          </Select>
+          <p class="mt-1 text-xs text-muted-foreground">
+            <template v-if="usersMergeMode === 'skip'">
+              已存在的用户将被保留，仅导入新用户
+            </template>
+            <template v-else-if="usersMergeMode === 'overwrite'">
+              已存在的用户将被导入的数据覆盖
+            </template>
+            <template v-else>
+              如果发现任何冲突，导入将中止并回滚
+            </template>
+          </p>
+        </div>
+
+        <p class="text-xs text-muted-foreground">
+          注意：用户 API Keys 需要目标系统使用相同的 ENCRYPTION_KEY 环境变量才能正常工作。
+        </p>
+      </div>
+
+      <template #footer>
+        <Button
+          variant="outline"
+          @click="importUsersDialogOpen = false; usersMergeModeSelectOpen = false"
+        >
+          取消
+        </Button>
+        <Button
+          :disabled="importUsersLoading"
+          @click="confirmImportUsers"
+        >
+          {{ importUsersLoading ? '导入中...' : '确认导入' }}
+        </Button>
+      </template>
+    </Dialog>
+
+    <!-- 用户数据导入结果对话框 -->
+    <Dialog
+      v-model:open="importUsersResultDialogOpen"
+      title="用户数据导入完成"
+    >
+      <div
+        v-if="importUsersResult"
+        class="space-y-4"
+      >
+        <div class="grid grid-cols-2 gap-4 text-sm">
+          <div class="p-3 bg-muted rounded-lg">
+            <p class="font-medium">
+              用户
+            </p>
+            <p class="text-muted-foreground">
+              创建: {{ importUsersResult.stats.users.created }},
+              更新: {{ importUsersResult.stats.users.updated }},
+              跳过: {{ importUsersResult.stats.users.skipped }}
+            </p>
+          </div>
+          <div class="p-3 bg-muted rounded-lg">
+            <p class="font-medium">
+              API Keys
+            </p>
+            <p class="text-muted-foreground">
+              创建: {{ importUsersResult.stats.api_keys.created }},
+              跳过: {{ importUsersResult.stats.api_keys.skipped }}
+            </p>
+          </div>
+        </div>
+
+        <div
+          v-if="importUsersResult.stats.errors.length > 0"
+          class="p-3 bg-destructive/10 rounded-lg"
+        >
+          <p class="font-medium text-destructive mb-2">
+            警告信息
+          </p>
+          <ul class="text-sm text-destructive space-y-1">
+            <li
+              v-for="(err, index) in importUsersResult.stats.errors"
+              :key="index"
+            >
+              {{ err }}
+            </li>
+          </ul>
+        </div>
+      </div>
+
+      <template #footer>
+        <Button @click="importUsersResultDialogOpen = false">
+          确定
+        </Button>
+      </template>
+    </Dialog>
   </PageContainer>
 </template>
 
 <script setup lang="ts">
 import { ref, computed, onMounted } from 'vue'
+import { Download, Upload } from 'lucide-vue-next'
 import Button from '@/components/ui/button.vue'
 import Input from '@/components/ui/input.vue'
 import Label from '@/components/ui/label.vue'
@@ -389,9 +781,12 @@ import SelectTrigger from '@/components/ui/select-trigger.vue'
 import SelectValue from '@/components/ui/select-value.vue'
 import SelectContent from '@/components/ui/select-content.vue'
 import SelectItem from '@/components/ui/select-item.vue'
+import {
+  Dialog,
+} from '@/components/ui'
 import { PageHeader, PageContainer, CardSection } from '@/components/layout'
 import { useToast } from '@/composables/useToast'
-import { adminApi } from '@/api/admin'
+import { adminApi, type ConfigExportData, type ConfigImportResponse, type UsersExportData, type UsersImportResponse } from '@/api/admin'
 import { log } from '@/utils/logger'
 
 const { success, error } = useToast()
@@ -403,8 +798,7 @@ interface SystemConfig {
   // 用户注册
   enable_registration: boolean
   require_email_verification: boolean
-  // API Key 管理
-  api_key_expire_days: number
+  // 独立余额 Key 过期管理
   auto_delete_expired_keys: boolean
   // 日志记录
   request_log_level: string
@@ -418,11 +812,34 @@ interface SystemConfig {
   header_retention_days: number
   log_retention_days: number
   cleanup_batch_size: number
+  audit_log_retention_days: number
 }
 
 const loading = ref(false)
 const logLevelSelectOpen = ref(false)
 
+// 导出/导入相关
+const exportLoading = ref(false)
+const importLoading = ref(false)
+const importDialogOpen = ref(false)
+const importResultDialogOpen = ref(false)
+const configFileInput = ref<HTMLInputElement | null>(null)
+const importPreview = ref<ConfigExportData | null>(null)
+const importResult = ref<ConfigImportResponse | null>(null)
+const mergeMode = ref<'skip' | 'overwrite' | 'error'>('skip')
+const mergeModeSelectOpen = ref(false)
+
+// 用户数据导出/导入相关
+const exportUsersLoading = ref(false)
+const importUsersLoading = ref(false)
+const importUsersDialogOpen = ref(false)
+const importUsersResultDialogOpen = ref(false)
+const usersFileInput = ref<HTMLInputElement | null>(null)
+const importUsersPreview = ref<UsersExportData | null>(null)
+const importUsersResult = ref<UsersImportResponse | null>(null)
+const usersMergeMode = ref<'skip' | 'overwrite' | 'error'>('skip')
+const usersMergeModeSelectOpen = ref(false)
+
 const systemConfig = ref<SystemConfig>({
   // 基础配置
   default_user_quota_usd: 10.0,
@@ -430,8 +847,7 @@ const systemConfig = ref<SystemConfig>({
   // 用户注册
   enable_registration: false,
   require_email_verification: false,
-  // API Key 管理
-  api_key_expire_days: 0,
+  // 独立余额 Key 过期管理
   auto_delete_expired_keys: false,
   // 日志记录
   request_log_level: 'basic',
@@ -445,6 +861,7 @@ const systemConfig = ref<SystemConfig>({
   header_retention_days: 90,
   log_retention_days: 365,
   cleanup_batch_size: 1000,
+  audit_log_retention_days: 30,
 })
 
 // 计算属性：KB 和 字节 之间的转换
@@ -486,8 +903,7 @@ async function loadSystemConfig() {
       // 用户注册
       'enable_registration',
       'require_email_verification',
-      // API Key 管理
-      'api_key_expire_days',
+      // 独立余额 Key 过期管理
       'auto_delete_expired_keys',
       // 日志记录
       'request_log_level',
@@ -501,6 +917,7 @@ async function loadSystemConfig() {
       'header_retention_days',
       'log_retention_days',
       'cleanup_batch_size',
+      'audit_log_retention_days',
     ]
 
     for (const key of configs) {
@@ -545,12 +962,7 @@ async function saveSystemConfig() {
         value: systemConfig.value.require_email_verification,
         description: '是否需要邮箱验证'
       },
-      // API Key 管理
-      {
-        key: 'api_key_expire_days',
-        value: systemConfig.value.api_key_expire_days,
-        description: 'API密钥过期天数'
-      },
+      // 独立余额 Key 过期管理
       {
         key: 'auto_delete_expired_keys',
         value: systemConfig.value.auto_delete_expired_keys,
@@ -608,6 +1020,11 @@ async function saveSystemConfig() {
         value: systemConfig.value.cleanup_batch_size,
         description: '每批次清理的记录数'
       },
+      {
+        key: 'audit_log_retention_days',
+        value: systemConfig.value.audit_log_retention_days,
+        description: '审计日志保留天数'
+      },
     ]
 
     const promises = configItems.map(item =>
@@ -623,4 +1040,185 @@ async function saveSystemConfig() {
     loading.value = false
   }
 }
+
+// 导出配置
+async function handleExportConfig() {
+  exportLoading.value = true
+  try {
+    const data = await adminApi.exportConfig()
+    const blob = new Blob([JSON.stringify(data, null, 2)], { type: 'application/json' })
+    const url = URL.createObjectURL(blob)
+    const a = document.createElement('a')
+    a.href = url
+    a.download = `aether-config-${new Date().toISOString().slice(0, 10)}.json`
+    document.body.appendChild(a)
+    a.click()
+    document.body.removeChild(a)
+    URL.revokeObjectURL(url)
+    success('配置已导出')
+  } catch (err) {
+    error('导出配置失败')
+    log.error('导出配置失败:', err)
+  } finally {
+    exportLoading.value = false
+  }
+}
+
+// 触发文件选择
+function triggerConfigFileSelect() {
+  configFileInput.value?.click()
+}
+
+// 文件大小限制 (10MB)
+const MAX_FILE_SIZE = 10 * 1024 * 1024
+
+// 处理文件选择
+function handleConfigFileSelect(event: Event) {
+  const input = event.target as HTMLInputElement
+  const file = input.files?.[0]
+  if (!file) return
+
+  if (file.size > MAX_FILE_SIZE) {
+    error('文件大小不能超过 10MB')
+    input.value = ''
+    return
+  }
+
+  const reader = new FileReader()
+  reader.onload = (e) => {
+    try {
+      const content = e.target?.result as string
+      const data = JSON.parse(content) as ConfigExportData
+
+      // 验证版本
+      if (data.version !== '1.0') {
+        error(`不支持的配置版本: ${data.version}`)
+        return
+      }
+
+      importPreview.value = data
+      mergeMode.value = 'skip'
+      importDialogOpen.value = true
+    } catch (err) {
+      error('解析配置文件失败，请确保是有效的 JSON 文件')
+      log.error('解析配置文件失败:', err)
+    }
+  }
+  reader.readAsText(file)
+
+  // 重置 input 以便能再次选择同一文件
+  input.value = ''
+}
+
+// 确认导入
+async function confirmImport() {
+  if (!importPreview.value) return
+
+  importLoading.value = true
+  try {
+    const result = await adminApi.importConfig({
+      ...importPreview.value,
+      merge_mode: mergeMode.value
+    })
+    importResult.value = result
+    importDialogOpen.value = false
+    mergeModeSelectOpen.value = false
+    importResultDialogOpen.value = true
+    success('配置导入成功')
+  } catch (err: any) {
+    error(err.response?.data?.detail || '导入配置失败')
+    log.error('导入配置失败:', err)
+  } finally {
+    importLoading.value = false
+  }
+}
+
+// 导出用户数据
+async function handleExportUsers() {
+  exportUsersLoading.value = true
+  try {
+    const data = await adminApi.exportUsers()
+    const blob = new Blob([JSON.stringify(data, null, 2)], { type: 'application/json' })
+    const url = URL.createObjectURL(blob)
+    const a = document.createElement('a')
+    a.href = url
+    a.download = `aether-users-${new Date().toISOString().slice(0, 10)}.json`
+    document.body.appendChild(a)
+    a.click()
+    document.body.removeChild(a)
+    URL.revokeObjectURL(url)
+    success('用户数据已导出')
+  } catch (err) {
+    error('导出用户数据失败')
+    log.error('导出用户数据失败:', err)
+  } finally {
+    exportUsersLoading.value = false
+  }
+}
+
+// 触发用户数据文件选择
+function triggerUsersFileSelect() {
+  usersFileInput.value?.click()
+}
+
+// 处理用户数据文件选择
+function handleUsersFileSelect(event: Event) {
+  const input = event.target as HTMLInputElement
+  const file = input.files?.[0]
+  if (!file) return
+
+  if (file.size > MAX_FILE_SIZE) {
+    error('文件大小不能超过 10MB')
+    input.value = ''
+    return
+  }
+
+  const reader = new FileReader()
+  reader.onload = (e) => {
+    try {
+      const content = e.target?.result as string
+      const data = JSON.parse(content) as UsersExportData
+
+      // 验证版本
+      if (data.version !== '1.0') {
+        error(`不支持的配置版本: ${data.version}`)
+        return
+      }
+
+      importUsersPreview.value = data
+      usersMergeMode.value = 'skip'
+      importUsersDialogOpen.value = true
+    } catch (err) {
+      error('解析用户数据文件失败，请确保是有效的 JSON 文件')
+      log.error('解析用户数据文件失败:', err)
+    }
+  }
+  reader.readAsText(file)
+
+  // 重置 input 以便能再次选择同一文件
+  input.value = ''
+}
+
+// 确认导入用户数据
+async function confirmImportUsers() {
+  if (!importUsersPreview.value) return
+
+  importUsersLoading.value = true
+  try {
+    const result = await adminApi.importUsers({
+      ...importUsersPreview.value,
+      merge_mode: usersMergeMode.value
+    })
+    importUsersResult.value = result
+    importUsersDialogOpen.value = false
+    usersMergeModeSelectOpen.value = false
+    importUsersResultDialogOpen.value = true
+    success('用户数据导入成功')
+  } catch (err: any) {
+    error(err.response?.data?.detail || '导入用户数据失败')
+    log.error('导入用户数据失败:', err)
+  } finally {
+    importUsersLoading.value = false
+  }
+}
 </script>

frontend/src/views/admin/Users.vue

@@ -791,11 +791,13 @@ const filteredUsers = computed(() => {
     return new Date(b.created_at).getTime() - new Date(a.created_at).getTime()
   })
 
+  // 搜索（支持空格分隔的多关键词 AND 搜索）
   if (searchQuery.value) {
-    const query = searchQuery.value.toLowerCase()
-    filtered = filtered.filter(
-      u => u.username.toLowerCase().includes(query) || u.email?.toLowerCase().includes(query)
-    )
+    const keywords = searchQuery.value.toLowerCase().split(/\s+/).filter(k => k.length > 0)
+    filtered = filtered.filter(u => {
+      const searchableText = `${u.username} ${u.email || ''}`.toLowerCase()
+      return keywords.every(keyword => searchableText.includes(keyword))
+    })
   }
 
   if (filterRole.value !== 'all') {

frontend/src/views/shared/Dashboard.vue

@@ -103,7 +103,7 @@
           </div>
           <div class="grid grid-cols-2 gap-2 sm:gap-3 xl:grid-cols-4">
             <Card class="relative p-3 sm:p-4 border-book-cloth/30">
-              <Clock class="absolute top-3 right-3 h-3.5 w-3.5 sm:h-4 sm:w-4 text-book-cloth" />
+              <Clock class="absolute top-3 right-3 h-3.5 w-3.5 sm:h-4 sm:w-4 text-muted-foreground" />
               <div class="pr-6">
                 <p class="text-[9px] sm:text-[10px] font-semibold uppercase tracking-[0.2em] sm:tracking-[0.3em] text-muted-foreground">
                   平均响应
@@ -114,7 +114,7 @@
               </div>
             </Card>
             <Card class="relative p-3 sm:p-4 border-kraft/30">
-              <AlertTriangle class="absolute top-3 right-3 h-3.5 w-3.5 sm:h-4 sm:w-4 text-kraft" />
+              <AlertTriangle class="absolute top-3 right-3 h-3.5 w-3.5 sm:h-4 sm:w-4 text-muted-foreground" />
               <div class="pr-6">
                 <p class="text-[9px] sm:text-[10px] font-semibold uppercase tracking-[0.2em] sm:tracking-[0.3em] text-muted-foreground">
                   错误率
@@ -128,7 +128,7 @@
               </div>
             </Card>
             <Card class="relative p-3 sm:p-4 border-book-cloth/25">
-              <Shuffle class="absolute top-3 right-3 h-3.5 w-3.5 sm:h-4 sm:w-4 text-kraft" />
+              <Shuffle class="absolute top-3 right-3 h-3.5 w-3.5 sm:h-4 sm:w-4 text-muted-foreground" />
               <div class="pr-6">
                 <p class="text-[9px] sm:text-[10px] font-semibold uppercase tracking-[0.2em] sm:tracking-[0.3em] text-muted-foreground">
                   转移次数
@@ -142,7 +142,7 @@
               v-if="costStats"
               class="relative p-3 sm:p-4 border-manilla/40"
             >
-              <DollarSign class="absolute top-3 right-3 h-3.5 w-3.5 sm:h-4 sm:w-4 text-book-cloth" />
+              <DollarSign class="absolute top-3 right-3 h-3.5 w-3.5 sm:h-4 sm:w-4 text-muted-foreground" />
               <div class="pr-6">
                 <p class="text-[9px] sm:text-[10px] font-semibold uppercase tracking-[0.2em] sm:tracking-[0.3em] text-muted-foreground">
                   实际成本
@@ -180,7 +180,7 @@
           </div>
           <div class="grid grid-cols-2 gap-2 sm:gap-3 xl:grid-cols-4">
             <Card class="relative p-3 sm:p-4 border-book-cloth/30">
-              <Database class="absolute top-3 right-3 h-3.5 w-3.5 sm:h-4 sm:w-4 text-book-cloth" />
+              <Database class="absolute top-3 right-3 h-3.5 w-3.5 sm:h-4 sm:w-4 text-muted-foreground" />
               <div class="pr-6">
                 <p class="text-[9px] sm:text-[10px] font-semibold uppercase tracking-[0.2em] sm:tracking-[0.3em] text-muted-foreground">
                   缓存命中率
@@ -191,7 +191,7 @@
               </div>
             </Card>
             <Card class="relative p-3 sm:p-4 border-kraft/30">
-              <Hash class="absolute top-3 right-3 h-3.5 w-3.5 sm:h-4 sm:w-4 text-kraft" />
+              <Hash class="absolute top-3 right-3 h-3.5 w-3.5 sm:h-4 sm:w-4 text-muted-foreground" />
               <div class="pr-6">
                 <p class="text-[9px] sm:text-[10px] font-semibold uppercase tracking-[0.2em] sm:tracking-[0.3em] text-muted-foreground">
                   缓存读取
@@ -202,7 +202,7 @@
               </div>
             </Card>
             <Card class="relative p-3 sm:p-4 border-book-cloth/25">
-              <Database class="absolute top-3 right-3 h-3.5 w-3.5 sm:h-4 sm:w-4 text-kraft" />
+              <Database class="absolute top-3 right-3 h-3.5 w-3.5 sm:h-4 sm:w-4 text-muted-foreground" />
               <div class="pr-6">
                 <p class="text-[9px] sm:text-[10px] font-semibold uppercase tracking-[0.2em] sm:tracking-[0.3em] text-muted-foreground">
                   缓存创建
@@ -216,7 +216,7 @@
               v-if="tokenBreakdown"
               class="relative p-3 sm:p-4 border-manilla/40"
             >
-              <Hash class="absolute top-3 right-3 h-3.5 w-3.5 sm:h-4 sm:w-4 text-book-cloth" />
+              <Hash class="absolute top-3 right-3 h-3.5 w-3.5 sm:h-4 sm:w-4 text-muted-foreground" />
               <div class="pr-6">
                 <p class="text-[9px] sm:text-[10px] font-semibold uppercase tracking-[0.2em] sm:tracking-[0.3em] text-muted-foreground">
                   总Token
@@ -254,16 +254,16 @@
         <Card class="overflow-hidden p-4 flex flex-col flex-1 min-h-0 h-full max-h-[280px] sm:max-h-none">
           <div
             v-if="loadingAnnouncements"
-            class="py-8 text-center"
+            class="flex-1 flex items-center justify-center"
           >
-            <Loader2 class="h-5 w-5 animate-spin mx-auto text-muted-foreground" />
+            <Loader2 class="h-5 w-5 animate-spin text-muted-foreground" />
           </div>
 
           <div
             v-else-if="announcements.length === 0"
-            class="py-8 text-center"
+            class="flex-1 flex flex-col items-center justify-center"
           >
-            <Bell class="h-8 w-8 mx-auto text-muted-foreground/40" />
+            <Bell class="h-8 w-8 text-muted-foreground/40" />
             <p class="mt-2 text-xs text-muted-foreground">
               暂无公告
             </p>
@@ -793,9 +793,8 @@ const statCardGlows = [
   'bg-kraft/30'
 ]
 
-const getStatIconColor = (index: number): string => {
-  const colors = ['text-book-cloth', 'text-kraft', 'text-book-cloth', 'text-kraft']
-  return colors[index % colors.length]
+const getStatIconColor = (_index: number): string => {
+  return 'text-muted-foreground'
 }
 
 // 统计数据

frontend/src/views/shared/Usage.vue

@@ -65,6 +65,7 @@
       :page-size="pageSize"
       :total-records="totalRecords"
       :page-size-options="pageSizeOptions"
+      :auto-refresh="globalAutoRefresh"
       @update:selected-period="handlePeriodChange"
       @update:filter-user="handleFilterUserChange"
       @update:filter-model="handleFilterModelChange"
@@ -72,6 +73,7 @@
       @update:filter-status="handleFilterStatusChange"
       @update:current-page="handlePageChange"
       @update:page-size="handlePageSizeChange"
+      @update:auto-refresh="handleAutoRefreshChange"
       @refresh="refreshData"
       @export="exportData"
       @show-detail="showRequestDetail"
@@ -214,7 +216,10 @@ const hasActiveRequests = computed(() => activeRequestIds.value.length > 0)
 
 // 自动刷新定时器
 let autoRefreshTimer: ReturnType<typeof setInterval> | null = null
-const AUTO_REFRESH_INTERVAL = 1000 // 1秒刷新一次
+let globalAutoRefreshTimer: ReturnType<typeof setInterval> | null = null
+const AUTO_REFRESH_INTERVAL = 1000 // 1秒刷新一次（用于活跃请求）
+const GLOBAL_AUTO_REFRESH_INTERVAL = 10000 // 10秒刷新一次（全局自动刷新）
+const globalAutoRefresh = ref(false) // 全局自动刷新开关
 
 // 轮询活跃请求状态（轻量级，只更新状态变化的记录）
 async function pollActiveRequests() {
@@ -278,9 +283,34 @@ watch(hasActiveRequests, (hasActive) => {
   }
 }, { immediate: true })
 
+// 启动全局自动刷新
+function startGlobalAutoRefresh() {
+  if (globalAutoRefreshTimer) return
+  globalAutoRefreshTimer = setInterval(refreshData, GLOBAL_AUTO_REFRESH_INTERVAL)
+}
+
+// 停止全局自动刷新
+function stopGlobalAutoRefresh() {
+  if (globalAutoRefreshTimer) {
+    clearInterval(globalAutoRefreshTimer)
+    globalAutoRefreshTimer = null
+  }
+}
+
+// 处理自动刷新开关变化
+function handleAutoRefreshChange(value: boolean) {
+  globalAutoRefresh.value = value
+  if (value) {
+    startGlobalAutoRefresh()
+  } else {
+    stopGlobalAutoRefresh()
+  }
+}
+
 // 组件卸载时清理定时器
 onUnmounted(() => {
   stopAutoRefresh()
+  stopGlobalAutoRefresh()
 })
 
 // 用户页面的前端分页

frontend/src/views/user/Announcements.vue

@@ -226,8 +226,8 @@
           <div
             v-for="announcement in announcements"
             :key="announcement.id"
+            class="p-4 space-y-2 cursor-pointer transition-colors"
             :class="[
-              'p-4 space-y-2 cursor-pointer transition-colors',
               announcement.is_read ? 'hover:bg-muted/30' : 'bg-primary/5 hover:bg-primary/10'
             ]"
             @click="viewAnnouncementDetail(announcement)"

frontend/src/views/user/ModelCatalog.vue

@@ -165,17 +165,17 @@
                 <TableCell class="py-4">
                   <div class="flex gap-1.5">
                     <Eye
-                      v-if="model.default_supports_vision"
+                      v-if="model.config?.vision === true"
                       class="w-4 h-4 text-muted-foreground"
                       title="Vision"
                     />
                     <Wrench
-                      v-if="model.default_supports_function_calling"
+                      v-if="model.config?.function_calling === true"
                       class="w-4 h-4 text-muted-foreground"
                       title="Tool Use"
                     />
                     <Brain
-                      v-if="model.default_supports_extended_thinking"
+                      v-if="model.config?.extended_thinking === true"
                       class="w-4 h-4 text-muted-foreground"
                       title="Extended Thinking"
                     />
@@ -253,15 +253,15 @@
             <!-- 第二行：能力图标 -->
             <div class="flex gap-1.5">
               <Eye
-                v-if="model.default_supports_vision"
+                v-if="model.config?.vision === true"
                 class="w-4 h-4 text-muted-foreground"
               />
               <Wrench
-                v-if="model.default_supports_function_calling"
+                v-if="model.config?.function_calling === true"
                 class="w-4 h-4 text-muted-foreground"
               />
               <Brain
-                v-if="model.default_supports_extended_thinking"
+                v-if="model.config?.extended_thinking === true"
                 class="w-4 h-4 text-muted-foreground"
               />
             </div>
@@ -474,24 +474,24 @@ async function toggleCapability(modelName: string, capName: string) {
 const filteredModels = computed(() => {
   let result = models.value
 
-  // 搜索
+  // 搜索（支持空格分隔的多关键词 AND 搜索）
   if (searchQuery.value) {
-    const query = searchQuery.value.toLowerCase()
-    result = result.filter(m =>
-      m.name.toLowerCase().includes(query) ||
-      m.display_name?.toLowerCase().includes(query)
-    )
+    const keywords = searchQuery.value.toLowerCase().split(/\s+/).filter(k => k.length > 0)
+    result = result.filter(m => {
+      const searchableText = `${m.name} ${m.display_name || ''}`.toLowerCase()
+      return keywords.every(keyword => searchableText.includes(keyword))
+    })
   }
 
   // 能力筛选
   if (capabilityFilters.value.vision) {
-    result = result.filter(m => m.default_supports_vision)
+    result = result.filter(m => m.config?.vision === true)
   }
   if (capabilityFilters.value.toolUse) {
-    result = result.filter(m => m.default_supports_function_calling)
+    result = result.filter(m => m.config?.function_calling === true)
   }
   if (capabilityFilters.value.extendedThinking) {
-    result = result.filter(m => m.default_supports_extended_thinking)
+    result = result.filter(m => m.config?.extended_thinking === true)
   }
 
   return result

frontend/src/views/user/Settings.vue

@@ -62,6 +62,7 @@
             <Button
               type="submit"
               :disabled="savingProfile"
+              class="shadow-none hover:shadow-none"
             >
               {{ savingProfile ? '保存中...' : '保存修改' }}
             </Button>
@@ -107,6 +108,7 @@
             <Button
               type="submit"
               :disabled="changingPassword"
+              class="shadow-none hover:shadow-none"
             >
               {{ changingPassword ? '修改中...' : '修改密码' }}
             </Button>
@@ -320,6 +322,7 @@
 import { ref, onMounted } from 'vue'
 import { useAuthStore } from '@/stores/auth'
 import { meApi, type Profile } from '@/api/me'
+import { useDarkMode, type ThemeMode } from '@/composables/useDarkMode'
 import Card from '@/components/ui/card.vue'
 import Button from '@/components/ui/button.vue'
 import Badge from '@/components/ui/badge.vue'
@@ -338,6 +341,7 @@ import { log } from '@/utils/logger'
 
 const authStore = useAuthStore()
 const { success, error: showError } = useToast()
+const { setThemeMode } = useDarkMode()
 
 const profile = ref<Profile | null>(null)
 
@@ -375,20 +379,8 @@ function handleThemeChange(value: string) {
   themeSelectOpen.value = false
   updatePreferences()
 
-  // 应用主题
-  if (value === 'dark') {
-    document.documentElement.classList.add('dark')
-  } else if (value === 'light') {
-    document.documentElement.classList.remove('dark')
-  } else {
-    // system: 跟随系统
-    const prefersDark = window.matchMedia('(prefers-color-scheme: dark)').matches
-    if (prefersDark) {
-      document.documentElement.classList.add('dark')
-    } else {
-      document.documentElement.classList.remove('dark')
-    }
-  }
+  // 使用 useDarkMode 统一切换主题
+  setThemeMode(value as ThemeMode)
 }
 
 function handleLanguageChange(value: string) {
@@ -418,10 +410,16 @@ async function loadProfile() {
 async function loadPreferences() {
   try {
     const prefs = await meApi.getPreferences()
+
+    // 主题以本地 localStorage 为准（useDarkMode 在应用启动时已初始化）
+    // 这样可以避免刷新页面时主题被服务端旧值覆盖
+    const { themeMode: currentThemeMode } = useDarkMode()
+    const localTheme = currentThemeMode.value
+
     preferencesForm.value = {
       avatar_url: prefs.avatar_url || '',
       bio: prefs.bio || '',
-      theme: prefs.theme || 'light',
+      theme: localTheme,  // 使用本地主题，而非服务端返回值
       language: prefs.language || 'zh-CN',
       timezone: prefs.timezone || 'Asia/Shanghai',
       notifications: {
@@ -431,11 +429,12 @@ async function loadPreferences() {
       }
     }
 
-    // 应用主题
-    if (preferencesForm.value.theme === 'dark') {
-      document.documentElement.classList.add('dark')
-    } else if (preferencesForm.value.theme === 'light') {
-      document.documentElement.classList.remove('dark')
+    // 如果本地主题和服务端不一致，同步到服务端（静默更新，不提示用户）
+    const serverTheme = prefs.theme || 'light'
+    if (localTheme !== serverTheme) {
+      meApi.updatePreferences({ theme: localTheme }).catch(() => {
+        // 静默失败，不影响用户体验
+      })
     }
   } catch (error) {
     log.error('加载偏好设置失败:', error)

frontend/src/views/user/components/UserModelDetailDrawer.vue

@@ -38,10 +38,10 @@
                   </button>
                 </div>
                 <p
-                  v-if="model.description"
+                  v-if="model.config?.description"
                   class="text-xs text-muted-foreground"
                 >
-                  {{ model.description }}
+                  {{ model.config?.description }}
                 </p>
               </div>
               <Button
@@ -73,10 +73,10 @@
                     </p>
                   </div>
                   <Badge
-                    :variant="model.default_supports_streaming ?? false ? 'default' : 'secondary'"
+                    :variant="model.config?.streaming !== false ? 'default' : 'secondary'"
                     class="text-xs"
                   >
-                    {{ model.default_supports_streaming ?? false ? '支持' : '不支持' }}
+                    {{ model.config?.streaming !== false ? '支持' : '不支持' }}
                   </Badge>
                 </div>
                 <div class="flex items-center gap-2 p-3 rounded-lg border">
@@ -90,10 +90,10 @@
                     </p>
                   </div>
                   <Badge
-                    :variant="model.default_supports_image_generation ?? false ? 'default' : 'secondary'"
+                    :variant="model.config?.image_generation === true ? 'default' : 'secondary'"
                     class="text-xs"
                   >
-                    {{ model.default_supports_image_generation ?? false ? '支持' : '不支持' }}
+                    {{ model.config?.image_generation === true ? '支持' : '不支持' }}
                   </Badge>
                 </div>
                 <div class="flex items-center gap-2 p-3 rounded-lg border">
@@ -107,10 +107,10 @@
                     </p>
                   </div>
                   <Badge
-                    :variant="model.default_supports_vision ?? false ? 'default' : 'secondary'"
+                    :variant="model.config?.vision === true ? 'default' : 'secondary'"
                     class="text-xs"
                   >
-                    {{ model.default_supports_vision ?? false ? '支持' : '不支持' }}
+                    {{ model.config?.vision === true ? '支持' : '不支持' }}
                   </Badge>
                 </div>
                 <div class="flex items-center gap-2 p-3 rounded-lg border">
@@ -124,10 +124,10 @@
                     </p>
                   </div>
                   <Badge
-                    :variant="model.default_supports_function_calling ?? false ? 'default' : 'secondary'"
+                    :variant="model.config?.function_calling === true ? 'default' : 'secondary'"
                     class="text-xs"
                   >
-                    {{ model.default_supports_function_calling ?? false ? '支持' : '不支持' }}
+                    {{ model.config?.function_calling === true ? '支持' : '不支持' }}
                   </Badge>
                 </div>
                 <div class="flex items-center gap-2 p-3 rounded-lg border">
@@ -141,10 +141,10 @@
                     </p>
                   </div>
                   <Badge
-                    :variant="model.default_supports_extended_thinking ?? false ? 'default' : 'secondary'"
+                    :variant="model.config?.extended_thinking === true ? 'default' : 'secondary'"
                     class="text-xs"
                   >
-                    {{ model.default_supports_extended_thinking ?? false ? '支持' : '不支持' }}
+                    {{ model.config?.extended_thinking === true ? '支持' : '不支持' }}
                   </Badge>
                 </div>
               </div>
@@ -350,6 +350,7 @@ import {
   Layers,
   Image as ImageIcon
 } from 'lucide-vue-next'
+import { useEscapeKey } from '@/composables/useEscapeKey'
 import { useToast } from '@/composables/useToast'
 import Card from '@/components/ui/card.vue'
 import Badge from '@/components/ui/badge.vue'
@@ -453,6 +454,16 @@ function getFirst1hCachePrice(tieredPricing: TieredPricingConfig | undefined | n
   if (!tieredPricing?.tiers?.length) return '-'
   return get1hCachePrice(tieredPricing.tiers[0])
 }
+
+// 添加 ESC 键监听
+useEscapeKey(() => {
+  if (props.open) {
+    handleClose()
+  }
+}, {
+  disableOnInput: true,
+  once: false
+})
 </script>
 
 <style scoped>

migrate.sh

@@ -0,0 +1,12 @@
+#!/bin/bash
+# 数据库迁移脚本 - 在 Docker 容器内执行 Alembic 迁移
+
+set -e
+
+CONTAINER_NAME="aether-app"
+
+echo "Running database migrations in container: $CONTAINER_NAME"
+
+docker exec $CONTAINER_NAME alembic upgrade head
+
+echo "Database migration completed successfully"

src/__init__.py

@@ -3,10 +3,8 @@
 A proxy server that enables AI models to work with multiple API providers.
 """
 
-from dotenv import load_dotenv
-
-# Load environment variables from .env file
-load_dotenv()
+# 注意: dotenv 加载已统一移至 src/config/settings.py
+# 不要在此处重复加载
 
 try:
     from src._version import __version__

src/api/admin/__init__.py

@@ -7,6 +7,7 @@ from .api_keys import router as api_keys_router
 from .endpoints import router as endpoints_router
 from .models import router as models_router
 from .monitoring import router as monitoring_router
+from .provider_query import router as provider_query_router
 from .provider_strategy import router as provider_strategy_router
 from .providers import router as providers_router
 from .security import router as security_router
@@ -26,5 +27,6 @@ router.include_router(provider_strategy_router)
 router.include_router(adaptive_router)
 router.include_router(models_router)
 router.include_router(security_router)
+router.include_router(provider_query_router)
 
 __all__ = ["router"]

src/api/admin/api_keys/routes.py

@@ -223,7 +223,7 @@ class AdminCreateStandaloneKeyAdapter(AdminApiAdapter):
             allowed_providers=self.key_data.allowed_providers,
             allowed_api_formats=self.key_data.allowed_api_formats,
             allowed_models=self.key_data.allowed_models,
-            rate_limit=self.key_data.rate_limit or 100,
+            rate_limit=self.key_data.rate_limit,  # None 表示不限制
             expire_days=self.key_data.expire_days,
             initial_balance_usd=self.key_data.initial_balance_usd,
             is_standalone=True,  # 标记为独立Key

src/api/admin/endpoints/routes.py

@@ -5,7 +5,7 @@ ProviderEndpoint CRUD 管理 API
 import uuid
 from dataclasses import dataclass
 from datetime import datetime, timezone
-from typing import List
+from typing import List, Optional
 
 from fastapi import APIRouter, Depends, Query, Request
 from sqlalchemy import and_, func
@@ -27,6 +27,16 @@ router = APIRouter(tags=["Endpoint Management"])
 pipeline = ApiRequestPipeline()
 
 
+def mask_proxy_password(proxy_config: Optional[dict]) -> Optional[dict]:
+    """对代理配置中的密码进行脱敏处理"""
+    if not proxy_config:
+        return None
+    masked = dict(proxy_config)
+    if masked.get("password"):
+        masked["password"] = "***"
+    return masked
+
+
 @router.get("/providers/{provider_id}/endpoints", response_model=List[ProviderEndpointResponse])
 async def list_provider_endpoints(
     provider_id: str,
@@ -153,6 +163,7 @@ class AdminListProviderEndpointsAdapter(AdminApiAdapter):
                 "api_format": endpoint.api_format,
                 "total_keys": total_keys_map.get(endpoint.id, 0),
                 "active_keys": active_keys_map.get(endpoint.id, 0),
+                "proxy": mask_proxy_password(endpoint.proxy),
             }
             endpoint_dict.pop("_sa_instance_state", None)
             result.append(ProviderEndpointResponse(**endpoint_dict))
@@ -202,6 +213,7 @@ class AdminCreateProviderEndpointAdapter(AdminApiAdapter):
             rate_limit=self.endpoint_data.rate_limit,
             is_active=True,
             config=self.endpoint_data.config,
+            proxy=self.endpoint_data.proxy.model_dump() if self.endpoint_data.proxy else None,
             created_at=now,
             updated_at=now,
         )
@@ -215,12 +227,13 @@ class AdminCreateProviderEndpointAdapter(AdminApiAdapter):
         endpoint_dict = {
             k: v
             for k, v in new_endpoint.__dict__.items()
-            if k not in {"api_format", "_sa_instance_state"}
+            if k not in {"api_format", "_sa_instance_state", "proxy"}
         }
         return ProviderEndpointResponse(
             **endpoint_dict,
             provider_name=provider.name,
             api_format=new_endpoint.api_format,
+            proxy=mask_proxy_password(new_endpoint.proxy),
             total_keys=0,
             active_keys=0,
         )
@@ -259,12 +272,13 @@ class AdminGetProviderEndpointAdapter(AdminApiAdapter):
         endpoint_dict = {
             k: v
             for k, v in endpoint_obj.__dict__.items()
-            if k not in {"api_format", "_sa_instance_state"}
+            if k not in {"api_format", "_sa_instance_state", "proxy"}
         }
         return ProviderEndpointResponse(
             **endpoint_dict,
             provider_name=provider.name,
             api_format=endpoint_obj.api_format,
+            proxy=mask_proxy_password(endpoint_obj.proxy),
             total_keys=total_keys,
             active_keys=active_keys,
         )
@@ -284,6 +298,17 @@ class AdminUpdateProviderEndpointAdapter(AdminApiAdapter):
             raise NotFoundException(f"Endpoint {self.endpoint_id} 不存在")
 
         update_data = self.endpoint_data.model_dump(exclude_unset=True)
+        # 把 proxy 转换为 dict 存储，支持显式设置为 None 清除代理
+        if "proxy" in update_data:
+            if update_data["proxy"] is not None:
+                new_proxy = dict(update_data["proxy"])
+                # 只有当密码字段未提供时才保留原密码（空字符串视为显式清除）
+                if "password" not in new_proxy and endpoint.proxy:
+                    old_password = endpoint.proxy.get("password")
+                    if old_password:
+                        new_proxy["password"] = old_password
+                update_data["proxy"] = new_proxy
+            # proxy 为 None 时保留，用于清除代理配置
         for field, value in update_data.items():
             setattr(endpoint, field, value)
         endpoint.updated_at = datetime.now(timezone.utc)
@@ -311,12 +336,13 @@ class AdminUpdateProviderEndpointAdapter(AdminApiAdapter):
         endpoint_dict = {
             k: v
             for k, v in endpoint.__dict__.items()
-            if k not in {"api_format", "_sa_instance_state"}
+            if k not in {"api_format", "_sa_instance_state", "proxy"}
         }
         return ProviderEndpointResponse(
             **endpoint_dict,
             provider_name=provider.name if provider else "Unknown",
             api_format=endpoint.api_format,
+            proxy=mask_proxy_password(endpoint.proxy),
             total_keys=total_keys,
             active_keys=active_keys,
         )

src/api/admin/models/__init__.py

@@ -5,12 +5,12 @@
 from fastapi import APIRouter
 
 from .catalog import router as catalog_router
+from .external import router as external_router
 from .global_models import router as global_models_router
-from .mappings import router as mappings_router
 
 router = APIRouter(prefix="/api/admin/models", tags=["Admin - Model Management"])
 
 # 挂载子路由
 router.include_router(catalog_router)
 router.include_router(global_models_router)
-router.include_router(mappings_router)
+router.include_router(external_router)

src/api/admin/models/catalog.py

@@ -1,38 +1,26 @@
 """
 统一模型目录 Admin API
 
-阶段一：基于 ModelMapping 和 Model 的聚合视图
+基于 GlobalModel 的聚合视图
 """
 
 from dataclasses import dataclass
-from typing import Dict, List, Optional, Set
+from typing import Dict, List
 
-from fastapi import APIRouter, Depends, HTTPException, Request
-from sqlalchemy import func, or_
+from fastapi import APIRouter, Depends, Request
 from sqlalchemy.orm import Session, joinedload
 
 from src.api.base.admin_adapter import AdminApiAdapter
 from src.api.base.pipeline import ApiRequestPipeline
-from src.core.logger import logger
 from src.database import get_db
-from src.models.database import GlobalModel, Model, ModelMapping, Provider
+from src.models.database import GlobalModel, Model
 from src.models.pydantic_models import (
-    BatchAssignError,
-    BatchAssignModelMappingRequest,
-    BatchAssignModelMappingResponse,
-    BatchAssignProviderResult,
-    DeleteModelMappingResponse,
     ModelCapabilities,
     ModelCatalogItem,
     ModelCatalogProviderDetail,
     ModelCatalogResponse,
     ModelPriceRange,
-    OrphanedModel,
-    UpdateModelMappingRequest,
-    UpdateModelMappingResponse,
 )
-from src.services.cache.invalidation import get_cache_invalidation_service
-from src.services.model.service import ModelService
 
 router = APIRouter(prefix="/catalog", tags=["Admin - Model Catalog"])
 pipeline = ApiRequestPipeline()
@@ -47,24 +35,13 @@ async def get_model_catalog(
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
-@router.post("/batch-assign", response_model=BatchAssignModelMappingResponse)
-async def batch_assign_model_mappings(
-    request: Request,
-    payload: BatchAssignModelMappingRequest,
-    db: Session = Depends(get_db),
-) -> BatchAssignModelMappingResponse:
-    adapter = AdminBatchAssignModelMappingsAdapter(payload=payload)
-    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
-
-
 @dataclass
 class AdminGetModelCatalogAdapter(AdminApiAdapter):
     """管理员查询统一模型目录
 
-    新架构说明：
+    架构说明：
     1. 以 GlobalModel 为中心聚合数据
-    2. ModelMapping 表提供别名信息（provider_id=NULL 表示全局）
-    3. Model 表提供关联提供商和价格
+    2. Model 表提供关联提供商和价格
     """
 
     async def handle(self, context):  # type: ignore[override]
@@ -75,29 +52,7 @@ class AdminGetModelCatalogAdapter(AdminApiAdapter):
             db.query(GlobalModel).filter(GlobalModel.is_active == True).all()
         )
 
-        # 2. 获取所有活跃的别名（含全局和 Provider 特定）
-        aliases_rows: List[ModelMapping] = (
-            db.query(ModelMapping)
-            .options(joinedload(ModelMapping.target_global_model))
-            .filter(
-                ModelMapping.is_active == True,
-                ModelMapping.provider_id.is_(None),
-            )
-            .all()
-        )
-
-        # 按 GlobalModel ID 组织别名
-        aliases_by_global_model: Dict[str, List[str]] = {}
-        for alias_row in aliases_rows:
-            if not alias_row.target_global_model_id:
-                continue
-            gm_id = alias_row.target_global_model_id
-            if gm_id not in aliases_by_global_model:
-                aliases_by_global_model[gm_id] = []
-            if alias_row.source_model not in aliases_by_global_model[gm_id]:
-                aliases_by_global_model[gm_id].append(alias_row.source_model)
-
-        # 3. 获取所有活跃的 Model 实现（包含 global_model 以便计算有效价格）
+        # 2. 获取所有活跃的 Model 实现（包含 global_model 以便计算有效价格）
         models: List[Model] = (
             db.query(Model)
             .options(joinedload(Model.provider), joinedload(Model.global_model))
@@ -111,16 +66,18 @@ class AdminGetModelCatalogAdapter(AdminApiAdapter):
             if model.global_model_id:
                 models_by_global_model.setdefault(model.global_model_id, []).append(model)
 
-        # 4. 为每个 GlobalModel 构建 catalog item
+        # 3. 为每个 GlobalModel 构建 catalog item
         catalog_items: List[ModelCatalogItem] = []
 
         for gm in global_models:
             gm_id = gm.id
             provider_entries: List[ModelCatalogProviderDetail] = []
+            # 从 config JSON 读取能力标志
+            gm_config = gm.config or {}
             capability_flags = {
-                "supports_vision": gm.default_supports_vision or False,
-                "supports_function_calling": gm.default_supports_function_calling or False,
-                "supports_streaming": gm.default_supports_streaming or False,
+                "supports_vision": gm_config.get("vision", False),
+                "supports_function_calling": gm_config.get("function_calling", False),
+                "supports_streaming": gm_config.get("streaming", True),
             }
 
             # 遍历该 GlobalModel 的所有关联提供商
@@ -168,7 +125,6 @@ class AdminGetModelCatalogAdapter(AdminApiAdapter):
                         supports_function_calling=model.get_effective_supports_function_calling(),
                         supports_streaming=model.get_effective_supports_streaming(),
                         is_active=bool(model.is_active),
-                        mapping_id=None,  # 新架构中不再有 mapping_id
                     )
                 )
 
@@ -186,8 +142,7 @@ class AdminGetModelCatalogAdapter(AdminApiAdapter):
                 ModelCatalogItem(
                     global_model_name=gm.name,
                     display_name=gm.display_name,
-                    description=gm.description,
-                    aliases=aliases_by_global_model.get(gm_id, []),
+                    description=gm_config.get("description"),
                     providers=provider_entries,
                     price_range=price_range,
                     total_providers=len(provider_entries),
@@ -195,238 +150,7 @@ class AdminGetModelCatalogAdapter(AdminApiAdapter):
                 )
             )
 
-        # 5. 查找孤立的别名（别名指向的 GlobalModel 不存在或不活跃）
-        orphaned_rows = (
-            db.query(ModelMapping.source_model, GlobalModel.name, func.count(ModelMapping.id))
-            .outerjoin(GlobalModel, ModelMapping.target_global_model_id == GlobalModel.id)
-            .filter(
-                ModelMapping.is_active == True,
-                ModelMapping.provider_id.is_(None),
-                or_(GlobalModel.id == None, GlobalModel.is_active == False),
-            )
-            .group_by(ModelMapping.source_model, GlobalModel.name)
-            .all()
-        )
-        orphaned_models = [
-            OrphanedModel(alias=row[0], global_model_name=row[1], mapping_count=row[2])
-            for row in orphaned_rows
-            if row[0]
-        ]
-
         return ModelCatalogResponse(
             models=catalog_items,
             total=len(catalog_items),
-            orphaned_models=orphaned_models,
-        )
-
-
-@dataclass
-class AdminBatchAssignModelMappingsAdapter(AdminApiAdapter):
-    payload: BatchAssignModelMappingRequest
-
-    async def handle(self, context):  # type: ignore[override]
-        db: Session = context.db
-        created: List[BatchAssignProviderResult] = []
-        errors: List[BatchAssignError] = []
-
-        for provider_config in self.payload.providers:
-            provider_id = provider_config.provider_id
-            try:
-                provider: Provider = db.query(Provider).filter(Provider.id == provider_id).first()
-                if not provider:
-                    errors.append(
-                        BatchAssignError(provider_id=provider_id, error="Provider 不存在")
-                    )
-                    continue
-
-                model_id: Optional[str] = None
-                created_model = False
-
-                if provider_config.create_model:
-                    model_data = provider_config.model_data
-                    if not model_data:
-                        errors.append(
-                            BatchAssignError(provider_id=provider_id, error="缺少 model_data 配置")
-                        )
-                        continue
-
-                    existing_model = ModelService.get_model_by_name(
-                        db, provider_id, model_data.provider_model_name
-                    )
-                    if existing_model:
-                        model_id = existing_model.id
-                        logger.info("模型 %s 已存在于 Provider %s，复用现有模型",
-                            model_data.provider_model_name,
-                            provider.name,
-                        )
-                    else:
-                        model = ModelService.create_model(db, provider_id, model_data)
-                        model_id = model.id
-                        created_model = True
-                else:
-                    model_id = provider_config.model_id
-                    if not model_id:
-                        errors.append(
-                            BatchAssignError(provider_id=provider_id, error="缺少 model_id")
-                        )
-                        continue
-                    model = (
-                        db.query(Model)
-                        .filter(Model.id == model_id, Model.provider_id == provider_id)
-                        .first()
-                    )
-                    if not model:
-                        errors.append(
-                            BatchAssignError(
-                                provider_id=provider_id, error="模型不存在或不属于当前 Provider")
-                        )
-                        continue
-
-                # 批量分配功能需要适配 GlobalModel 架构
-                # 参见 docs/optimization-backlog.md 中的待办项
-                errors.append(
-                    BatchAssignError(
-                        provider_id=provider_id,
-                        error="批量分配功能暂时不可用，需要适配新的 GlobalModel 架构",
-                    )
-                )
-                continue
-
-            except Exception as exc:
-                db.rollback()
-                logger.error("批量添加模型映射失败（需要适配新架构）")
-                errors.append(BatchAssignError(provider_id=provider_id, error=str(exc)))
-
-        return BatchAssignModelMappingResponse(
-            success=len(created) > 0,
-            created_mappings=created,
-            errors=errors,
-        )
-
-
-@router.put("/mappings/{mapping_id}", response_model=UpdateModelMappingResponse)
-async def update_model_mapping(
-    request: Request,
-    mapping_id: str,
-    payload: UpdateModelMappingRequest,
-    db: Session = Depends(get_db),
-) -> UpdateModelMappingResponse:
-    """更新模型映射"""
-    adapter = AdminUpdateModelMappingAdapter(mapping_id=mapping_id, payload=payload)
-    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
-
-
-@router.delete("/mappings/{mapping_id}", response_model=DeleteModelMappingResponse)
-async def delete_model_mapping(
-    request: Request,
-    mapping_id: str,
-    db: Session = Depends(get_db),
-) -> DeleteModelMappingResponse:
-    """删除模型映射"""
-    adapter = AdminDeleteModelMappingAdapter(mapping_id=mapping_id)
-    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
-
-
-@dataclass
-class AdminUpdateModelMappingAdapter(AdminApiAdapter):
-    """更新模型映射"""
-
-    mapping_id: str
-    payload: UpdateModelMappingRequest
-
-    async def handle(self, context):  # type: ignore[override]
-        db: Session = context.db
-
-        mapping: Optional[ModelMapping] = (
-            db.query(ModelMapping).filter(ModelMapping.id == self.mapping_id).first()
-        )
-
-        if not mapping:
-            raise HTTPException(status_code=404, detail="映射不存在")
-
-        update_data = self.payload.model_dump(exclude_unset=True)
-
-        if "provider_id" in update_data:
-            new_provider_id = update_data["provider_id"]
-            if new_provider_id:
-                provider = db.query(Provider).filter(Provider.id == new_provider_id).first()
-                if not provider:
-                    raise HTTPException(status_code=404, detail="Provider 不存在")
-            mapping.provider_id = new_provider_id
-
-        if "target_global_model_id" in update_data:
-            target_model = (
-                db.query(GlobalModel)
-                .filter(
-                    GlobalModel.id == update_data["target_global_model_id"],
-                    GlobalModel.is_active == True,
-                )
-                .first()
-            )
-            if not target_model:
-                raise HTTPException(status_code=404, detail="目标 GlobalModel 不存在或未激活")
-            mapping.target_global_model_id = update_data["target_global_model_id"]
-
-        if "source_model" in update_data:
-            new_source = update_data["source_model"].strip()
-            if not new_source:
-                raise HTTPException(status_code=400, detail="source_model 不能为空")
-            mapping.source_model = new_source
-
-        if "is_active" in update_data:
-            mapping.is_active = update_data["is_active"]
-
-        duplicate = (
-            db.query(ModelMapping)
-            .filter(
-                ModelMapping.source_model == mapping.source_model,
-                ModelMapping.provider_id == mapping.provider_id,
-                ModelMapping.id != mapping.id,
-            )
-            .first()
-        )
-        if duplicate:
-            raise HTTPException(status_code=400, detail="映射已存在")
-
-        db.commit()
-        db.refresh(mapping)
-
-        cache_service = get_cache_invalidation_service()
-        cache_service.on_model_mapping_changed(mapping.source_model, mapping.provider_id)
-
-        return UpdateModelMappingResponse(
-            success=True,
-            mapping_id=mapping.id,
-            message="映射更新成功",
-        )
-
-
-@dataclass
-class AdminDeleteModelMappingAdapter(AdminApiAdapter):
-    """删除模型映射"""
-
-    mapping_id: str
-
-    async def handle(self, context):  # type: ignore[override]
-        db: Session = context.db
-
-        mapping: Optional[ModelMapping] = (
-            db.query(ModelMapping).filter(ModelMapping.id == self.mapping_id).first()
-        )
-
-        if not mapping:
-            raise HTTPException(status_code=404, detail="映射不存在")
-
-        source_model = mapping.source_model
-        provider_id = mapping.provider_id
-
-        db.delete(mapping)
-        db.commit()
-
-        cache_service = get_cache_invalidation_service()
-        cache_service.on_model_mapping_changed(source_model, provider_id)
-
-        return DeleteModelMappingResponse(
-            success=True,
-            message=f"映射 {self.mapping_id} 已删除",
         )

src/api/admin/models/external.py

@@ -0,0 +1,141 @@
+"""
+models.dev 外部模型数据代理
+"""
+
+import json
+
+from typing import Any, Optional
+
+import httpx
+from fastapi import APIRouter, Depends, HTTPException
+from fastapi.responses import JSONResponse
+
+from src.clients import get_redis_client
+from src.core.logger import logger
+from src.models.database import User
+from src.utils.auth_utils import require_admin
+
+router = APIRouter()
+
+CACHE_KEY = "aether:external:models_dev"
+CACHE_TTL = 15 * 60  # 15 分钟
+
+# 标记官方/一手提供商，前端可据此过滤第三方转售商
+OFFICIAL_PROVIDERS = {
+    "anthropic",  # Claude 官方
+    "openai",  # OpenAI 官方
+    "google",  # Gemini 官方
+    "google-vertex",  # Google Vertex AI
+    "azure",  # Azure OpenAI
+    "amazon-bedrock",  # AWS Bedrock
+    "xai",  # Grok 官方
+    "meta",  # Llama 官方
+    "deepseek",  # DeepSeek 官方
+    "mistral",  # Mistral 官方
+    "cohere",  # Cohere 官方
+    "zhipuai",  # 智谱 AI 官方
+    "alibaba",  # 阿里云（通义千问）
+    "minimax",  # MiniMax 官方
+    "moonshot",  # 月之暗面（Kimi）
+    "baichuan",  # 百川智能
+    "ai21",  # AI21 Labs
+}
+
+
+async def _get_cached_data() -> Optional[dict[str, Any]]:
+    """从 Redis 获取缓存数据"""
+    redis = await get_redis_client()
+    if redis is None:
+        return None
+    try:
+        cached = await redis.get(CACHE_KEY)
+        if cached:
+            result: dict[str, Any] = json.loads(cached)
+            return result
+    except Exception as e:
+        logger.warning(f"读取 models.dev 缓存失败: {e}")
+    return None
+
+
+async def _set_cached_data(data: dict) -> None:
+    """将数据写入 Redis 缓存"""
+    redis = await get_redis_client()
+    if redis is None:
+        return
+    try:
+        await redis.setex(CACHE_KEY, CACHE_TTL, json.dumps(data, ensure_ascii=False))
+    except Exception as e:
+        logger.warning(f"写入 models.dev 缓存失败: {e}")
+
+
+def _mark_official_providers(data: dict[str, Any]) -> dict[str, Any]:
+    """为每个提供商标记是否为官方"""
+    result = {}
+    for provider_id, provider_data in data.items():
+        result[provider_id] = {
+            **provider_data,
+            "official": provider_id in OFFICIAL_PROVIDERS,
+        }
+    return result
+
+
+@router.get("/external")
+async def get_external_models(_: User = Depends(require_admin)) -> JSONResponse:
+    """
+    获取 models.dev 的模型数据（代理请求，解决跨域问题）
+    数据缓存 15 分钟（使用 Redis，多 worker 共享）
+    每个提供商会标记 official 字段，前端可据此过滤
+    """
+    # 检查缓存
+    cached = await _get_cached_data()
+    if cached is not None:
+        # 兼容旧缓存：如果没有 official 字段则补全并回写
+        try:
+            needs_mark = False
+            for provider_data in cached.values():
+                if not isinstance(provider_data, dict) or "official" not in provider_data:
+                    needs_mark = True
+                    break
+            if needs_mark:
+                marked_cached = _mark_official_providers(cached)
+                await _set_cached_data(marked_cached)
+                return JSONResponse(content=marked_cached)
+        except Exception as e:
+            logger.warning(f"处理 models.dev 缓存数据失败，将直接返回原缓存: {e}")
+        return JSONResponse(content=cached)
+
+    # 从 models.dev 获取数据
+    try:
+        async with httpx.AsyncClient(timeout=30.0) as client:
+            response = await client.get("https://models.dev/api.json")
+            response.raise_for_status()
+            data = response.json()
+
+            # 标记官方提供商
+            marked_data = _mark_official_providers(data)
+
+            # 写入缓存
+            await _set_cached_data(marked_data)
+
+            return JSONResponse(content=marked_data)
+    except httpx.TimeoutException:
+        raise HTTPException(status_code=504, detail="请求 models.dev 超时")
+    except httpx.HTTPStatusError as e:
+        raise HTTPException(
+            status_code=502, detail=f"models.dev 返回错误: {e.response.status_code}"
+        )
+    except Exception as e:
+        raise HTTPException(status_code=502, detail=f"获取外部模型数据失败: {str(e)}")
+
+
+@router.delete("/external/cache")
+async def clear_external_models_cache(_: User = Depends(require_admin)) -> dict:
+    """清除 models.dev 缓存"""
+    redis = await get_redis_client()
+    if redis is None:
+        return {"cleared": False, "message": "Redis 未启用"}
+    try:
+        await redis.delete(CACHE_KEY)
+        return {"cleared": True}
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=f"清除缓存失败: {str(e)}")

src/api/admin/models/global_models.py

@@ -123,7 +123,7 @@ class AdminListGlobalModelsAdapter(AdminApiAdapter):
     async def handle(self, context):  # type: ignore[override]
         from sqlalchemy import func
 
-        from src.models.database import Model, ModelMapping
+        from src.models.database import Model
 
         models = GlobalModelService.list_global_models(
             db=context.db,
@@ -144,17 +144,8 @@ class AdminListGlobalModelsAdapter(AdminApiAdapter):
                 or 0
             )
 
-            # 统计别名数量
-            alias_count = (
-                context.db.query(func.count(ModelMapping.id))
-                .filter(ModelMapping.target_global_model_id == gm.id)
-                .scalar()
-                or 0
-            )
-
             response = GlobalModelResponse.model_validate(gm)
             response.provider_count = provider_count
-            response.alias_count = alias_count
             # usage_count 直接从 GlobalModel 表读取，已在 model_validate 中自动映射
             model_responses.append(response)
 
@@ -196,21 +187,15 @@ class AdminCreateGlobalModelAdapter(AdminApiAdapter):
             db=context.db,
             name=self.payload.name,
             display_name=self.payload.display_name,
-            description=self.payload.description,
-            official_url=self.payload.official_url,
-            icon_url=self.payload.icon_url,
             is_active=self.payload.is_active,
             # 按次计费配置
             default_price_per_request=self.payload.default_price_per_request,
             # 阶梯计费配置
             default_tiered_pricing=tiered_pricing_dict,
-            # 默认能力配置
-            default_supports_vision=self.payload.default_supports_vision,
-            default_supports_function_calling=self.payload.default_supports_function_calling,
-            default_supports_streaming=self.payload.default_supports_streaming,
-            default_supports_extended_thinking=self.payload.default_supports_extended_thinking,
             # Key 能力配置
             supported_capabilities=self.payload.supported_capabilities,
+            # 模型配置（JSON）
+            config=self.payload.config,
         )
 
         logger.info(f"GlobalModel 已创建: id={global_model.id} name={global_model.name}")

src/api/admin/models/mappings.py

@@ -1,303 +0,0 @@
-"""模型映射管理 API
-
-提供模型映射的 CRUD 操作。
-
-模型映射（Mapping）用于将源模型映射到目标模型，例如：
-- 请求 gpt-5.1 → Provider A 映射到 gpt-4
-- 用于处理 Provider 不支持请求模型的情况
-
-映射必须关联到特定的 Provider。
-"""
-
-import uuid
-from datetime import datetime, timezone
-from typing import List, Optional
-
-from fastapi import APIRouter, Depends, HTTPException, Query
-from sqlalchemy.orm import Session, joinedload
-
-from src.core.logger import logger
-from src.database import get_db
-from src.models.api import (
-    ModelMappingCreate,
-    ModelMappingResponse,
-    ModelMappingUpdate,
-)
-from src.models.database import GlobalModel, ModelMapping, Provider, User
-from src.services.cache.invalidation import get_cache_invalidation_service
-
-
-router = APIRouter(prefix="/mappings", tags=["Model Mappings"])
-
-
-def _serialize_mapping(mapping: ModelMapping) -> ModelMappingResponse:
-    target = mapping.target_global_model
-    provider = mapping.provider
-    scope = "provider" if mapping.provider_id else "global"
-    return ModelMappingResponse(
-        id=mapping.id,
-        source_model=mapping.source_model,
-        target_global_model_id=mapping.target_global_model_id,
-        target_global_model_name=target.name if target else None,
-        target_global_model_display_name=target.display_name if target else None,
-        provider_id=mapping.provider_id,
-        provider_name=provider.name if provider else None,
-        scope=scope,
-        mapping_type=mapping.mapping_type,
-        is_active=mapping.is_active,
-        created_at=mapping.created_at,
-        updated_at=mapping.updated_at,
-    )
-
-
-@router.get("", response_model=List[ModelMappingResponse])
-async def list_mappings(
-    provider_id: Optional[str] = Query(None, description="按 Provider 筛选"),
-    source_model: Optional[str] = Query(None, description="按源模型名筛选"),
-    target_global_model_id: Optional[str] = Query(None, description="按目标模型筛选"),
-    scope: Optional[str] = Query(None, description="global 或 provider"),
-    mapping_type: Optional[str] = Query(None, description="映射类型: alias 或 mapping"),
-    is_active: Optional[bool] = Query(None, description="按状态筛选"),
-    skip: int = Query(0, ge=0, description="跳过记录数"),
-    limit: int = Query(100, ge=1, le=1000, description="返回记录数"),
-    db: Session = Depends(get_db),
-):
-    """获取模型映射列表"""
-    query = db.query(ModelMapping).options(
-        joinedload(ModelMapping.target_global_model),
-        joinedload(ModelMapping.provider),
-    )
-
-    if provider_id is not None:
-        query = query.filter(ModelMapping.provider_id == provider_id)
-    if scope == "global":
-        query = query.filter(ModelMapping.provider_id.is_(None))
-    elif scope == "provider":
-        query = query.filter(ModelMapping.provider_id.isnot(None))
-    if mapping_type is not None:
-        query = query.filter(ModelMapping.mapping_type == mapping_type)
-    if source_model:
-        query = query.filter(ModelMapping.source_model.ilike(f"%{source_model}%"))
-    if target_global_model_id is not None:
-        query = query.filter(ModelMapping.target_global_model_id == target_global_model_id)
-    if is_active is not None:
-        query = query.filter(ModelMapping.is_active == is_active)
-
-    mappings = query.offset(skip).limit(limit).all()
-    return [_serialize_mapping(mapping) for mapping in mappings]
-
-
-@router.get("/{mapping_id}", response_model=ModelMappingResponse)
-async def get_mapping(
-    mapping_id: str,
-    db: Session = Depends(get_db),
-):
-    """获取单个模型映射"""
-    mapping = (
-        db.query(ModelMapping)
-        .options(
-            joinedload(ModelMapping.target_global_model),
-            joinedload(ModelMapping.provider),
-        )
-        .filter(ModelMapping.id == mapping_id)
-        .first()
-    )
-
-    if not mapping:
-        raise HTTPException(status_code=404, detail=f"映射 {mapping_id} 不存在")
-
-    return _serialize_mapping(mapping)
-
-
-@router.post("", response_model=ModelMappingResponse, status_code=201)
-async def create_mapping(
-    data: ModelMappingCreate,
-    db: Session = Depends(get_db),
-):
-    """创建模型映射"""
-    source_model = data.source_model.strip()
-    if not source_model:
-        raise HTTPException(status_code=400, detail="source_model 不能为空")
-
-    # 验证 mapping_type
-    if data.mapping_type not in ("alias", "mapping"):
-        raise HTTPException(status_code=400, detail="mapping_type 必须是 'alias' 或 'mapping'")
-
-    # 验证目标 GlobalModel 存在
-    target_model = (
-        db.query(GlobalModel)
-        .filter(GlobalModel.id == data.target_global_model_id, GlobalModel.is_active == True)
-        .first()
-    )
-    if not target_model:
-        raise HTTPException(
-            status_code=404, detail=f"目标模型 {data.target_global_model_id} 不存在或未激活"
-        )
-
-    # 验证 Provider 存在
-    provider = None
-    provider_id = data.provider_id
-    if provider_id:
-        provider = db.query(Provider).filter(Provider.id == provider_id).first()
-        if not provider:
-            raise HTTPException(status_code=404, detail=f"Provider {provider_id} 不存在")
-
-    # 检查映射是否已存在（全局或同一 Provider 下不可重复）
-    existing = (
-        db.query(ModelMapping)
-        .filter(
-            ModelMapping.source_model == source_model,
-            ModelMapping.provider_id == provider_id,
-        )
-        .first()
-    )
-    if existing:
-        raise HTTPException(status_code=400, detail="映射已存在")
-
-    # 创建映射
-    mapping = ModelMapping(
-        id=str(uuid.uuid4()),
-        source_model=source_model,
-        target_global_model_id=data.target_global_model_id,
-        provider_id=provider_id,
-        mapping_type=data.mapping_type,
-        is_active=data.is_active,
-        created_at=datetime.now(timezone.utc),
-        updated_at=datetime.now(timezone.utc),
-    )
-
-    db.add(mapping)
-    db.commit()
-    mapping = (
-        db.query(ModelMapping)
-        .options(
-            joinedload(ModelMapping.target_global_model),
-            joinedload(ModelMapping.provider),
-        )
-        .filter(ModelMapping.id == mapping.id)
-        .first()
-    )
-
-    logger.info(f"创建模型映射: {source_model} -> {target_model.name} "
-        f"(Provider: {provider.name if provider else 'global'}, ID: {mapping.id})")
-
-    cache_service = get_cache_invalidation_service()
-    cache_service.on_model_mapping_changed(source_model, provider_id)
-
-    return _serialize_mapping(mapping)
-
-
-@router.patch("/{mapping_id}", response_model=ModelMappingResponse)
-async def update_mapping(
-    mapping_id: str,
-    data: ModelMappingUpdate,
-    db: Session = Depends(get_db),
-):
-    """更新模型映射"""
-    mapping = db.query(ModelMapping).filter(ModelMapping.id == mapping_id).first()
-    if not mapping:
-        raise HTTPException(status_code=404, detail=f"映射 {mapping_id} 不存在")
-
-    update_data = data.model_dump(exclude_unset=True)
-
-    # 更新 Provider
-    if "provider_id" in update_data:
-        new_provider_id = update_data["provider_id"]
-        if new_provider_id:
-            provider = db.query(Provider).filter(Provider.id == new_provider_id).first()
-            if not provider:
-                raise HTTPException(status_code=404, detail=f"Provider {new_provider_id} 不存在")
-        mapping.provider_id = new_provider_id
-
-    # 更新目标模型
-    if "target_global_model_id" in update_data:
-        target_model = (
-            db.query(GlobalModel)
-            .filter(
-                GlobalModel.id == update_data["target_global_model_id"],
-                GlobalModel.is_active == True,
-            )
-            .first()
-        )
-        if not target_model:
-            raise HTTPException(
-                status_code=404,
-                detail=f"目标模型 {update_data['target_global_model_id']} 不存在或未激活",
-            )
-        mapping.target_global_model_id = update_data["target_global_model_id"]
-
-    # 更新源模型名
-    if "source_model" in update_data:
-        new_source = update_data["source_model"].strip()
-        if not new_source:
-            raise HTTPException(status_code=400, detail="source_model 不能为空")
-        mapping.source_model = new_source
-
-    # 检查唯一约束
-    duplicate = (
-        db.query(ModelMapping)
-        .filter(
-            ModelMapping.source_model == mapping.source_model,
-            ModelMapping.provider_id == mapping.provider_id,
-            ModelMapping.id != mapping_id,
-        )
-        .first()
-    )
-    if duplicate:
-        raise HTTPException(status_code=400, detail="映射已存在")
-
-    # 更新映射类型
-    if "mapping_type" in update_data:
-        if update_data["mapping_type"] not in ("alias", "mapping"):
-            raise HTTPException(status_code=400, detail="mapping_type 必须是 'alias' 或 'mapping'")
-        mapping.mapping_type = update_data["mapping_type"]
-
-    # 更新状态
-    if "is_active" in update_data:
-        mapping.is_active = update_data["is_active"]
-
-    mapping.updated_at = datetime.now(timezone.utc)
-    db.commit()
-    db.refresh(mapping)
-
-    logger.info(f"更新模型映射 (ID: {mapping.id})")
-
-    mapping = (
-        db.query(ModelMapping)
-        .options(
-            joinedload(ModelMapping.target_global_model),
-            joinedload(ModelMapping.provider),
-        )
-        .filter(ModelMapping.id == mapping.id)
-        .first()
-    )
-
-    cache_service = get_cache_invalidation_service()
-    cache_service.on_model_mapping_changed(mapping.source_model, mapping.provider_id)
-
-    return _serialize_mapping(mapping)
-
-
-@router.delete("/{mapping_id}", status_code=204)
-async def delete_mapping(
-    mapping_id: str,
-    db: Session = Depends(get_db),
-):
-    """删除模型映射"""
-    mapping = db.query(ModelMapping).filter(ModelMapping.id == mapping_id).first()
-
-    if not mapping:
-        raise HTTPException(status_code=404, detail=f"映射 {mapping_id} 不存在")
-
-    source_model = mapping.source_model
-    provider_id = mapping.provider_id
-
-    logger.info(f"删除模型映射: {source_model} -> {mapping.target_global_model_id} (ID: {mapping.id})")
-
-    db.delete(mapping)
-    db.commit()
-
-    cache_service = get_cache_invalidation_service()
-    cache_service.on_model_mapping_changed(source_model, provider_id)
-
-    return None

src/api/admin/monitoring/cache.py

@@ -12,6 +12,7 @@ from fastapi.responses import PlainTextResponse
 from sqlalchemy.orm import Session
 
 from src.api.base.admin_adapter import AdminApiAdapter
+from src.api.base.context import ApiRequestContext
 from src.api.base.pagination import PaginationMeta, build_pagination_payload, paginate_sequence
 from src.api.base.pipeline import ApiRequestPipeline
 from src.clients.redis_client import get_redis_client_sync
@@ -20,7 +21,8 @@ from src.core.logger import logger
 from src.database import get_db
 from src.models.database import ApiKey, User
 from src.services.cache.affinity_manager import get_affinity_manager
-from src.services.cache.aware_scheduler import get_cache_aware_scheduler
+from src.services.cache.aware_scheduler import CacheAwareScheduler, get_cache_aware_scheduler
+from src.services.system.config import SystemConfigService
 
 router = APIRouter(prefix="/api/admin/monitoring/cache", tags=["Admin - Monitoring: Cache"])
 pipeline = ApiRequestPipeline()
@@ -87,19 +89,19 @@ def resolve_user_identifier(db: Session, identifier: str) -> Optional[str]:
     # 2. 尝试作为 Username 查询
     user = db.query(User).filter(User.username == identifier).first()
     if user:
-        logger.debug(f"通过Username解析: {identifier} -> {user.id[:8]}...")
+        logger.debug(f"通过Username解析: {identifier} -> {user.id[:8]}...")  # type: ignore[index]
         return user.id
 
     # 3. 尝试作为 Email 查询
     user = db.query(User).filter(User.email == identifier).first()
     if user:
-        logger.debug(f"通过Email解析: {identifier} -> {user.id[:8]}...")
+        logger.debug(f"通过Email解析: {identifier} -> {user.id[:8]}...")  # type: ignore[index]
         return user.id
 
     # 4. 尝试作为 API Key ID 查询
     api_key = db.query(ApiKey).filter(ApiKey.id == identifier).first()
     if api_key:
-        logger.debug(f"通过API Key ID解析: {identifier[:8]}... -> User ID: {api_key.user_id[:8]}...")
+        logger.debug(f"通过API Key ID解析: {identifier[:8]}... -> User ID: {api_key.user_id[:8]}...")  # type: ignore[index]
         return api_key.user_id
 
     # 无法识别
@@ -111,7 +113,7 @@ def resolve_user_identifier(db: Session, identifier: str) -> Optional[str]:
 async def get_cache_stats(
     request: Request,
     db: Session = Depends(get_db),
-):
+) -> Any:
     """
     获取缓存亲和性统计信息
 
@@ -131,7 +133,7 @@ async def get_user_affinity(
     user_identifier: str,
     request: Request,
     db: Session = Depends(get_db),
-):
+) -> Any:
     """
     查询指定用户的所有缓存亲和性
 
@@ -157,7 +159,7 @@ async def list_affinities(
     limit: int = Query(100, ge=1, le=1000, description="返回数量限制"),
     offset: int = Query(0, ge=0, description="偏移量"),
     db: Session = Depends(get_db),
-):
+) -> Any:
     """
     获取所有缓存亲和性列表，可选按关键词过滤
 
@@ -173,7 +175,7 @@ async def clear_user_cache(
     user_identifier: str,
     request: Request,
     db: Session = Depends(get_db),
-):
+) -> Any:
     """
     Clear cache affinity for a specific user
 
@@ -184,11 +186,35 @@ async def clear_user_cache(
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
+@router.delete("/affinity/{affinity_key}/{endpoint_id}/{model_id}/{api_format}")
+async def clear_single_affinity(
+    affinity_key: str,
+    endpoint_id: str,
+    model_id: str,
+    api_format: str,
+    request: Request,
+    db: Session = Depends(get_db),
+) -> Any:
+    """
+    Clear a single cache affinity entry
+
+    Parameters:
+    - affinity_key: API Key ID
+    - endpoint_id: Endpoint ID
+    - model_id: Model ID (GlobalModel ID)
+    - api_format: API format (claude/openai)
+    """
+    adapter = AdminClearSingleAffinityAdapter(
+        affinity_key=affinity_key, endpoint_id=endpoint_id, model_id=model_id, api_format=api_format
+    )
+    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
+
+
 @router.delete("")
 async def clear_all_cache(
     request: Request,
     db: Session = Depends(get_db),
-):
+) -> Any:
     """
     Clear all cache affinities
 
@@ -203,7 +229,7 @@ async def clear_provider_cache(
     provider_id: str,
     request: Request,
     db: Session = Depends(get_db),
-):
+) -> Any:
     """
     Clear cache affinities for a specific provider
 
@@ -218,7 +244,7 @@ async def clear_provider_cache(
 async def get_cache_config(
     request: Request,
     db: Session = Depends(get_db),
-):
+) -> Any:
     """
     获取缓存相关配置
 
@@ -234,7 +260,7 @@ async def get_cache_config(
 async def get_cache_metrics(
     request: Request,
     db: Session = Depends(get_db),
-):
+) -> Any:
     """
     以 Prometheus 文本格式暴露缓存调度指标，方便接入 Grafana。
     """
@@ -246,10 +272,25 @@ async def get_cache_metrics(
 
 
 class AdminCacheStatsAdapter(AdminApiAdapter):
-    async def handle(self, context):  # type: ignore[override]
+    async def handle(self, context: ApiRequestContext) -> Dict[str, Any]:  # type: ignore[override]
         try:
             redis_client = get_redis_client_sync()
-            scheduler = await get_cache_aware_scheduler(redis_client)
+            # 读取系统配置，确保监控接口与编排器使用一致的模式
+            priority_mode = SystemConfigService.get_config(
+                context.db,
+                "provider_priority_mode",
+                CacheAwareScheduler.PRIORITY_MODE_PROVIDER,
+            )
+            scheduling_mode = SystemConfigService.get_config(
+                context.db,
+                "scheduling_mode",
+                CacheAwareScheduler.SCHEDULING_MODE_CACHE_AFFINITY,
+            )
+            scheduler = await get_cache_aware_scheduler(
+                redis_client,
+                priority_mode=priority_mode,
+                scheduling_mode=scheduling_mode,
+            )
             stats = await scheduler.get_stats()
             logger.info("缓存统计信息查询成功")
             context.add_audit_metadata(
@@ -266,10 +307,25 @@ class AdminCacheStatsAdapter(AdminApiAdapter):
 
 
 class AdminCacheMetricsAdapter(AdminApiAdapter):
-    async def handle(self, context):  # type: ignore[override]
+    async def handle(self, context: ApiRequestContext) -> PlainTextResponse:
         try:
             redis_client = get_redis_client_sync()
-            scheduler = await get_cache_aware_scheduler(redis_client)
+            # 读取系统配置，确保监控接口与编排器使用一致的模式
+            priority_mode = SystemConfigService.get_config(
+                context.db,
+                "provider_priority_mode",
+                CacheAwareScheduler.PRIORITY_MODE_PROVIDER,
+            )
+            scheduling_mode = SystemConfigService.get_config(
+                context.db,
+                "scheduling_mode",
+                CacheAwareScheduler.SCHEDULING_MODE_CACHE_AFFINITY,
+            )
+            scheduler = await get_cache_aware_scheduler(
+                redis_client,
+                priority_mode=priority_mode,
+                scheduling_mode=scheduling_mode,
+            )
             stats = await scheduler.get_stats()
             payload = self._format_prometheus(stats)
             context.add_audit_metadata(
@@ -391,7 +447,7 @@ class AdminCacheMetricsAdapter(AdminApiAdapter):
 class AdminGetUserAffinityAdapter(AdminApiAdapter):
     user_identifier: str
 
-    async def handle(self, context):  # type: ignore[override]
+    async def handle(self, context: ApiRequestContext) -> Dict[str, Any]:  # type: ignore[override]
         db = context.db
         try:
             user_id = resolve_user_identifier(db, self.user_identifier)
@@ -472,7 +528,7 @@ class AdminListAffinitiesAdapter(AdminApiAdapter):
     limit: int
     offset: int
 
-    async def handle(self, context):  # type: ignore[override]
+    async def handle(self, context: ApiRequestContext) -> Dict[str, Any]:  # type: ignore[override]
         db = context.db
         redis_client = get_redis_client_sync()
         if not redis_client:
@@ -623,6 +679,7 @@ class AdminListAffinitiesAdapter(AdminApiAdapter):
                 "key_name": key.name if key else None,
                 "key_prefix": provider_key_masked,
                 "rate_multiplier": key.rate_multiplier if key else 1.0,
+                "global_model_id": affinity.get("model_name"),  # 原始的 global_model_id
                 "model_name": (
                     global_model_map.get(affinity.get("model_name")).name
                     if affinity.get("model_name") and global_model_map.get(affinity.get("model_name"))
@@ -682,7 +739,7 @@ class AdminListAffinitiesAdapter(AdminApiAdapter):
 class AdminClearUserCacheAdapter(AdminApiAdapter):
     user_identifier: str
 
-    async def handle(self, context):  # type: ignore[override]
+    async def handle(self, context: ApiRequestContext) -> Dict[str, Any]:  # type: ignore[override]
         db = context.db
         try:
             redis_client = get_redis_client_sync()
@@ -785,8 +842,67 @@ class AdminClearUserCacheAdapter(AdminApiAdapter):
             raise HTTPException(status_code=500, detail=f"清除失败: {exc}")
 
 
+@dataclass
+class AdminClearSingleAffinityAdapter(AdminApiAdapter):
+    affinity_key: str
+    endpoint_id: str
+    model_id: str
+    api_format: str
+
+    async def handle(self, context: ApiRequestContext) -> Dict[str, Any]:  # type: ignore[override]
+        db = context.db
+        try:
+            redis_client = get_redis_client_sync()
+            affinity_mgr = await get_affinity_manager(redis_client)
+
+            # 直接获取指定的亲和性记录（无需遍历全部）
+            existing_affinity = await affinity_mgr.get_affinity(
+                self.affinity_key, self.api_format, self.model_id
+            )
+
+            if not existing_affinity:
+                raise HTTPException(status_code=404, detail="未找到指定的缓存亲和性记录")
+
+            # 验证 endpoint_id 是否匹配
+            if existing_affinity.endpoint_id != self.endpoint_id:
+                raise HTTPException(status_code=404, detail="未找到指定的缓存亲和性记录")
+
+            # 失效单条记录
+            await affinity_mgr.invalidate_affinity(
+                self.affinity_key, self.api_format, self.model_id, endpoint_id=self.endpoint_id
+            )
+
+            # 获取用于日志的信息
+            api_key = db.query(ApiKey).filter(ApiKey.id == self.affinity_key).first()
+            api_key_name = api_key.name if api_key else None
+
+            logger.info(
+                f"已清除单条缓存亲和性: affinity_key={self.affinity_key[:8]}..., "
+                f"endpoint_id={self.endpoint_id[:8]}..., model_id={self.model_id[:8]}..."
+            )
+
+            context.add_audit_metadata(
+                action="cache_clear_single",
+                affinity_key=self.affinity_key,
+                endpoint_id=self.endpoint_id,
+                model_id=self.model_id,
+            )
+            return {
+                "status": "ok",
+                "message": f"已清除缓存亲和性: {api_key_name or self.affinity_key[:8]}",
+                "affinity_key": self.affinity_key,
+                "endpoint_id": self.endpoint_id,
+                "model_id": self.model_id,
+            }
+        except HTTPException:
+            raise
+        except Exception as exc:
+            logger.exception(f"清除单条缓存亲和性失败: {exc}")
+            raise HTTPException(status_code=500, detail=f"清除失败: {exc}")
+
+
 class AdminClearAllCacheAdapter(AdminApiAdapter):
-    async def handle(self, context):  # type: ignore[override]
+    async def handle(self, context: ApiRequestContext) -> Dict[str, Any]:  # type: ignore[override]
         try:
             redis_client = get_redis_client_sync()
             affinity_mgr = await get_affinity_manager(redis_client)
@@ -806,7 +922,7 @@ class AdminClearAllCacheAdapter(AdminApiAdapter):
 class AdminClearProviderCacheAdapter(AdminApiAdapter):
     provider_id: str
 
-    async def handle(self, context):  # type: ignore[override]
+    async def handle(self, context: ApiRequestContext) -> Dict[str, Any]:  # type: ignore[override]
         try:
             redis_client = get_redis_client_sync()
             affinity_mgr = await get_affinity_manager(redis_client)
@@ -829,7 +945,7 @@ class AdminClearProviderCacheAdapter(AdminApiAdapter):
 
 
 class AdminCacheConfigAdapter(AdminApiAdapter):
-    async def handle(self, context):  # type: ignore[override]
+    async def handle(self, context: ApiRequestContext) -> Dict[str, Any]:  # type: ignore[override]
         from src.services.cache.affinity_manager import CacheAffinityManager
         from src.services.cache.aware_scheduler import CacheAwareScheduler
         from src.services.rate_limit.adaptive_reservation import get_adaptive_reservation_manager
@@ -869,3 +985,464 @@ class AdminCacheConfigAdapter(AdminApiAdapter):
             dynamic_reservation_enabled=True,
         )
         return response
+
+
+# ==================== 模型映射缓存管理 ====================
+
+
+@router.get("/model-mapping/stats")
+async def get_model_mapping_cache_stats(
+    request: Request,
+    db: Session = Depends(get_db),
+) -> Any:
+    """
+    获取模型映射缓存统计信息
+
+    返回:
+    - 缓存键数量
+    - 缓存 TTL 配置
+    - 各类型缓存数量
+    """
+    adapter = AdminModelMappingCacheStatsAdapter()
+    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
+
+
+@router.delete("/model-mapping")
+async def clear_all_model_mapping_cache(
+    request: Request,
+    db: Session = Depends(get_db),
+) -> Any:
+    """
+    清除所有模型映射缓存
+
+    警告: 这会影响所有模型解析，请谨慎使用
+    """
+    adapter = AdminClearAllModelMappingCacheAdapter()
+    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
+
+
+@router.delete("/model-mapping/{model_name}")
+async def clear_model_mapping_cache_by_name(
+    model_name: str,
+    request: Request,
+    db: Session = Depends(get_db),
+) -> Any:
+    """
+    清除指定模型名称的映射缓存
+
+    参数:
+    - model_name: 模型名称（可以是 GlobalModel.name 或映射名称）
+    """
+    adapter = AdminClearModelMappingCacheByNameAdapter(model_name=model_name)
+    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
+
+
+@router.delete("/model-mapping/provider/{provider_id}/{global_model_id}")
+async def clear_provider_model_mapping_cache(
+    provider_id: str,
+    global_model_id: str,
+    request: Request,
+    db: Session = Depends(get_db),
+) -> Any:
+    """
+    清除指定 Provider 和 GlobalModel 的模型映射缓存
+
+    参数:
+    - provider_id: Provider ID
+    - global_model_id: GlobalModel ID
+    """
+    adapter = AdminClearProviderModelMappingCacheAdapter(
+        provider_id=provider_id, global_model_id=global_model_id
+    )
+    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
+
+
+class AdminModelMappingCacheStatsAdapter(AdminApiAdapter):
+    async def handle(self, context: ApiRequestContext) -> Dict[str, Any]:  # type: ignore[override]
+        import json
+
+        from src.clients.redis_client import get_redis_client
+        from src.config.constants import CacheTTL
+        from src.models.database import GlobalModel, Model, Provider
+
+        db = context.db
+
+        try:
+            redis = await get_redis_client(require_redis=False)
+            if not redis:
+                return {
+                    "status": "ok",
+                    "data": {
+                        "available": False,
+                        "message": "Redis 未启用，模型映射缓存不可用",
+                    },
+                }
+
+            # 统计各类型缓存键数量
+            model_id_keys = []
+            global_model_id_keys = []
+            global_model_name_keys = []
+            global_model_resolve_keys = []
+            provider_global_keys = []
+
+            # 扫描所有模型相关的缓存键
+            async for key in redis.scan_iter(match="model:*", count=100):
+                key_str = key.decode() if isinstance(key, bytes) else key
+                if key_str.startswith("model:id:"):
+                    model_id_keys.append(key_str)
+                elif key_str.startswith("model:provider_global:"):
+                    # 过滤掉 hits 统计键，只保留实际的缓存键
+                    if not key_str.startswith("model:provider_global:hits:"):
+                        provider_global_keys.append(key_str)
+
+            async for key in redis.scan_iter(match="global_model:*", count=100):
+                key_str = key.decode() if isinstance(key, bytes) else key
+                if key_str.startswith("global_model:id:"):
+                    global_model_id_keys.append(key_str)
+                elif key_str.startswith("global_model:name:"):
+                    global_model_name_keys.append(key_str)
+                elif key_str.startswith("global_model:resolve:"):
+                    global_model_resolve_keys.append(key_str)
+
+            total_keys = (
+                len(model_id_keys)
+                + len(global_model_id_keys)
+                + len(global_model_name_keys)
+                + len(global_model_resolve_keys)
+                + len(provider_global_keys)
+            )
+
+            # 解析缓存内容，构建映射列表
+            mappings = []
+            unmapped_entries = []
+
+            for key in global_model_resolve_keys[:100]:  # 最多处理 100 个
+                mapping_name = key.replace("global_model:resolve:", "")
+                try:
+                    cached_value = await redis.get(key)
+                    ttl = await redis.ttl(key)
+
+                    if cached_value:
+                        cached_str = (
+                            cached_value.decode()
+                            if isinstance(cached_value, bytes)
+                            else cached_value
+                        )
+
+                        if cached_str == "NOT_FOUND":
+                            unmapped_entries.append({
+                                "mapping_name": mapping_name,
+                                "status": "not_found",
+                                "ttl": ttl if ttl > 0 else None,
+                            })
+                        else:
+                            try:
+                                cached_data = json.loads(cached_str)
+                                global_model_id = cached_data.get("id")
+                                global_model_name = cached_data.get("name")
+                                global_model_display_name = cached_data.get("display_name")
+
+                                # 跳过 mapping_name == global_model_name 的情况（直接匹配，不是映射）
+                                if mapping_name == global_model_name:
+                                    continue
+
+                                # 查询哪些 Provider 配置了这个映射名称
+                                provider_names = []
+                                if global_model_id:
+                                    models = (
+                                        db.query(Model, Provider)
+                                        .join(Provider, Model.provider_id == Provider.id)
+                                        .filter(
+                                            Model.global_model_id == global_model_id,
+                                            Model.is_active,
+                                            Provider.is_active,
+                                        )
+                                        .all()
+                                    )
+                                    # 只显示配置了该映射名称的 Provider
+                                    for model, provider in models:
+                                        # 检查是否是主模型名称
+                                        if model.provider_model_name == mapping_name:
+                                            provider_names.append(
+                                                provider.display_name or provider.name
+                                            )
+                                            continue
+                                        # 检查是否在映射列表中
+                                        if model.provider_model_mappings:
+                                            mapping_list = [
+                                                a.get("name")
+                                                for a in model.provider_model_mappings
+                                                if isinstance(a, dict)
+                                            ]
+                                            if mapping_name in mapping_list:
+                                                provider_names.append(
+                                                    provider.display_name or provider.name
+                                                )
+                                    provider_names = sorted(list(set(provider_names)))
+
+                                mappings.append({
+                                    "mapping_name": mapping_name,
+                                    "global_model_name": global_model_name,
+                                    "global_model_display_name": global_model_display_name,
+                                    "providers": provider_names,
+                                    "ttl": ttl if ttl > 0 else None,
+                                })
+
+                            except json.JSONDecodeError:
+                                unmapped_entries.append({
+                                    "mapping_name": mapping_name,
+                                    "status": "invalid",
+                                    "ttl": ttl if ttl > 0 else None,
+                                })
+                except Exception as e:
+                    logger.warning(f"解析缓存键 {key} 失败: {e}")
+                    unmapped_entries.append({
+                        "mapping_name": mapping_name,
+                        "status": "error",
+                        "ttl": None,
+                    })
+
+            # 按 mapping_name 排序
+            mappings.sort(key=lambda x: x["mapping_name"])
+
+            # 3. 解析 provider_global 缓存（Provider 级别的模型解析缓存）
+            provider_model_mappings = []
+            # 预加载 Provider 和 GlobalModel 数据
+            provider_map = {str(p.id): p for p in db.query(Provider).filter(Provider.is_active.is_(True)).all()}
+            global_model_map = {str(gm.id): gm for gm in db.query(GlobalModel).filter(GlobalModel.is_active.is_(True)).all()}
+
+            for key in provider_global_keys[:100]:  # 最多处理 100 个
+                # key 格式: model:provider_global:{provider_id}:{global_model_id}
+                try:
+                    parts = key.replace("model:provider_global:", "").split(":")
+                    if len(parts) != 2:
+                        continue
+                    provider_id, global_model_id = parts
+
+                    cached_value = await redis.get(key)
+                    ttl = await redis.ttl(key)
+
+                    # 获取命中次数
+                    hit_count_key = f"model:provider_global:hits:{provider_id}:{global_model_id}"
+                    hit_count_raw = await redis.get(hit_count_key)
+                    hit_count = int(hit_count_raw) if hit_count_raw else 0
+
+                    if cached_value:
+                        cached_str = (
+                            cached_value.decode()
+                            if isinstance(cached_value, bytes)
+                            else cached_value
+                        )
+                        try:
+                            cached_data = json.loads(cached_str)
+                            provider_model_name = cached_data.get("provider_model_name")
+                            provider_model_mappings = cached_data.get("provider_model_mappings", [])
+
+                            # 获取 Provider 和 GlobalModel 信息
+                            provider = provider_map.get(provider_id)
+                            global_model = global_model_map.get(global_model_id)
+
+                            if provider and global_model:
+                                # 提取映射名称
+                                mapping_names = []
+                                if provider_model_mappings:
+                                    for mapping_entry in provider_model_mappings:
+                                        if isinstance(mapping_entry, dict) and mapping_entry.get("name"):
+                                            mapping_names.append(mapping_entry["name"])
+
+                                # provider_model_name 为空时跳过
+                                if not provider_model_name:
+                                    continue
+
+                                # 只显示有实际映射的条目：
+                                # 1. 全局模型名 != Provider 模型名（模型名称映射）
+                                # 2. 或者有映射配置
+                                has_name_mapping = global_model.name != provider_model_name
+                                has_mappings = len(mapping_names) > 0
+
+                                if has_name_mapping or has_mappings:
+                                    # 构建用于展示的映射列表
+                                    # 如果只有名称映射没有额外映射，则用 global_model_name 作为"请求名称"
+                                    display_mappings = mapping_names if mapping_names else [global_model.name]
+
+                                    provider_model_mappings.append({
+                                        "provider_id": provider_id,
+                                        "provider_name": provider.display_name or provider.name,
+                                        "global_model_id": global_model_id,
+                                        "global_model_name": global_model.name,
+                                        "global_model_display_name": global_model.display_name,
+                                        "provider_model_name": provider_model_name,
+                                        "aliases": display_mappings,
+                                        "ttl": ttl if ttl > 0 else None,
+                                        "hit_count": hit_count,
+                                    })
+                        except json.JSONDecodeError:
+                            pass
+                except Exception as e:
+                    logger.warning(f"解析 provider_global 缓存键 {key} 失败: {e}")
+
+            # 按 provider_name + global_model_name 排序
+            provider_model_mappings.sort(key=lambda x: (x["provider_name"], x["global_model_name"]))
+
+            response_data = {
+                "available": True,
+                "ttl_seconds": CacheTTL.MODEL,
+                "total_keys": total_keys,
+                "breakdown": {
+                    "model_by_id": len(model_id_keys),
+                    "model_by_provider_global": len(provider_global_keys),
+                    "global_model_by_id": len(global_model_id_keys),
+                    "global_model_by_name": len(global_model_name_keys),
+                    "global_model_resolve": len(global_model_resolve_keys),
+                },
+                "mappings": mappings,
+                "provider_model_mappings": provider_model_mappings if provider_model_mappings else None,
+                "unmapped": unmapped_entries if unmapped_entries else None,
+            }
+
+            context.add_audit_metadata(
+                action="model_mapping_cache_stats",
+                total_keys=total_keys,
+            )
+            return {"status": "ok", "data": response_data}
+
+        except Exception as exc:
+            logger.exception(f"获取模型映射缓存统计失败: {exc}")
+            raise HTTPException(status_code=500, detail=f"获取统计失败: {exc}")
+
+
+class AdminClearAllModelMappingCacheAdapter(AdminApiAdapter):
+    async def handle(self, context: ApiRequestContext) -> Dict[str, Any]:  # type: ignore[override]
+        from src.clients.redis_client import get_redis_client
+
+        try:
+            redis = await get_redis_client(require_redis=False)
+            if not redis:
+                raise HTTPException(status_code=503, detail="Redis 未启用")
+
+            deleted_count = 0
+
+            # 删除所有模型相关的缓存键
+            keys_to_delete = []
+            async for key in redis.scan_iter(match="model:*", count=100):
+                keys_to_delete.append(key)
+            async for key in redis.scan_iter(match="global_model:*", count=100):
+                keys_to_delete.append(key)
+
+            if keys_to_delete:
+                deleted_count = await redis.delete(*keys_to_delete)
+
+            logger.warning(f"已清除所有模型映射缓存（管理员操作）: {deleted_count} 个键")
+            context.add_audit_metadata(
+                action="model_mapping_cache_clear_all",
+                deleted_count=deleted_count,
+            )
+            return {
+                "status": "ok",
+                "message": f"已清除所有模型映射缓存",
+                "deleted_count": deleted_count,
+            }
+
+        except HTTPException:
+            raise
+        except Exception as exc:
+            logger.exception(f"清除模型映射缓存失败: {exc}")
+            raise HTTPException(status_code=500, detail=f"清除失败: {exc}")
+
+
+@dataclass
+class AdminClearModelMappingCacheByNameAdapter(AdminApiAdapter):
+    model_name: str
+
+    async def handle(self, context: ApiRequestContext) -> Dict[str, Any]:  # type: ignore[override]
+        from src.clients.redis_client import get_redis_client
+
+        try:
+            redis = await get_redis_client(require_redis=False)
+            if not redis:
+                raise HTTPException(status_code=503, detail="Redis 未启用")
+
+            deleted_keys = []
+
+            # 清除 resolve 缓存
+            resolve_key = f"global_model:resolve:{self.model_name}"
+            if await redis.exists(resolve_key):
+                await redis.delete(resolve_key)
+                deleted_keys.append(resolve_key)
+
+            # 清除 name 缓存
+            name_key = f"global_model:name:{self.model_name}"
+            if await redis.exists(name_key):
+                await redis.delete(name_key)
+                deleted_keys.append(name_key)
+
+            logger.info(f"已清除模型映射缓存: model_name={self.model_name}, 删除键={deleted_keys}")
+            context.add_audit_metadata(
+                action="model_mapping_cache_clear_by_name",
+                model_name=self.model_name,
+                deleted_keys=deleted_keys,
+            )
+            return {
+                "status": "ok",
+                "message": f"已清除模型 {self.model_name} 的映射缓存",
+                "model_name": self.model_name,
+                "deleted_keys": deleted_keys,
+            }
+
+        except HTTPException:
+            raise
+        except Exception as exc:
+            logger.exception(f"清除模型映射缓存失败: {exc}")
+            raise HTTPException(status_code=500, detail=f"清除失败: {exc}")
+
+
+@dataclass
+class AdminClearProviderModelMappingCacheAdapter(AdminApiAdapter):
+    provider_id: str
+    global_model_id: str
+
+    async def handle(self, context: ApiRequestContext) -> Dict[str, Any]:  # type: ignore[override]
+        from src.clients.redis_client import get_redis_client
+
+        try:
+            redis = await get_redis_client(require_redis=False)
+            if not redis:
+                raise HTTPException(status_code=503, detail="Redis 未启用")
+
+            deleted_keys = []
+
+            # 清除 provider_global 缓存
+            provider_global_key = f"model:provider_global:{self.provider_id}:{self.global_model_id}"
+            if await redis.exists(provider_global_key):
+                await redis.delete(provider_global_key)
+                deleted_keys.append(provider_global_key)
+
+            # 清除对应的 hit_count 缓存
+            hit_count_key = f"model:provider_global:hits:{self.provider_id}:{self.global_model_id}"
+            if await redis.exists(hit_count_key):
+                await redis.delete(hit_count_key)
+                deleted_keys.append(hit_count_key)
+
+            logger.info(
+                f"已清除 Provider 模型映射缓存: provider_id={self.provider_id[:8]}..., "
+                f"global_model_id={self.global_model_id[:8]}..., 删除键={deleted_keys}"
+            )
+            context.add_audit_metadata(
+                action="provider_model_mapping_cache_clear",
+                provider_id=self.provider_id,
+                global_model_id=self.global_model_id,
+                deleted_keys=deleted_keys,
+            )
+            return {
+                "status": "ok",
+                "message": "已清除 Provider 模型映射缓存",
+                "provider_id": self.provider_id,
+                "global_model_id": self.global_model_id,
+                "deleted_keys": deleted_keys,
+            }
+
+        except HTTPException:
+            raise
+        except Exception as exc:
+            logger.exception(f"清除 Provider 模型映射缓存失败: {exc}")
+            raise HTTPException(status_code=500, detail=f"清除失败: {exc}")

src/api/admin/monitoring/trace.py

@@ -7,7 +7,7 @@ from datetime import datetime
 from typing import List, Optional
 
 from fastapi import APIRouter, Depends, HTTPException, Query, Request
-from pydantic import BaseModel
+from pydantic import BaseModel, ConfigDict
 from sqlalchemy.orm import Session
 
 from src.api.base.admin_adapter import AdminApiAdapter
@@ -52,8 +52,7 @@ class CandidateResponse(BaseModel):
     started_at: Optional[datetime] = None
     finished_at: Optional[datetime] = None
 
-    class Config:
-        from_attributes = True
+    model_config = ConfigDict(from_attributes=True)
 
 
 class RequestTraceResponse(BaseModel):

src/api/admin/provider_query.py

@@ -1,46 +1,30 @@
 """
 Provider Query API 端点
-用于查询提供商的余额、使用记录等信息
+用于查询提供商的模型列表等信息
 """
 
-from datetime import datetime
+import asyncio
 from typing import Optional
 
-from fastapi import APIRouter, Depends, HTTPException, Query
+import httpx
+from fastapi import APIRouter, Depends, HTTPException
 from pydantic import BaseModel
-from sqlalchemy.ext.asyncio import AsyncSession
+from sqlalchemy.orm import Session, joinedload
 
+from src.api.handlers.base.chat_adapter_base import get_adapter_class
+from src.api.handlers.base.cli_adapter_base import get_cli_adapter_class
+from src.core.crypto import crypto_service
 from src.core.logger import logger
 from src.database.database import get_db
-from src.models.database import Provider, ProviderAPIKey, ProviderEndpoint, User
-
-# 初始化适配器注册
-from src.plugins.provider_query import init  # noqa
-from src.plugins.provider_query import get_query_registry
-from src.plugins.provider_query.base import QueryCapability
+from src.models.database import Provider, ProviderEndpoint, User
 from src.utils.auth_utils import get_current_user
 
-router = APIRouter(prefix="/provider-query", tags=["Provider Query"])
+router = APIRouter(prefix="/api/admin/provider-query", tags=["Provider Query"])
 
 
 # ============ Request/Response Models ============
 
 
-class BalanceQueryRequest(BaseModel):
-    """余额查询请求"""
-
-    provider_id: str
-    api_key_id: Optional[str] = None  # 如果不指定，使用提供商的第一个可用 API Key
-
-
-class UsageSummaryQueryRequest(BaseModel):
-    """使用汇总查询请求"""
-
-    provider_id: str
-    api_key_id: Optional[str] = None
-    period: str = "month"  # day, week, month, year
-
-
 class ModelsQueryRequest(BaseModel):
     """模型列表查询请求"""
 
@@ -51,360 +35,174 @@ class ModelsQueryRequest(BaseModel):
 # ============ API Endpoints ============
 
 
-@router.get("/adapters")
-async def list_adapters(
-    current_user: User = Depends(get_current_user),
-):
-    """
-    获取所有可用的查询适配器
+def _get_adapter_for_format(api_format: str):
+    """根据 API 格式获取对应的 Adapter 类"""
+    # 先检查 Chat Adapter 注册表
+    adapter_class = get_adapter_class(api_format)
+    if adapter_class:
+        return adapter_class
 
-    Returns:
-        适配器列表
-    """
-    registry = get_query_registry()
-    adapters = registry.list_adapters()
+    # 再检查 CLI Adapter 注册表
+    cli_adapter_class = get_cli_adapter_class(api_format)
+    if cli_adapter_class:
+        return cli_adapter_class
 
-    return {"success": True, "data": adapters}
-
-
-@router.get("/capabilities/{provider_id}")
-async def get_provider_capabilities(
-    provider_id: str,
-    db: AsyncSession = Depends(get_db),
-    current_user: User = Depends(get_current_user),
-):
-    """
-    获取提供商支持的查询能力
-
-    Args:
-        provider_id: 提供商 ID
-
-    Returns:
-        支持的查询能力列表
-    """
-    # 获取提供商
-    from sqlalchemy import select
-
-    result = await db.execute(select(Provider).where(Provider.id == provider_id))
-    provider = result.scalar_one_or_none()
-
-    if not provider:
-        raise HTTPException(status_code=404, detail="Provider not found")
-
-    registry = get_query_registry()
-    capabilities = registry.get_capabilities_for_provider(provider.name)
-
-    if capabilities is None:
-        return {
-            "success": True,
-            "data": {
-                "provider_id": provider_id,
-                "provider_name": provider.name,
-                "capabilities": [],
-                "has_adapter": False,
-                "message": "No query adapter available for this provider",
-            },
-        }
-
-    return {
-        "success": True,
-        "data": {
-            "provider_id": provider_id,
-            "provider_name": provider.name,
-            "capabilities": [c.name for c in capabilities],
-            "has_adapter": True,
-        },
-    }
-
-
-@router.post("/balance")
-async def query_balance(
-    request: BalanceQueryRequest,
-    db: AsyncSession = Depends(get_db),
-    current_user: User = Depends(get_current_user),
-):
-    """
-    查询提供商余额
-
-    Args:
-        request: 查询请求
-
-    Returns:
-        余额信息
-    """
-    from sqlalchemy import select
-    from sqlalchemy.orm import selectinload
-
-    # 获取提供商及其端点
-    result = await db.execute(
-        select(Provider)
-        .options(selectinload(Provider.endpoints).selectinload(ProviderEndpoint.api_keys))
-        .where(Provider.id == request.provider_id)
-    )
-    provider = result.scalar_one_or_none()
-
-    if not provider:
-        raise HTTPException(status_code=404, detail="Provider not found")
-
-    # 获取 API Key
-    api_key_value = None
-    endpoint_config = None
-
-    if request.api_key_id:
-        # 查找指定的 API Key
-        for endpoint in provider.endpoints:
-            for api_key in endpoint.api_keys:
-                if api_key.id == request.api_key_id:
-                    api_key_value = api_key.api_key
-                    endpoint_config = {
-                        "base_url": endpoint.base_url,
-                        "api_format": endpoint.api_format if endpoint.api_format else None,
-                    }
-                    break
-            if api_key_value:
-                break
-
-        if not api_key_value:
-            raise HTTPException(status_code=404, detail="API Key not found")
-    else:
-        # 使用第一个可用的 API Key
-        for endpoint in provider.endpoints:
-            if endpoint.is_active and endpoint.api_keys:
-                for api_key in endpoint.api_keys:
-                    if api_key.is_active:
-                        api_key_value = api_key.api_key
-                        endpoint_config = {
-                            "base_url": endpoint.base_url,
-                            "api_format": endpoint.api_format if endpoint.api_format else None,
-                        }
-                        break
-                if api_key_value:
-                    break
-
-        if not api_key_value:
-            raise HTTPException(status_code=400, detail="No active API Key found for this provider")
-
-    # 查询余额
-    registry = get_query_registry()
-    query_result = await registry.query_provider_balance(
-        provider_type=provider.name, api_key=api_key_value, endpoint_config=endpoint_config
-    )
-
-    if not query_result.success:
-        logger.warning(f"Balance query failed for provider {provider.name}: {query_result.error}")
-
-    return {
-        "success": query_result.success,
-        "data": query_result.to_dict(),
-        "provider": {
-            "id": provider.id,
-            "name": provider.name,
-            "display_name": provider.display_name,
-        },
-    }
-
-
-@router.post("/usage-summary")
-async def query_usage_summary(
-    request: UsageSummaryQueryRequest,
-    db: AsyncSession = Depends(get_db),
-    current_user: User = Depends(get_current_user),
-):
-    """
-    查询提供商使用汇总
-
-    Args:
-        request: 查询请求
-
-    Returns:
-        使用汇总信息
-    """
-    from sqlalchemy import select
-    from sqlalchemy.orm import selectinload
-
-    # 获取提供商及其端点
-    result = await db.execute(
-        select(Provider)
-        .options(selectinload(Provider.endpoints).selectinload(ProviderEndpoint.api_keys))
-        .where(Provider.id == request.provider_id)
-    )
-    provider = result.scalar_one_or_none()
-
-    if not provider:
-        raise HTTPException(status_code=404, detail="Provider not found")
-
-    # 获取 API Key（逻辑同上）
-    api_key_value = None
-    endpoint_config = None
-
-    if request.api_key_id:
-        for endpoint in provider.endpoints:
-            for api_key in endpoint.api_keys:
-                if api_key.id == request.api_key_id:
-                    api_key_value = api_key.api_key
-                    endpoint_config = {"base_url": endpoint.base_url}
-                    break
-            if api_key_value:
-                break
-
-        if not api_key_value:
-            raise HTTPException(status_code=404, detail="API Key not found")
-    else:
-        for endpoint in provider.endpoints:
-            if endpoint.is_active and endpoint.api_keys:
-                for api_key in endpoint.api_keys:
-                    if api_key.is_active:
-                        api_key_value = api_key.api_key
-                        endpoint_config = {"base_url": endpoint.base_url}
-                        break
-                if api_key_value:
-                    break
-
-        if not api_key_value:
-            raise HTTPException(status_code=400, detail="No active API Key found for this provider")
-
-    # 查询使用汇总
-    registry = get_query_registry()
-    query_result = await registry.query_provider_usage(
-        provider_type=provider.name,
-        api_key=api_key_value,
-        period=request.period,
-        endpoint_config=endpoint_config,
-    )
-
-    return {
-        "success": query_result.success,
-        "data": query_result.to_dict(),
-        "provider": {
-            "id": provider.id,
-            "name": provider.name,
-            "display_name": provider.display_name,
-        },
-    }
+    return None
 
 
 @router.post("/models")
 async def query_available_models(
     request: ModelsQueryRequest,
-    db: AsyncSession = Depends(get_db),
+    db: Session = Depends(get_db),
     current_user: User = Depends(get_current_user),
 ):
     """
     查询提供商可用模型
 
+    遍历所有活跃端点，根据端点的 API 格式选择正确的 Adapter 进行请求：
+    - OPENAI/OPENAI_CLI: 使用 OpenAIChatAdapter.fetch_models
+    - CLAUDE/CLAUDE_CLI: 使用 ClaudeChatAdapter.fetch_models
+    - GEMINI/GEMINI_CLI: 使用 GeminiChatAdapter.fetch_models
+
     Args:
         request: 查询请求
 
     Returns:
-        模型列表
+        所有端点的模型列表（合并）
     """
-    from sqlalchemy import select
-    from sqlalchemy.orm import selectinload
-
     # 获取提供商及其端点
-    result = await db.execute(
-        select(Provider)
-        .options(selectinload(Provider.endpoints).selectinload(ProviderEndpoint.api_keys))
-        .where(Provider.id == request.provider_id)
+    provider = (
+        db.query(Provider)
+        .options(joinedload(Provider.endpoints).joinedload(ProviderEndpoint.api_keys))
+        .filter(Provider.id == request.provider_id)
+        .first()
     )
-    provider = result.scalar_one_or_none()
 
     if not provider:
         raise HTTPException(status_code=404, detail="Provider not found")
 
-    # 获取 API Key
-    api_key_value = None
-    endpoint_config = None
+    # 收集所有活跃端点的配置
+    endpoint_configs: list[dict] = []
 
     if request.api_key_id:
+        # 指定了特定的 API Key，只使用该 Key 对应的端点
         for endpoint in provider.endpoints:
             for api_key in endpoint.api_keys:
                 if api_key.id == request.api_key_id:
-                    api_key_value = api_key.api_key
-                    endpoint_config = {"base_url": endpoint.base_url}
+                    try:
+                        api_key_value = crypto_service.decrypt(api_key.api_key)
+                    except Exception as e:
+                        logger.error(f"Failed to decrypt API key: {e}")
+                        raise HTTPException(status_code=500, detail="Failed to decrypt API key")
+                    endpoint_configs.append({
+                        "api_key": api_key_value,
+                        "base_url": endpoint.base_url,
+                        "api_format": endpoint.api_format,
+                        "extra_headers": endpoint.headers,
+                    })
                     break
-            if api_key_value:
+            if endpoint_configs:
                 break
 
-        if not api_key_value:
+        if not endpoint_configs:
             raise HTTPException(status_code=404, detail="API Key not found")
     else:
+        # 遍历所有活跃端点，每个端点取第一个可用的 Key
         for endpoint in provider.endpoints:
-            if endpoint.is_active and endpoint.api_keys:
-                for api_key in endpoint.api_keys:
-                    if api_key.is_active:
-                        api_key_value = api_key.api_key
-                        endpoint_config = {"base_url": endpoint.base_url}
-                        break
-                if api_key_value:
-                    break
+            if not endpoint.is_active or not endpoint.api_keys:
+                continue
 
-        if not api_key_value:
+            # 找第一个可用的 Key
+            for api_key in endpoint.api_keys:
+                if api_key.is_active:
+                    try:
+                        api_key_value = crypto_service.decrypt(api_key.api_key)
+                    except Exception as e:
+                        logger.error(f"Failed to decrypt API key: {e}")
+                        continue  # 尝试下一个 Key
+                    endpoint_configs.append({
+                        "api_key": api_key_value,
+                        "base_url": endpoint.base_url,
+                        "api_format": endpoint.api_format,
+                        "extra_headers": endpoint.headers,
+                    })
+                    break  # 只取第一个可用的 Key
+
+        if not endpoint_configs:
             raise HTTPException(status_code=400, detail="No active API Key found for this provider")
 
-    # 查询模型
-    registry = get_query_registry()
-    adapter = registry.get_adapter_for_provider(provider.name)
+    # 并发请求所有端点的模型列表
+    all_models: list = []
+    errors: list[str] = []
 
-    if not adapter:
-        raise HTTPException(
-            status_code=400, detail=f"No query adapter available for provider: {provider.name}"
+    async def fetch_endpoint_models(
+        client: httpx.AsyncClient, config: dict
+    ) -> tuple[list, Optional[str]]:
+        base_url = config["base_url"]
+        if not base_url:
+            return [], None
+        base_url = base_url.rstrip("/")
+        api_format = config["api_format"]
+        api_key_value = config["api_key"]
+        extra_headers = config.get("extra_headers")
+
+        try:
+            # 获取对应的 Adapter 类并调用 fetch_models
+            adapter_class = _get_adapter_for_format(api_format)
+            if not adapter_class:
+                return [], f"Unknown API format: {api_format}"
+            models, error = await adapter_class.fetch_models(
+                client, base_url, api_key_value, extra_headers
+            )
+            # 确保所有模型都有 api_format 字段
+            for m in models:
+                if "api_format" not in m:
+                    m["api_format"] = api_format
+            return models, error
+        except Exception as e:
+            logger.error(f"Error fetching models from {api_format} endpoint: {e}")
+            return [], f"{api_format}: {str(e)}"
+
+    # 限制并发请求数量，避免触发上游速率限制
+    MAX_CONCURRENT_REQUESTS = 5
+    semaphore = asyncio.Semaphore(MAX_CONCURRENT_REQUESTS)
+
+    async def fetch_with_semaphore(
+        client: httpx.AsyncClient, config: dict
+    ) -> tuple[list, Optional[str]]:
+        async with semaphore:
+            return await fetch_endpoint_models(client, config)
+
+    async with httpx.AsyncClient(timeout=30.0) as client:
+        results = await asyncio.gather(
+            *[fetch_with_semaphore(client, c) for c in endpoint_configs]
         )
+        for models, error in results:
+            all_models.extend(models)
+            if error:
+                errors.append(error)
 
-    query_result = await adapter.query_available_models(
-        api_key=api_key_value, endpoint_config=endpoint_config
-    )
+    # 按 model id + api_format 去重（保留第一个）
+    seen_keys: set[str] = set()
+    unique_models: list = []
+    for model in all_models:
+        model_id = model.get("id")
+        api_format = model.get("api_format", "")
+        unique_key = f"{model_id}:{api_format}"
+        if model_id and unique_key not in seen_keys:
+            seen_keys.add(unique_key)
+            unique_models.append(model)
+
+    error = "; ".join(errors) if errors else None
+    if not unique_models and not error:
+        error = "No models returned from any endpoint"
 
     return {
-        "success": query_result.success,
-        "data": query_result.to_dict(),
+        "success": len(unique_models) > 0,
+        "data": {"models": unique_models, "error": error},
         "provider": {
             "id": provider.id,
             "name": provider.name,
             "display_name": provider.display_name,
         },
     }
-
-
-@router.delete("/cache/{provider_id}")
-async def clear_query_cache(
-    provider_id: str,
-    api_key_id: Optional[str] = None,
-    db: AsyncSession = Depends(get_db),
-    current_user: User = Depends(get_current_user),
-):
-    """
-    清除查询缓存
-
-    Args:
-        provider_id: 提供商 ID
-        api_key_id: 可选，指定清除某个 API Key 的缓存
-
-    Returns:
-        清除结果
-    """
-    from sqlalchemy import select
-
-    # 获取提供商
-    result = await db.execute(select(Provider).where(Provider.id == provider_id))
-    provider = result.scalar_one_or_none()
-
-    if not provider:
-        raise HTTPException(status_code=404, detail="Provider not found")
-
-    registry = get_query_registry()
-    adapter = registry.get_adapter_for_provider(provider.name)
-
-    if adapter:
-        if api_key_id:
-            # 获取 API Key 值来清除缓存
-            from sqlalchemy.orm import selectinload
-
-            result = await db.execute(select(ProviderAPIKey).where(ProviderAPIKey.id == api_key_id))
-            api_key = result.scalar_one_or_none()
-            if api_key:
-                adapter.clear_cache(api_key.api_key)
-        else:
-            adapter.clear_cache()
-
-    return {"success": True, "message": "Cache cleared successfully"}

src/api/admin/providers/models.py

@@ -6,10 +6,10 @@ from dataclasses import dataclass
 from typing import Any, Dict, List, Optional
 
 from fastapi import APIRouter, Depends, Request
-from sqlalchemy import or_
 from sqlalchemy.orm import Session, joinedload
 
 from src.api.base.admin_adapter import AdminApiAdapter
+from src.api.base.models_service import invalidate_models_list_cache
 from src.api.base.pipeline import ApiRequestPipeline
 from src.core.exceptions import InvalidRequestException, NotFoundException
 from src.core.logger import logger
@@ -22,17 +22,18 @@ from src.models.api import (
 from src.models.pydantic_models import (
     BatchAssignModelsToProviderRequest,
     BatchAssignModelsToProviderResponse,
+    ImportFromUpstreamRequest,
+    ImportFromUpstreamResponse,
+    ImportFromUpstreamSuccessItem,
+    ImportFromUpstreamErrorItem,
+    ProviderAvailableSourceModel,
+    ProviderAvailableSourceModelsResponse,
 )
 from src.models.database import (
     GlobalModel,
     Model,
-    ModelMapping,
     Provider,
 )
-from src.models.pydantic_models import (
-    ProviderAvailableSourceModel,
-    ProviderAvailableSourceModelsResponse,
-)
 from src.services.model.service import ModelService
 
 router = APIRouter(tags=["Model Management"])
@@ -136,8 +137,7 @@ async def get_provider_available_source_models(
     获取该 Provider 支持的所有统一模型名（source_model）
 
     包括：
-    1. 通过 ModelMapping 映射的模型
-    2. 直连模型（Model.provider_model_name 直接作为统一模型名）
+    1. 直连模型（Model.provider_model_name 直接作为统一模型名）
     """
     adapter = AdminGetProviderAvailableSourceModelsAdapter(provider_id=provider_id)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
@@ -160,6 +160,28 @@ async def batch_assign_global_models_to_provider(
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
+@router.post(
+    "/{provider_id}/import-from-upstream",
+    response_model=ImportFromUpstreamResponse,
+)
+async def import_models_from_upstream(
+    provider_id: str,
+    payload: ImportFromUpstreamRequest,
+    request: Request,
+    db: Session = Depends(get_db),
+) -> ImportFromUpstreamResponse:
+    """
+    从上游提供商导入模型
+
+    流程：
+    1. 根据 model_ids 检查全局模型是否存在（按 name 匹配）
+    2. 如不存在，自动创建新的 GlobalModel（使用默认配置）
+    3. 创建 Model 关联到当前 Provider
+    """
+    adapter = AdminImportFromUpstreamAdapter(provider_id=provider_id, payload=payload)
+    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
+
+
 # -------- Adapters --------
 
 
@@ -294,10 +316,9 @@ class AdminGetProviderAvailableSourceModelsAdapter(AdminApiAdapter):
         """
         返回 Provider 支持的所有 GlobalModel
 
-        方案 A 逻辑：
+        逻辑：
         1. 查询该 Provider 的所有 Model
         2. 通过 Model.global_model_id 获取 GlobalModel
-        3. 查询所有指向该 GlobalModel 的别名（ModelMapping.alias）
         """
         db = context.db
         provider = db.query(Provider).filter(Provider.id == self.provider_id).first()
@@ -324,27 +345,10 @@ class AdminGetProviderAvailableSourceModelsAdapter(AdminApiAdapter):
 
             # 如果该 GlobalModel 还未处理，初始化
             if global_model_name not in global_models_dict:
-                # 查询指向该 GlobalModel 的所有别名/映射
-                alias_rows = (
-                    db.query(ModelMapping.source_model)
-                    .filter(
-                        ModelMapping.target_global_model_id == global_model.id,
-                        ModelMapping.is_active == True,
-                        or_(
-                            ModelMapping.provider_id == self.provider_id,
-                            ModelMapping.provider_id.is_(None),
-                        ),
-                    )
-                    .all()
-                )
-                alias_list = [alias[0] for alias in alias_rows]
-
                 global_models_dict[global_model_name] = {
                     "global_model_name": global_model_name,
                     "display_name": global_model.display_name,
                     "provider_model_name": model.provider_model_name,
-                    "has_alias": len(alias_list) > 0,
-                    "aliases": alias_list,
                     "model_id": model.id,
                     "price": {
                         "input_price_per_1m": model.get_effective_input_price(),
@@ -440,4 +444,135 @@ class AdminBatchAssignModelsToProviderAdapter(AdminApiAdapter):
             f"Batch assigned {len(success)} GlobalModels to provider {provider.name} by {context.user.username}"
         )
 
+        # 清除 /v1/models 列表缓存
+        if success:
+            await invalidate_models_list_cache()
+
         return BatchAssignModelsToProviderResponse(success=success, errors=errors)
+
+
+@dataclass
+class AdminImportFromUpstreamAdapter(AdminApiAdapter):
+    """从上游提供商导入模型"""
+
+    provider_id: str
+    payload: ImportFromUpstreamRequest
+
+    async def handle(self, context):  # type: ignore[override]
+        db = context.db
+        provider = db.query(Provider).filter(Provider.id == self.provider_id).first()
+        if not provider:
+            raise NotFoundException("Provider not found", "provider")
+
+        success: list[ImportFromUpstreamSuccessItem] = []
+        errors: list[ImportFromUpstreamErrorItem] = []
+
+        # 默认阶梯计费配置（免费）
+        default_tiered_pricing = {
+            "tiers": [
+                {
+                    "up_to": None,
+                    "input_price_per_1m": 0.0,
+                    "output_price_per_1m": 0.0,
+                }
+            ]
+        }
+
+        for model_id in self.payload.model_ids:
+            # 输入验证：检查 model_id 长度
+            if not model_id or len(model_id) > 100:
+                errors.append(
+                    ImportFromUpstreamErrorItem(
+                        model_id=model_id[:50] + "..." if model_id and len(model_id) > 50 else model_id or "<empty>",
+                        error="Invalid model_id: must be 1-100 characters",
+                    )
+                )
+                continue
+
+            try:
+                # 使用 savepoint 确保单个模型导入的原子性
+                savepoint = db.begin_nested()
+                try:
+                    # 1. 检查是否已存在同名的 GlobalModel
+                    global_model = (
+                        db.query(GlobalModel).filter(GlobalModel.name == model_id).first()
+                    )
+                    created_global_model = False
+
+                    if not global_model:
+                        # 2. 创建新的 GlobalModel
+                        global_model = GlobalModel(
+                            name=model_id,
+                            display_name=model_id,
+                            default_tiered_pricing=default_tiered_pricing,
+                            is_active=True,
+                        )
+                        db.add(global_model)
+                        db.flush()
+                        created_global_model = True
+                        logger.info(
+                            f"Created new GlobalModel: {model_id} during upstream import"
+                        )
+
+                    # 3. 检查是否已存在关联
+                    existing = (
+                        db.query(Model)
+                        .filter(
+                            Model.provider_id == self.provider_id,
+                            Model.global_model_id == global_model.id,
+                        )
+                        .first()
+                    )
+                    if existing:
+                        # 已存在关联，提交 savepoint 并记录成功
+                        savepoint.commit()
+                        success.append(
+                            ImportFromUpstreamSuccessItem(
+                                model_id=model_id,
+                                global_model_id=global_model.id,
+                                global_model_name=global_model.name,
+                                provider_model_id=existing.id,
+                                created_global_model=created_global_model,
+                            )
+                        )
+                        continue
+
+                    # 4. 创建新的 Model 记录
+                    new_model = Model(
+                        provider_id=self.provider_id,
+                        global_model_id=global_model.id,
+                        provider_model_name=global_model.name,
+                        is_active=True,
+                    )
+                    db.add(new_model)
+                    db.flush()
+
+                    # 提交 savepoint
+                    savepoint.commit()
+                    success.append(
+                        ImportFromUpstreamSuccessItem(
+                            model_id=model_id,
+                            global_model_id=global_model.id,
+                            global_model_name=global_model.name,
+                            provider_model_id=new_model.id,
+                            created_global_model=created_global_model,
+                        )
+                    )
+                except Exception as e:
+                    # 回滚到 savepoint
+                    savepoint.rollback()
+                    raise e
+            except Exception as e:
+                logger.error(f"Error importing model {model_id}: {e}")
+                errors.append(ImportFromUpstreamErrorItem(model_id=model_id, error=str(e)))
+
+        db.commit()
+        logger.info(
+            f"Imported {len(success)} models from upstream to provider {provider.name} by {context.user.username}"
+        )
+
+        # 清除 /v1/models 列表缓存
+        if success:
+            await invalidate_models_list_cache()
+
+        return ImportFromUpstreamResponse(success=success, errors=errors)

src/api/admin/system.py

@@ -91,6 +91,34 @@ async def get_api_formats(request: Request, db: Session = Depends(get_db)):
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
+@router.get("/config/export")
+async def export_config(request: Request, db: Session = Depends(get_db)):
+    """导出提供商和模型配置（管理员）"""
+    adapter = AdminExportConfigAdapter()
+    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
+
+
+@router.post("/config/import")
+async def import_config(request: Request, db: Session = Depends(get_db)):
+    """导入提供商和模型配置（管理员）"""
+    adapter = AdminImportConfigAdapter()
+    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
+
+
+@router.get("/users/export")
+async def export_users(request: Request, db: Session = Depends(get_db)):
+    """导出用户数据（管理员）"""
+    adapter = AdminExportUsersAdapter()
+    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
+
+
+@router.post("/users/import")
+async def import_users(request: Request, db: Session = Depends(get_db)):
+    """导入用户数据（管理员）"""
+    adapter = AdminImportUsersAdapter()
+    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
+
+
 # -------- 系统设置适配器 --------
 
 
@@ -310,3 +338,749 @@ class AdminGetApiFormatsAdapter(AdminApiAdapter):
             )
 
         return {"formats": formats}
+
+
+class AdminExportConfigAdapter(AdminApiAdapter):
+    async def handle(self, context):  # type: ignore[override]
+        """导出提供商和模型配置（解密数据）"""
+        from datetime import datetime, timezone
+
+        from src.core.crypto import crypto_service
+        from src.models.database import GlobalModel, Model, ProviderAPIKey, ProviderEndpoint
+
+        db = context.db
+
+        # 导出 GlobalModels
+        global_models = db.query(GlobalModel).all()
+        global_models_data = []
+        for gm in global_models:
+            global_models_data.append(
+                {
+                    "name": gm.name,
+                    "display_name": gm.display_name,
+                    "default_price_per_request": gm.default_price_per_request,
+                    "default_tiered_pricing": gm.default_tiered_pricing,
+                    "supported_capabilities": gm.supported_capabilities,
+                    "config": gm.config,
+                    "is_active": gm.is_active,
+                }
+            )
+
+        # 导出 Providers 及其关联数据
+        providers = db.query(Provider).all()
+        providers_data = []
+        for provider in providers:
+            # 导出 Endpoints
+            endpoints = (
+                db.query(ProviderEndpoint)
+                .filter(ProviderEndpoint.provider_id == provider.id)
+                .all()
+            )
+            endpoints_data = []
+            for ep in endpoints:
+                # 导出 Endpoint Keys
+                keys = (
+                    db.query(ProviderAPIKey).filter(ProviderAPIKey.endpoint_id == ep.id).all()
+                )
+                keys_data = []
+                for key in keys:
+                    # 解密 API Key
+                    try:
+                        decrypted_key = crypto_service.decrypt(key.api_key)
+                    except Exception:
+                        decrypted_key = ""
+
+                    keys_data.append(
+                        {
+                            "api_key": decrypted_key,
+                            "name": key.name,
+                            "note": key.note,
+                            "rate_multiplier": key.rate_multiplier,
+                            "internal_priority": key.internal_priority,
+                            "global_priority": key.global_priority,
+                            "max_concurrent": key.max_concurrent,
+                            "rate_limit": key.rate_limit,
+                            "daily_limit": key.daily_limit,
+                            "monthly_limit": key.monthly_limit,
+                            "allowed_models": key.allowed_models,
+                            "capabilities": key.capabilities,
+                            "is_active": key.is_active,
+                        }
+                    )
+
+                endpoints_data.append(
+                    {
+                        "api_format": ep.api_format,
+                        "base_url": ep.base_url,
+                        "headers": ep.headers,
+                        "timeout": ep.timeout,
+                        "max_retries": ep.max_retries,
+                        "max_concurrent": ep.max_concurrent,
+                        "rate_limit": ep.rate_limit,
+                        "is_active": ep.is_active,
+                        "custom_path": ep.custom_path,
+                        "config": ep.config,
+                        "keys": keys_data,
+                    }
+                )
+
+            # 导出 Provider Models
+            models = db.query(Model).filter(Model.provider_id == provider.id).all()
+            models_data = []
+            for model in models:
+                # 获取关联的 GlobalModel 名称
+                global_model = (
+                    db.query(GlobalModel).filter(GlobalModel.id == model.global_model_id).first()
+                )
+                models_data.append(
+                    {
+                        "global_model_name": global_model.name if global_model else None,
+                        "provider_model_name": model.provider_model_name,
+                        "provider_model_mappings": model.provider_model_mappings,
+                        "price_per_request": model.price_per_request,
+                        "tiered_pricing": model.tiered_pricing,
+                        "supports_vision": model.supports_vision,
+                        "supports_function_calling": model.supports_function_calling,
+                        "supports_streaming": model.supports_streaming,
+                        "supports_extended_thinking": model.supports_extended_thinking,
+                        "supports_image_generation": model.supports_image_generation,
+                        "is_active": model.is_active,
+                        "config": model.config,
+                    }
+                )
+
+            providers_data.append(
+                {
+                    "name": provider.name,
+                    "display_name": provider.display_name,
+                    "description": provider.description,
+                    "website": provider.website,
+                    "billing_type": provider.billing_type.value if provider.billing_type else None,
+                    "monthly_quota_usd": provider.monthly_quota_usd,
+                    "quota_reset_day": provider.quota_reset_day,
+                    "rpm_limit": provider.rpm_limit,
+                    "provider_priority": provider.provider_priority,
+                    "is_active": provider.is_active,
+                    "rate_limit": provider.rate_limit,
+                    "concurrent_limit": provider.concurrent_limit,
+                    "config": provider.config,
+                    "endpoints": endpoints_data,
+                    "models": models_data,
+                }
+            )
+
+        return {
+            "version": "1.0",
+            "exported_at": datetime.now(timezone.utc).isoformat(),
+            "global_models": global_models_data,
+            "providers": providers_data,
+        }
+
+
+MAX_IMPORT_SIZE = 10 * 1024 * 1024  # 10MB
+
+
+class AdminImportConfigAdapter(AdminApiAdapter):
+    async def handle(self, context):  # type: ignore[override]
+        """导入提供商和模型配置"""
+        import uuid
+        from datetime import datetime, timezone
+
+        from src.core.crypto import crypto_service
+        from src.core.enums import ProviderBillingType
+        from src.models.database import GlobalModel, Model, ProviderAPIKey, ProviderEndpoint
+
+        # 检查请求体大小
+        if context.raw_body and len(context.raw_body) > MAX_IMPORT_SIZE:
+            raise InvalidRequestException("请求体大小不能超过 10MB")
+
+        db = context.db
+        payload = context.ensure_json_body()
+
+        # 验证配置版本
+        version = payload.get("version")
+        if version != "1.0":
+            raise InvalidRequestException(f"不支持的配置版本: {version}")
+
+        # 获取导入选项
+        merge_mode = payload.get("merge_mode", "skip")  # skip, overwrite, error
+        global_models_data = payload.get("global_models", [])
+        providers_data = payload.get("providers", [])
+
+        stats = {
+            "global_models": {"created": 0, "updated": 0, "skipped": 0},
+            "providers": {"created": 0, "updated": 0, "skipped": 0},
+            "endpoints": {"created": 0, "updated": 0, "skipped": 0},
+            "keys": {"created": 0, "updated": 0, "skipped": 0},
+            "models": {"created": 0, "updated": 0, "skipped": 0},
+            "errors": [],
+        }
+
+        try:
+            # 导入 GlobalModels
+            global_model_map = {}  # name -> id 映射
+            for gm_data in global_models_data:
+                existing = (
+                    db.query(GlobalModel).filter(GlobalModel.name == gm_data["name"]).first()
+                )
+
+                if existing:
+                    global_model_map[gm_data["name"]] = existing.id
+                    if merge_mode == "skip":
+                        stats["global_models"]["skipped"] += 1
+                        continue
+                    elif merge_mode == "error":
+                        raise InvalidRequestException(
+                            f"GlobalModel '{gm_data['name']}' 已存在"
+                        )
+                    elif merge_mode == "overwrite":
+                        # 更新现有记录
+                        existing.display_name = gm_data.get(
+                            "display_name", existing.display_name
+                        )
+                        existing.default_price_per_request = gm_data.get(
+                            "default_price_per_request"
+                        )
+                        existing.default_tiered_pricing = gm_data.get(
+                            "default_tiered_pricing", existing.default_tiered_pricing
+                        )
+                        existing.supported_capabilities = gm_data.get(
+                            "supported_capabilities"
+                        )
+                        existing.config = gm_data.get("config")
+                        existing.is_active = gm_data.get("is_active", True)
+                        existing.updated_at = datetime.now(timezone.utc)
+                        stats["global_models"]["updated"] += 1
+                else:
+                    # 创建新记录
+                    new_gm = GlobalModel(
+                        id=str(uuid.uuid4()),
+                        name=gm_data["name"],
+                        display_name=gm_data.get("display_name", gm_data["name"]),
+                        default_price_per_request=gm_data.get("default_price_per_request"),
+                        default_tiered_pricing=gm_data.get(
+                            "default_tiered_pricing",
+                            {"tiers": [{"up_to": None, "input_price_per_1m": 0, "output_price_per_1m": 0}]},
+                        ),
+                        supported_capabilities=gm_data.get("supported_capabilities"),
+                        config=gm_data.get("config"),
+                        is_active=gm_data.get("is_active", True),
+                    )
+                    db.add(new_gm)
+                    db.flush()
+                    global_model_map[gm_data["name"]] = new_gm.id
+                    stats["global_models"]["created"] += 1
+
+            # 导入 Providers
+            for prov_data in providers_data:
+                existing_provider = (
+                    db.query(Provider).filter(Provider.name == prov_data["name"]).first()
+                )
+
+                if existing_provider:
+                    provider_id = existing_provider.id
+                    if merge_mode == "skip":
+                        stats["providers"]["skipped"] += 1
+                        # 仍然需要处理 endpoints 和 models（如果存在）
+                    elif merge_mode == "error":
+                        raise InvalidRequestException(
+                            f"Provider '{prov_data['name']}' 已存在"
+                        )
+                    elif merge_mode == "overwrite":
+                        # 更新现有记录
+                        existing_provider.display_name = prov_data.get(
+                            "display_name", existing_provider.display_name
+                        )
+                        existing_provider.description = prov_data.get("description")
+                        existing_provider.website = prov_data.get("website")
+                        if prov_data.get("billing_type"):
+                            existing_provider.billing_type = ProviderBillingType(
+                                prov_data["billing_type"]
+                            )
+                        existing_provider.monthly_quota_usd = prov_data.get(
+                            "monthly_quota_usd"
+                        )
+                        existing_provider.quota_reset_day = prov_data.get(
+                            "quota_reset_day", 30
+                        )
+                        existing_provider.rpm_limit = prov_data.get("rpm_limit")
+                        existing_provider.provider_priority = prov_data.get(
+                            "provider_priority", 100
+                        )
+                        existing_provider.is_active = prov_data.get("is_active", True)
+                        existing_provider.rate_limit = prov_data.get("rate_limit")
+                        existing_provider.concurrent_limit = prov_data.get(
+                            "concurrent_limit"
+                        )
+                        existing_provider.config = prov_data.get("config")
+                        existing_provider.updated_at = datetime.now(timezone.utc)
+                        stats["providers"]["updated"] += 1
+                else:
+                    # 创建新 Provider
+                    billing_type = ProviderBillingType.PAY_AS_YOU_GO
+                    if prov_data.get("billing_type"):
+                        billing_type = ProviderBillingType(prov_data["billing_type"])
+
+                    new_provider = Provider(
+                        id=str(uuid.uuid4()),
+                        name=prov_data["name"],
+                        display_name=prov_data.get("display_name", prov_data["name"]),
+                        description=prov_data.get("description"),
+                        website=prov_data.get("website"),
+                        billing_type=billing_type,
+                        monthly_quota_usd=prov_data.get("monthly_quota_usd"),
+                        quota_reset_day=prov_data.get("quota_reset_day", 30),
+                        rpm_limit=prov_data.get("rpm_limit"),
+                        provider_priority=prov_data.get("provider_priority", 100),
+                        is_active=prov_data.get("is_active", True),
+                        rate_limit=prov_data.get("rate_limit"),
+                        concurrent_limit=prov_data.get("concurrent_limit"),
+                        config=prov_data.get("config"),
+                    )
+                    db.add(new_provider)
+                    db.flush()
+                    provider_id = new_provider.id
+                    stats["providers"]["created"] += 1
+
+                # 导入 Endpoints
+                for ep_data in prov_data.get("endpoints", []):
+                    existing_ep = (
+                        db.query(ProviderEndpoint)
+                        .filter(
+                            ProviderEndpoint.provider_id == provider_id,
+                            ProviderEndpoint.api_format == ep_data["api_format"],
+                        )
+                        .first()
+                    )
+
+                    if existing_ep:
+                        endpoint_id = existing_ep.id
+                        if merge_mode == "skip":
+                            stats["endpoints"]["skipped"] += 1
+                        elif merge_mode == "error":
+                            raise InvalidRequestException(
+                                f"Endpoint '{ep_data['api_format']}' 已存在于 Provider '{prov_data['name']}'"
+                            )
+                        elif merge_mode == "overwrite":
+                            existing_ep.base_url = ep_data.get(
+                                "base_url", existing_ep.base_url
+                            )
+                            existing_ep.headers = ep_data.get("headers")
+                            existing_ep.timeout = ep_data.get("timeout", 300)
+                            existing_ep.max_retries = ep_data.get("max_retries", 3)
+                            existing_ep.max_concurrent = ep_data.get("max_concurrent")
+                            existing_ep.rate_limit = ep_data.get("rate_limit")
+                            existing_ep.is_active = ep_data.get("is_active", True)
+                            existing_ep.custom_path = ep_data.get("custom_path")
+                            existing_ep.config = ep_data.get("config")
+                            existing_ep.updated_at = datetime.now(timezone.utc)
+                            stats["endpoints"]["updated"] += 1
+                    else:
+                        new_ep = ProviderEndpoint(
+                            id=str(uuid.uuid4()),
+                            provider_id=provider_id,
+                            api_format=ep_data["api_format"],
+                            base_url=ep_data["base_url"],
+                            headers=ep_data.get("headers"),
+                            timeout=ep_data.get("timeout", 300),
+                            max_retries=ep_data.get("max_retries", 3),
+                            max_concurrent=ep_data.get("max_concurrent"),
+                            rate_limit=ep_data.get("rate_limit"),
+                            is_active=ep_data.get("is_active", True),
+                            custom_path=ep_data.get("custom_path"),
+                            config=ep_data.get("config"),
+                        )
+                        db.add(new_ep)
+                        db.flush()
+                        endpoint_id = new_ep.id
+                        stats["endpoints"]["created"] += 1
+
+                    # 导入 Keys
+                    # 获取当前 endpoint 下所有已有的 keys，用于去重
+                    existing_keys = (
+                        db.query(ProviderAPIKey)
+                        .filter(ProviderAPIKey.endpoint_id == endpoint_id)
+                        .all()
+                    )
+                    # 解密已有 keys 用于比对
+                    existing_key_values = set()
+                    for ek in existing_keys:
+                        try:
+                            decrypted = crypto_service.decrypt(ek.api_key)
+                            existing_key_values.add(decrypted)
+                        except Exception:
+                            pass
+
+                    for key_data in ep_data.get("keys", []):
+                        if not key_data.get("api_key"):
+                            stats["errors"].append(
+                                f"跳过空 API Key (Endpoint: {ep_data['api_format']})"
+                            )
+                            continue
+
+                        # 检查是否已存在相同的 Key（通过明文比对）
+                        if key_data["api_key"] in existing_key_values:
+                            stats["keys"]["skipped"] += 1
+                            continue
+
+                        encrypted_key = crypto_service.encrypt(key_data["api_key"])
+
+                        new_key = ProviderAPIKey(
+                            id=str(uuid.uuid4()),
+                            endpoint_id=endpoint_id,
+                            api_key=encrypted_key,
+                            name=key_data.get("name"),
+                            note=key_data.get("note"),
+                            rate_multiplier=key_data.get("rate_multiplier", 1.0),
+                            internal_priority=key_data.get("internal_priority", 100),
+                            global_priority=key_data.get("global_priority"),
+                            max_concurrent=key_data.get("max_concurrent"),
+                            rate_limit=key_data.get("rate_limit"),
+                            daily_limit=key_data.get("daily_limit"),
+                            monthly_limit=key_data.get("monthly_limit"),
+                            allowed_models=key_data.get("allowed_models"),
+                            capabilities=key_data.get("capabilities"),
+                            is_active=key_data.get("is_active", True),
+                        )
+                        db.add(new_key)
+                        # 添加到已有集合，防止同一批导入中重复
+                        existing_key_values.add(key_data["api_key"])
+                        stats["keys"]["created"] += 1
+
+                # 导入 Models
+                for model_data in prov_data.get("models", []):
+                    global_model_name = model_data.get("global_model_name")
+                    if not global_model_name:
+                        stats["errors"].append(
+                            f"跳过无 global_model_name 的模型 (Provider: {prov_data['name']})"
+                        )
+                        continue
+
+                    global_model_id = global_model_map.get(global_model_name)
+                    if not global_model_id:
+                        # 尝试从数据库查找
+                        existing_gm = (
+                            db.query(GlobalModel)
+                            .filter(GlobalModel.name == global_model_name)
+                            .first()
+                        )
+                        if existing_gm:
+                            global_model_id = existing_gm.id
+                        else:
+                            stats["errors"].append(
+                                f"GlobalModel '{global_model_name}' 不存在，跳过模型"
+                            )
+                            continue
+
+                    existing_model = (
+                        db.query(Model)
+                        .filter(
+                            Model.provider_id == provider_id,
+                            Model.provider_model_name == model_data["provider_model_name"],
+                        )
+                        .first()
+                    )
+
+                    if existing_model:
+                        if merge_mode == "skip":
+                            stats["models"]["skipped"] += 1
+                        elif merge_mode == "error":
+                            raise InvalidRequestException(
+                                f"Model '{model_data['provider_model_name']}' 已存在于 Provider '{prov_data['name']}'"
+                            )
+                        elif merge_mode == "overwrite":
+                            existing_model.global_model_id = global_model_id
+                            existing_model.provider_model_mappings = model_data.get(
+                                "provider_model_mappings"
+                            )
+                            existing_model.price_per_request = model_data.get(
+                                "price_per_request"
+                            )
+                            existing_model.tiered_pricing = model_data.get(
+                                "tiered_pricing"
+                            )
+                            existing_model.supports_vision = model_data.get(
+                                "supports_vision"
+                            )
+                            existing_model.supports_function_calling = model_data.get(
+                                "supports_function_calling"
+                            )
+                            existing_model.supports_streaming = model_data.get(
+                                "supports_streaming"
+                            )
+                            existing_model.supports_extended_thinking = model_data.get(
+                                "supports_extended_thinking"
+                            )
+                            existing_model.supports_image_generation = model_data.get(
+                                "supports_image_generation"
+                            )
+                            existing_model.is_active = model_data.get("is_active", True)
+                            existing_model.config = model_data.get("config")
+                            existing_model.updated_at = datetime.now(timezone.utc)
+                            stats["models"]["updated"] += 1
+                    else:
+                        new_model = Model(
+                            id=str(uuid.uuid4()),
+                            provider_id=provider_id,
+                            global_model_id=global_model_id,
+                            provider_model_name=model_data["provider_model_name"],
+                            provider_model_mappings=model_data.get(
+                                "provider_model_mappings"
+                            ),
+                            price_per_request=model_data.get("price_per_request"),
+                            tiered_pricing=model_data.get("tiered_pricing"),
+                            supports_vision=model_data.get("supports_vision"),
+                            supports_function_calling=model_data.get(
+                                "supports_function_calling"
+                            ),
+                            supports_streaming=model_data.get("supports_streaming"),
+                            supports_extended_thinking=model_data.get(
+                                "supports_extended_thinking"
+                            ),
+                            supports_image_generation=model_data.get(
+                                "supports_image_generation"
+                            ),
+                            is_active=model_data.get("is_active", True),
+                            config=model_data.get("config"),
+                        )
+                        db.add(new_model)
+                        stats["models"]["created"] += 1
+
+            db.commit()
+
+            # 失效缓存
+            from src.services.cache.invalidation import get_cache_invalidation_service
+
+            cache_service = get_cache_invalidation_service()
+            cache_service.clear_all_caches()
+
+            return {
+                "message": "配置导入成功",
+                "stats": stats,
+            }
+
+        except InvalidRequestException:
+            db.rollback()
+            raise
+        except Exception as e:
+            db.rollback()
+            raise InvalidRequestException(f"导入失败: {str(e)}")
+
+
+class AdminExportUsersAdapter(AdminApiAdapter):
+    async def handle(self, context):  # type: ignore[override]
+        """导出用户数据（保留加密数据，排除管理员）"""
+        from datetime import datetime, timezone
+
+        from src.core.enums import UserRole
+        from src.models.database import ApiKey, User
+
+        db = context.db
+
+        # 导出 Users（排除管理员）
+        users = db.query(User).filter(
+            User.is_deleted.is_(False),
+            User.role != UserRole.ADMIN
+        ).all()
+        users_data = []
+        for user in users:
+            # 导出用户的 API Keys（保留加密数据）
+            api_keys = db.query(ApiKey).filter(ApiKey.user_id == user.id).all()
+            api_keys_data = []
+            for key in api_keys:
+                api_keys_data.append(
+                    {
+                        "key_hash": key.key_hash,
+                        "key_encrypted": key.key_encrypted,
+                        "name": key.name,
+                        "is_standalone": key.is_standalone,
+                        "balance_used_usd": key.balance_used_usd,
+                        "current_balance_usd": key.current_balance_usd,
+                        "allowed_providers": key.allowed_providers,
+                        "allowed_endpoints": key.allowed_endpoints,
+                        "allowed_api_formats": key.allowed_api_formats,
+                        "allowed_models": key.allowed_models,
+                        "rate_limit": key.rate_limit,
+                        "concurrent_limit": key.concurrent_limit,
+                        "force_capabilities": key.force_capabilities,
+                        "is_active": key.is_active,
+                        "auto_delete_on_expiry": key.auto_delete_on_expiry,
+                        "total_requests": key.total_requests,
+                        "total_cost_usd": key.total_cost_usd,
+                    }
+                )
+
+            users_data.append(
+                {
+                    "email": user.email,
+                    "username": user.username,
+                    "password_hash": user.password_hash,
+                    "role": user.role.value if user.role else "user",
+                    "allowed_providers": user.allowed_providers,
+                    "allowed_endpoints": user.allowed_endpoints,
+                    "allowed_models": user.allowed_models,
+                    "model_capability_settings": user.model_capability_settings,
+                    "quota_usd": user.quota_usd,
+                    "used_usd": user.used_usd,
+                    "total_usd": user.total_usd,
+                    "is_active": user.is_active,
+                    "api_keys": api_keys_data,
+                }
+            )
+
+        return {
+            "version": "1.0",
+            "exported_at": datetime.now(timezone.utc).isoformat(),
+            "users": users_data,
+        }
+
+
+class AdminImportUsersAdapter(AdminApiAdapter):
+    async def handle(self, context):  # type: ignore[override]
+        """导入用户数据"""
+        import uuid
+        from datetime import datetime, timezone
+
+        from src.core.enums import UserRole
+        from src.models.database import ApiKey, User
+
+        # 检查请求体大小
+        if context.raw_body and len(context.raw_body) > MAX_IMPORT_SIZE:
+            raise InvalidRequestException("请求体大小不能超过 10MB")
+
+        db = context.db
+        payload = context.ensure_json_body()
+
+        # 验证配置版本
+        version = payload.get("version")
+        if version != "1.0":
+            raise InvalidRequestException(f"不支持的配置版本: {version}")
+
+        # 获取导入选项
+        merge_mode = payload.get("merge_mode", "skip")  # skip, overwrite, error
+        users_data = payload.get("users", [])
+
+        stats = {
+            "users": {"created": 0, "updated": 0, "skipped": 0},
+            "api_keys": {"created": 0, "skipped": 0},
+            "errors": [],
+        }
+
+        try:
+            for user_data in users_data:
+                # 跳过管理员角色的导入（不区分大小写）
+                role_str = str(user_data.get("role", "")).lower()
+                if role_str == "admin":
+                    stats["errors"].append(f"跳过管理员用户: {user_data.get('email')}")
+                    stats["users"]["skipped"] += 1
+                    continue
+
+                existing_user = (
+                    db.query(User).filter(User.email == user_data["email"]).first()
+                )
+
+                if existing_user:
+                    user_id = existing_user.id
+                    if merge_mode == "skip":
+                        stats["users"]["skipped"] += 1
+                    elif merge_mode == "error":
+                        raise InvalidRequestException(
+                            f"用户 '{user_data['email']}' 已存在"
+                        )
+                    elif merge_mode == "overwrite":
+                        # 更新现有用户
+                        existing_user.username = user_data.get(
+                            "username", existing_user.username
+                        )
+                        if user_data.get("password_hash"):
+                            existing_user.password_hash = user_data["password_hash"]
+                        if user_data.get("role"):
+                            existing_user.role = UserRole(user_data["role"])
+                        existing_user.allowed_providers = user_data.get("allowed_providers")
+                        existing_user.allowed_endpoints = user_data.get("allowed_endpoints")
+                        existing_user.allowed_models = user_data.get("allowed_models")
+                        existing_user.model_capability_settings = user_data.get(
+                            "model_capability_settings"
+                        )
+                        existing_user.quota_usd = user_data.get("quota_usd")
+                        existing_user.used_usd = user_data.get("used_usd", 0.0)
+                        existing_user.total_usd = user_data.get("total_usd", 0.0)
+                        existing_user.is_active = user_data.get("is_active", True)
+                        existing_user.updated_at = datetime.now(timezone.utc)
+                        stats["users"]["updated"] += 1
+                else:
+                    # 创建新用户
+                    role = UserRole.USER
+                    if user_data.get("role"):
+                        role = UserRole(user_data["role"])
+
+                    new_user = User(
+                        id=str(uuid.uuid4()),
+                        email=user_data["email"],
+                        username=user_data.get("username", user_data["email"].split("@")[0]),
+                        password_hash=user_data.get("password_hash", ""),
+                        role=role,
+                        allowed_providers=user_data.get("allowed_providers"),
+                        allowed_endpoints=user_data.get("allowed_endpoints"),
+                        allowed_models=user_data.get("allowed_models"),
+                        model_capability_settings=user_data.get("model_capability_settings"),
+                        quota_usd=user_data.get("quota_usd"),
+                        used_usd=user_data.get("used_usd", 0.0),
+                        total_usd=user_data.get("total_usd", 0.0),
+                        is_active=user_data.get("is_active", True),
+                    )
+                    db.add(new_user)
+                    db.flush()
+                    user_id = new_user.id
+                    stats["users"]["created"] += 1
+
+                # 导入 API Keys
+                for key_data in user_data.get("api_keys", []):
+                    # 检查是否已存在相同的 key_hash
+                    if key_data.get("key_hash"):
+                        existing_key = (
+                            db.query(ApiKey)
+                            .filter(ApiKey.key_hash == key_data["key_hash"])
+                            .first()
+                        )
+                        if existing_key:
+                            stats["api_keys"]["skipped"] += 1
+                            continue
+
+                    new_key = ApiKey(
+                        id=str(uuid.uuid4()),
+                        user_id=user_id,
+                        key_hash=key_data.get("key_hash", ""),
+                        key_encrypted=key_data.get("key_encrypted"),
+                        name=key_data.get("name"),
+                        is_standalone=key_data.get("is_standalone", False),
+                        balance_used_usd=key_data.get("balance_used_usd", 0.0),
+                        current_balance_usd=key_data.get("current_balance_usd"),
+                        allowed_providers=key_data.get("allowed_providers"),
+                        allowed_endpoints=key_data.get("allowed_endpoints"),
+                        allowed_api_formats=key_data.get("allowed_api_formats"),
+                        allowed_models=key_data.get("allowed_models"),
+                        rate_limit=key_data.get("rate_limit", 100),
+                        concurrent_limit=key_data.get("concurrent_limit", 5),
+                        force_capabilities=key_data.get("force_capabilities"),
+                        is_active=key_data.get("is_active", True),
+                        auto_delete_on_expiry=key_data.get("auto_delete_on_expiry", False),
+                        total_requests=key_data.get("total_requests", 0),
+                        total_cost_usd=key_data.get("total_cost_usd", 0.0),
+                    )
+                    db.add(new_key)
+                    stats["api_keys"]["created"] += 1
+
+            db.commit()
+
+            return {
+                "message": "用户数据导入成功",
+                "stats": stats,
+            }
+
+        except InvalidRequestException:
+            db.rollback()
+            raise
+        except Exception as e:
+            db.rollback()
+            raise InvalidRequestException(f"导入失败: {str(e)}")

src/api/admin/usage/routes.py

@@ -628,6 +628,7 @@ class AdminUsageRecordsAdapter(AdminApiAdapter):
                     "actual_cost": actual_cost,
                     "rate_multiplier": rate_multiplier,
                     "response_time_ms": usage.response_time_ms,
+                    "first_byte_time_ms": usage.first_byte_time_ms,  # 首字时间 (TTFB)
                     "created_at": usage.created_at.isoformat(),
                     "is_stream": usage.is_stream,
                     "input_price_per_1m": usage.input_price_per_1m,
@@ -738,6 +739,7 @@ class AdminUsageDetailAdapter(AdminApiAdapter):
             "status_code": usage_record.status_code,
             "error_message": usage_record.error_message,
             "response_time_ms": usage_record.response_time_ms,
+            "first_byte_time_ms": usage_record.first_byte_time_ms,  # 首字时间 (TTFB)
             "created_at": usage_record.created_at.isoformat() if usage_record.created_at else None,
             "request_headers": usage_record.request_headers,
             "request_body": usage_record.get_request_body(),

src/api/announcements/routes.py

@@ -140,9 +140,9 @@ class AnnouncementOptionalAuthAdapter(ApiAdapter):
         if not authorization or not authorization.lower().startswith("bearer "):
             return None
 
-        token = authorization.replace("Bearer ", "").strip()
+        token = authorization[7:].strip()
         try:
-            payload = await AuthService.verify_token(token)
+            payload = await AuthService.verify_token(token, token_type="access")
             user_id = payload.get("user_id")
             if not user_id:
                 return None

src/api/auth/routes.py

@@ -211,7 +211,7 @@ class AuthRefreshAdapter(AuthPublicAdapter):
 
 class AuthRegisterAdapter(AuthPublicAdapter):
     async def handle(self, context):  # type: ignore[override]
-        from ..models.database import SystemConfig
+        from src.models.database import SystemConfig
 
         db = context.db
         payload = context.ensure_json_body()

src/api/base/models_service.py

@@ -0,0 +1,403 @@
+"""
+公共模型查询服务
+
+为 Claude/OpenAI/Gemini 的 /models 端点提供统一的查询逻辑
+
+查询逻辑:
+1. 找到指定 api_format 的活跃端点
+2. 端点下有活跃的 Key
+3. Provider 关联了该模型（Model 表）
+4. Key 的 allowed_models 允许该模型（null = 允许所有）
+"""
+
+from dataclasses import asdict, dataclass
+from typing import Any, Optional
+
+from sqlalchemy.orm import Session, joinedload
+
+from src.config.constants import CacheTTL
+from src.core.cache_service import CacheService
+from src.core.logger import logger
+from src.models.database import GlobalModel, Model, Provider, ProviderAPIKey, ProviderEndpoint
+
+# 缓存 key 前缀
+_CACHE_KEY_PREFIX = "models:list"
+_CACHE_TTL = CacheTTL.MODEL  # 300 秒
+
+
+def _get_cache_key(api_formats: list[str]) -> str:
+    """生成缓存 key"""
+    formats_str = ",".join(sorted(api_formats))
+    return f"{_CACHE_KEY_PREFIX}:{formats_str}"
+
+
+async def _get_cached_models(api_formats: list[str]) -> Optional[list["ModelInfo"]]:
+    """从缓存获取模型列表"""
+    cache_key = _get_cache_key(api_formats)
+    try:
+        cached = await CacheService.get(cache_key)
+        if cached:
+            logger.debug(f"[ModelsService] 缓存命中: {cache_key}, {len(cached)} 个模型")
+            return [ModelInfo(**item) for item in cached]
+    except Exception as e:
+        logger.warning(f"[ModelsService] 缓存读取失败: {e}")
+    return None
+
+
+async def _set_cached_models(api_formats: list[str], models: list["ModelInfo"]) -> None:
+    """将模型列表写入缓存"""
+    cache_key = _get_cache_key(api_formats)
+    try:
+        data = [asdict(m) for m in models]
+        await CacheService.set(cache_key, data, ttl_seconds=_CACHE_TTL)
+        logger.debug(f"[ModelsService] 已缓存: {cache_key}, {len(models)} 个模型, TTL={_CACHE_TTL}s")
+    except Exception as e:
+        logger.warning(f"[ModelsService] 缓存写入失败: {e}")
+
+
+async def invalidate_models_list_cache() -> None:
+    """
+    清除所有 /v1/models 列表缓存
+
+    在模型创建、更新、删除时调用，确保模型列表实时更新
+    """
+    # 清除所有格式的缓存
+    all_formats = ["CLAUDE", "OPENAI", "GEMINI"]
+    for fmt in all_formats:
+        cache_key = f"{_CACHE_KEY_PREFIX}:{fmt}"
+        try:
+            await CacheService.delete(cache_key)
+            logger.debug(f"[ModelsService] 已清除缓存: {cache_key}")
+        except Exception as e:
+            logger.warning(f"[ModelsService] 清除缓存失败 {cache_key}: {e}")
+
+
+@dataclass
+class ModelInfo:
+    """统一的模型信息结构"""
+
+    id: str  # 模型 ID (GlobalModel.name 或 provider_model_name)
+    display_name: str
+    description: Optional[str]
+    created_at: Optional[str]  # ISO 格式
+    created_timestamp: int  # Unix 时间戳
+    provider_name: str
+    # 能力配置
+    streaming: bool = True
+    vision: bool = False
+    function_calling: bool = False
+    extended_thinking: bool = False
+    image_generation: bool = False
+    structured_output: bool = False
+    # 规格参数
+    context_limit: Optional[int] = None
+    output_limit: Optional[int] = None
+    # 元信息
+    family: Optional[str] = None
+    knowledge_cutoff: Optional[str] = None
+    input_modalities: Optional[list[str]] = None
+    output_modalities: Optional[list[str]] = None
+
+
+def get_available_provider_ids(db: Session, api_formats: list[str]) -> set[str]:
+    """
+    返回有可用端点的 Provider IDs
+
+    条件:
+    - 端点 api_format 匹配
+    - 端点是活跃的
+    - 端点下有活跃的 Key
+    """
+    rows = (
+        db.query(ProviderEndpoint.provider_id)
+        .join(ProviderAPIKey, ProviderAPIKey.endpoint_id == ProviderEndpoint.id)
+        .filter(
+            ProviderEndpoint.api_format.in_(api_formats),
+            ProviderEndpoint.is_active.is_(True),
+            ProviderAPIKey.is_active.is_(True),
+        )
+        .distinct()
+        .all()
+    )
+    return {row[0] for row in rows}
+
+
+def _get_available_model_ids_for_format(db: Session, api_formats: list[str]) -> set[str]:
+    """
+    获取指定格式下真正可用的模型 ID 集合
+
+    一个模型可用需满足:
+    1. 端点 api_format 匹配且活跃
+    2. 端点下有活跃的 Key
+    3. **该端点的 Provider 关联了该模型**
+    4. Key 的 allowed_models 允许该模型（null = 允许该 Provider 关联的所有模型）
+    """
+    # 查询所有匹配格式的活跃端点及其活跃 Key，同时获取 endpoint_id
+    endpoint_keys = (
+        db.query(
+            ProviderEndpoint.id.label("endpoint_id"),
+            ProviderEndpoint.provider_id,
+            ProviderAPIKey.allowed_models,
+        )
+        .join(ProviderAPIKey, ProviderAPIKey.endpoint_id == ProviderEndpoint.id)
+        .filter(
+            ProviderEndpoint.api_format.in_(api_formats),
+            ProviderEndpoint.is_active.is_(True),
+            ProviderAPIKey.is_active.is_(True),
+        )
+        .all()
+    )
+
+    if not endpoint_keys:
+        return set()
+
+    # 收集每个 (provider_id, endpoint_id) 对应的 allowed_models
+    # 使用 provider_id 作为 key，因为模型是关联到 Provider 的
+    provider_allowed_models: dict[str, list[Optional[list[str]]]] = {}
+    provider_ids_with_format: set[str] = set()
+
+    for endpoint_id, provider_id, allowed_models in endpoint_keys:
+        provider_ids_with_format.add(provider_id)
+        if provider_id not in provider_allowed_models:
+            provider_allowed_models[provider_id] = []
+        provider_allowed_models[provider_id].append(allowed_models)
+
+    # 只查询那些有匹配格式端点的 Provider 下的模型
+    models = (
+        db.query(Model)
+        .options(joinedload(Model.global_model))
+        .join(Provider)
+        .filter(
+            Model.provider_id.in_(provider_ids_with_format),
+            Model.is_active.is_(True),
+            Provider.is_active.is_(True),
+        )
+        .all()
+    )
+
+    available_model_ids: set[str] = set()
+
+    for model in models:
+        model_provider_id = model.provider_id
+        global_model = model.global_model
+        model_id = global_model.name if global_model else model.provider_model_name  # type: ignore
+
+        if not model_provider_id or not model_id:
+            continue
+
+        # 该模型的 Provider 必须有匹配格式的端点
+        if model_provider_id not in provider_ids_with_format:
+            continue
+
+        # 检查该 provider 下是否有 Key 允许这个模型
+        allowed_lists = provider_allowed_models.get(model_provider_id, [])
+        for allowed_models in allowed_lists:
+            if allowed_models is None:
+                # null = 允许该 Provider 关联的所有模型（已通过上面的查询限制）
+                available_model_ids.add(model_id)
+                break
+            elif model_id in allowed_models:
+                # 明确在允许列表中
+                available_model_ids.add(model_id)
+                break
+            elif global_model and model.provider_model_name in allowed_models:
+                # 也检查 provider_model_name
+                available_model_ids.add(model_id)
+                break
+
+    return available_model_ids
+
+
+def _extract_model_info(model: Any) -> ModelInfo:
+    """从 Model 对象提取 ModelInfo"""
+    global_model = model.global_model
+    model_id: str = global_model.name if global_model else model.provider_model_name
+    display_name: str = global_model.display_name if global_model else model.provider_model_name
+    created_at: Optional[str] = (
+        model.created_at.strftime("%Y-%m-%dT%H:%M:%SZ") if model.created_at else None
+    )
+    created_timestamp: int = int(model.created_at.timestamp()) if model.created_at else 0
+    provider_name: str = model.provider.name if model.provider else "unknown"
+
+    # 从 GlobalModel.config 提取配置信息
+    config: dict = {}
+    description: Optional[str] = None
+    if global_model:
+        config = global_model.config or {}
+        description = config.get("description")
+
+    return ModelInfo(
+        id=model_id,
+        display_name=display_name,
+        description=description,
+        created_at=created_at,
+        created_timestamp=created_timestamp,
+        provider_name=provider_name,
+        # 能力配置
+        streaming=config.get("streaming", True),
+        vision=config.get("vision", False),
+        function_calling=config.get("function_calling", False),
+        extended_thinking=config.get("extended_thinking", False),
+        image_generation=config.get("image_generation", False),
+        structured_output=config.get("structured_output", False),
+        # 规格参数
+        context_limit=config.get("context_limit"),
+        output_limit=config.get("output_limit"),
+        # 元信息
+        family=config.get("family"),
+        knowledge_cutoff=config.get("knowledge_cutoff"),
+        input_modalities=config.get("input_modalities"),
+        output_modalities=config.get("output_modalities"),
+    )
+
+
+async def list_available_models(
+    db: Session,
+    available_provider_ids: set[str],
+    api_formats: Optional[list[str]] = None,
+) -> list[ModelInfo]:
+    """
+    获取可用模型列表（已去重，带缓存）
+
+    Args:
+        db: 数据库会话
+        available_provider_ids: 有可用端点的 Provider ID 集合
+        api_formats: API 格式列表，用于检查 Key 的 allowed_models
+
+    Returns:
+        去重后的 ModelInfo 列表，按创建时间倒序
+    """
+    if not available_provider_ids:
+        return []
+
+    # 尝试从缓存获取
+    if api_formats:
+        cached = await _get_cached_models(api_formats)
+        if cached is not None:
+            return cached
+
+    # 如果提供了 api_formats，获取真正可用的模型 ID
+    available_model_ids: Optional[set[str]] = None
+    if api_formats:
+        available_model_ids = _get_available_model_ids_for_format(db, api_formats)
+        if not available_model_ids:
+            return []
+
+    query = (
+        db.query(Model)
+        .options(joinedload(Model.global_model), joinedload(Model.provider))
+        .join(Provider)
+        .filter(
+            Model.is_active.is_(True),
+            Provider.is_active.is_(True),
+            Model.provider_id.in_(available_provider_ids),
+        )
+        .order_by(Model.created_at.desc())
+    )
+    all_models = query.all()
+
+    result: list[ModelInfo] = []
+    seen_model_ids: set[str] = set()
+
+    for model in all_models:
+        info = _extract_model_info(model)
+
+        # 如果有 available_model_ids 限制，检查是否在其中
+        if available_model_ids is not None and info.id not in available_model_ids:
+            continue
+
+        if info.id in seen_model_ids:
+            continue
+        seen_model_ids.add(info.id)
+
+        result.append(info)
+
+    # 写入缓存
+    if api_formats:
+        await _set_cached_models(api_formats, result)
+
+    return result
+
+
+def find_model_by_id(
+    db: Session,
+    model_id: str,
+    available_provider_ids: set[str],
+    api_formats: Optional[list[str]] = None,
+) -> Optional[ModelInfo]:
+    """
+    按 ID 查找模型
+
+    查找顺序：
+    1. 先按 GlobalModel.name 查找
+    2. 如果没找到任何候选，再按 provider_model_name 查找
+    3. 如果有候选但都不可用，返回 None（不回退）
+
+    Args:
+        db: 数据库会话
+        model_id: 模型 ID
+        available_provider_ids: 有可用端点的 Provider ID 集合
+        api_formats: API 格式列表，用于检查 Key 的 allowed_models
+
+    Returns:
+        ModelInfo 或 None
+    """
+    if not available_provider_ids:
+        return None
+
+    # 如果提供了 api_formats，获取真正可用的模型 ID
+    available_model_ids: Optional[set[str]] = None
+    if api_formats:
+        available_model_ids = _get_available_model_ids_for_format(db, api_formats)
+        # 快速检查：如果目标模型不在可用列表中，直接返回 None
+        if available_model_ids is not None and model_id not in available_model_ids:
+            return None
+
+    # 先按 GlobalModel.name 查找
+    models_by_global = (
+        db.query(Model)
+        .options(joinedload(Model.global_model), joinedload(Model.provider))
+        .join(Provider)
+        .join(GlobalModel, Model.global_model_id == GlobalModel.id)
+        .filter(
+            GlobalModel.name == model_id,
+            Model.is_active.is_(True),
+            Provider.is_active.is_(True),
+        )
+        .order_by(Model.created_at.desc())
+        .all()
+    )
+
+    model = next(
+        (m for m in models_by_global if m.provider_id in available_provider_ids),
+        None,
+    )
+
+    # 如果有候选但都不可用，直接返回 None（不回退 provider_model_name）
+    if not model and models_by_global:
+        return None
+
+    # 如果找不到任何候选，按 provider_model_name 查找
+    if not model:
+        models_by_provider_name = (
+            db.query(Model)
+            .options(joinedload(Model.global_model), joinedload(Model.provider))
+            .join(Provider)
+            .filter(
+                Model.provider_model_name == model_id,
+                Model.is_active.is_(True),
+                Provider.is_active.is_(True),
+            )
+            .order_by(Model.created_at.desc())
+            .all()
+        )
+
+        model = next(
+            (m for m in models_by_provider_name if m.provider_id in available_provider_ids),
+            None,
+        )
+
+    if not model:
+        return None
+
+    return _extract_model_info(model)

src/api/base/pagination.py

@@ -1,7 +1,7 @@
 from __future__ import annotations
 
 from dataclasses import asdict, dataclass
-from typing import List, Sequence, Tuple, TypeVar
+from typing import Any, List, Sequence, Tuple, TypeVar
 
 from sqlalchemy.orm import Query
 
@@ -40,10 +40,10 @@ def paginate_sequence(
     return sliced, meta
 
 
-def build_pagination_payload(items: List[dict], meta: PaginationMeta, **extra) -> dict:
+def build_pagination_payload(items: List[dict], meta: PaginationMeta, **extra: Any) -> dict:
     """
     构建标准分页响应 payload。
     """
-    payload = {"items": items, "meta": meta.to_dict()}
+    payload: dict = {"items": items, "meta": meta.to_dict()}
     payload.update(extra)
     return payload

src/api/base/pipeline.py

@@ -5,13 +5,12 @@ from enum import Enum
 from typing import Any, Optional, Tuple
 
 from fastapi import HTTPException, Request
-from sqlalchemy.orm import Session, sessionmaker
+from sqlalchemy.orm import Session
 
 from src.core.exceptions import QuotaExceededException
 from src.core.logger import logger
 from src.models.database import ApiKey, AuditEventType, User, UserRole
 from src.services.auth.service import AuthService
-from src.services.cache.user_cache import UserCacheService
 from src.services.system.audit import AuditService
 from src.services.usage.service import UsageService
 
@@ -178,9 +177,9 @@ class ApiRequestPipeline:
         if not authorization or not authorization.lower().startswith("bearer "):
             raise HTTPException(status_code=401, detail="缺少管理员凭证")
 
-        token = authorization.replace("Bearer ", "").strip()
+        token = authorization[7:].strip()
         try:
-            payload = await self.auth_service.verify_token(token)
+            payload = await self.auth_service.verify_token(token, token_type="access")
         except HTTPException:
             raise
         except Exception as exc:
@@ -191,8 +190,8 @@ class ApiRequestPipeline:
         if not user_id:
             raise HTTPException(status_code=401, detail="无效的管理员令牌")
 
-        # 使用缓存查询用户
-        user = await UserCacheService.get_user_by_id(db, user_id)
+        # 直接查询数据库，确保返回的是当前 Session 绑定的对象
+        user = db.query(User).filter(User.id == user_id).first()
         if not user or not user.is_active:
             raise HTTPException(status_code=403, detail="用户不存在或已禁用")
 
@@ -205,9 +204,9 @@ class ApiRequestPipeline:
         if not authorization or not authorization.lower().startswith("bearer "):
             raise HTTPException(status_code=401, detail="缺少用户凭证")
 
-        token = authorization.replace("Bearer ", "").strip()
+        token = authorization[7:].strip()
         try:
-            payload = await self.auth_service.verify_token(token)
+            payload = await self.auth_service.verify_token(token, token_type="access")
         except HTTPException:
             raise
         except Exception as exc:
@@ -218,8 +217,8 @@ class ApiRequestPipeline:
         if not user_id:
             raise HTTPException(status_code=401, detail="无效的用户令牌")
 
-        # 使用缓存查询用户
-        user = await UserCacheService.get_user_by_id(db, user_id)
+        # 直接查询数据库，确保返回的是当前 Session 绑定的对象
+        user = db.query(User).filter(User.id == user_id).first()
         if not user or not user.is_active:
             raise HTTPException(status_code=403, detail="用户不存在或已禁用")
 
@@ -242,11 +241,15 @@ class ApiRequestPipeline:
         status_code: Optional[int] = None,
         error: Optional[str] = None,
     ) -> None:
+        """记录审计事件
+
+        事务策略：复用请求级 Session，不单独提交。
+        审计记录随主事务一起提交，由中间件统一管理。
+        """
         if not getattr(adapter, "audit_log_enabled", True):
             return
 
-        bind = context.db.get_bind()
-        if bind is None:
+        if context.db is None:
             return
 
         event_type = adapter.audit_success_event if success else adapter.audit_failure_event
@@ -266,11 +269,11 @@ class ApiRequestPipeline:
             error=error,
         )
 
-        SessionMaker = sessionmaker(bind=bind)
-        audit_session = SessionMaker()
         try:
+            # 复用请求级 Session，不创建新的连接
+            # 审计记录随主事务一起提交，由中间件统一管理
             self.audit_service.log_event(
-                db=audit_session,
+                db=context.db,
                 event_type=event_type,
                 description=f"{context.request.method} {context.request.url.path} via {adapter.name}",
                 user_id=context.user.id if context.user else None,
@@ -282,12 +285,9 @@ class ApiRequestPipeline:
                 error_message=error,
                 metadata=metadata,
             )
-            audit_session.commit()
         except Exception as exc:
-            audit_session.rollback()
+            # 审计失败不应影响主请求，仅记录警告
             logger.warning(f"[Audit] Failed to record event for adapter={adapter.name}: {exc}")
-        finally:
-            audit_session.close()
 
     def _build_audit_metadata(
         self,

src/api/dashboard/routes.py

@@ -13,7 +13,7 @@ from src.api.base.admin_adapter import AdminApiAdapter
 from src.api.base.pipeline import ApiRequestPipeline
 from src.core.enums import UserRole
 from src.database import get_db
-from src.models.database import ApiKey, Provider, RequestCandidate, StatsDaily, Usage
+from src.models.database import ApiKey, Provider, RequestCandidate, StatsDaily, StatsDailyModel, Usage
 from src.models.database import User as DBUser
 from src.services.system.stats_aggregator import StatsAggregatorService
 from src.utils.cache_decorator import cache_result
@@ -731,8 +731,15 @@ class DashboardDailyStatsAdapter(DashboardAdapter):
             )
             # stats_daily.date 存储的是业务日期对应的 UTC 开始时间
             # 需要转回业务时区再取日期，才能与日期序列匹配
+            def _to_business_date_str(value: datetime) -> str:
+                if value.tzinfo is None:
+                    value_utc = value.replace(tzinfo=timezone.utc)
+                else:
+                    value_utc = value.astimezone(timezone.utc)
+                return value_utc.astimezone(app_tz).date().isoformat()
+
             stats_map = {
-                stat.date.replace(tzinfo=timezone.utc).astimezone(app_tz).date().isoformat(): {
+                _to_business_date_str(stat.date): {
                     "requests": stat.total_requests,
                     "tokens": stat.input_tokens + stat.output_tokens + stat.cache_creation_tokens + stat.cache_read_tokens,
                     "cost": stat.total_cost,
@@ -790,6 +797,38 @@ class DashboardDailyStatsAdapter(DashboardAdapter):
                     "unique_providers": today_unique_providers,
                     "fallback_count": today_fallback_count,
                 }
+
+            # 历史预聚合缺失时兜底：按业务日范围实时计算（仅补最近少量缺失，避免全表扫描）
+            yesterday_date = today_local.date() - timedelta(days=1)
+            historical_end = min(end_date_local.date(), yesterday_date)
+            missing_dates: list[str] = []
+            cursor = start_date_local.date()
+            while cursor <= historical_end:
+                date_str = cursor.isoformat()
+                if date_str not in stats_map:
+                    missing_dates.append(date_str)
+                cursor += timedelta(days=1)
+
+            if missing_dates:
+                for date_str in missing_dates[-7:]:
+                    target_local = datetime.fromisoformat(date_str).replace(tzinfo=app_tz)
+                    computed = StatsAggregatorService.compute_daily_stats(db, target_local)
+                    stats_map[date_str] = {
+                        "requests": computed["total_requests"],
+                        "tokens": (
+                            computed["input_tokens"]
+                            + computed["output_tokens"]
+                            + computed["cache_creation_tokens"]
+                            + computed["cache_read_tokens"]
+                        ),
+                        "cost": computed["total_cost"],
+                        "avg_response_time": computed["avg_response_time_ms"] / 1000.0
+                        if computed["avg_response_time_ms"]
+                        else 0,
+                        "unique_models": computed["unique_models"],
+                        "unique_providers": computed["unique_providers"],
+                        "fallback_count": computed["fallback_count"],
+                    }
         else:
             # 普通用户：仍需实时查询（用户级预聚合可选）
             query = db.query(Usage).filter(
@@ -854,69 +893,172 @@ class DashboardDailyStatsAdapter(DashboardAdapter):
                 })
             current_date += timedelta(days=1)
 
-        # ==================== 模型统计（仍需实时查询）====================
-        model_query = db.query(Usage)
-        if not is_admin:
-            model_query = model_query.filter(Usage.user_id == user.id)
-        model_query = model_query.filter(
-            and_(Usage.created_at >= start_date, Usage.created_at <= end_date)
-        )
-
-        model_stats = (
-            model_query.with_entities(
-                Usage.model,
-                func.count(Usage.id).label("requests"),
-                func.sum(Usage.total_tokens).label("tokens"),
-                func.sum(Usage.total_cost_usd).label("cost"),
-                func.avg(Usage.response_time_ms).label("avg_response_time"),
+        # ==================== 模型统计 ====================
+        if is_admin:
+            # 管理员：使用预聚合数据 + 今日实时数据
+            # 历史数据从 stats_daily_model 获取
+            historical_model_stats = (
+                db.query(StatsDailyModel)
+                .filter(and_(StatsDailyModel.date >= start_date, StatsDailyModel.date < today))
+                .all()
             )
-            .group_by(Usage.model)
-            .order_by(func.sum(Usage.total_cost_usd).desc())
-            .all()
-        )
 
-        model_summary = [
-            {
-                "model": stat.model,
-                "requests": stat.requests or 0,
-                "tokens": int(stat.tokens or 0),
-                "cost": float(stat.cost or 0),
-                "avg_response_time": (
-                    float(stat.avg_response_time or 0) / 1000.0 if stat.avg_response_time else 0
-                ),
-                "cost_per_request": float(stat.cost or 0) / max(stat.requests or 1, 1),
-                "tokens_per_request": int(stat.tokens or 0) / max(stat.requests or 1, 1),
-            }
-            for stat in model_stats
-        ]
+            # 按模型汇总历史数据
+            model_agg: dict = {}
+            daily_breakdown: dict = {}
 
-        daily_model_stats = (
-            model_query.with_entities(
-                func.date(Usage.created_at).label("date"),
-                Usage.model,
-                func.count(Usage.id).label("requests"),
-                func.sum(Usage.total_tokens).label("tokens"),
-                func.sum(Usage.total_cost_usd).label("cost"),
+            for stat in historical_model_stats:
+                model = stat.model
+                if model not in model_agg:
+                    model_agg[model] = {
+                        "requests": 0, "tokens": 0, "cost": 0.0,
+                        "total_response_time": 0.0, "response_count": 0
+                    }
+                model_agg[model]["requests"] += stat.total_requests
+                tokens = (stat.input_tokens + stat.output_tokens +
+                          stat.cache_creation_tokens + stat.cache_read_tokens)
+                model_agg[model]["tokens"] += tokens
+                model_agg[model]["cost"] += stat.total_cost
+                if stat.avg_response_time_ms is not None:
+                    model_agg[model]["total_response_time"] += stat.avg_response_time_ms * stat.total_requests
+                    model_agg[model]["response_count"] += stat.total_requests
+
+                # 按日期分组
+                if stat.date.tzinfo is None:
+                    date_utc = stat.date.replace(tzinfo=timezone.utc)
+                else:
+                    date_utc = stat.date.astimezone(timezone.utc)
+                date_str = date_utc.astimezone(app_tz).date().isoformat()
+
+                daily_breakdown.setdefault(date_str, []).append({
+                    "model": model,
+                    "requests": stat.total_requests,
+                    "tokens": tokens,
+                    "cost": stat.total_cost,
+                })
+
+            # 今日实时模型统计
+            today_model_stats = (
+                db.query(
+                    Usage.model,
+                    func.count(Usage.id).label("requests"),
+                    func.sum(Usage.total_tokens).label("tokens"),
+                    func.sum(Usage.total_cost_usd).label("cost"),
+                    func.avg(Usage.response_time_ms).label("avg_response_time"),
+                )
+                .filter(Usage.created_at >= today)
+                .group_by(Usage.model)
+                .all()
             )
-            .group_by(func.date(Usage.created_at), Usage.model)
-            .order_by(func.date(Usage.created_at).desc(), func.sum(Usage.total_cost_usd).desc())
-            .all()
-        )
 
-        breakdown = {}
-        for stat in daily_model_stats:
-            date_str = stat.date.isoformat()
-            breakdown.setdefault(date_str, []).append(
+            today_str = today_local.date().isoformat()
+            for stat in today_model_stats:
+                model = stat.model
+                if model not in model_agg:
+                    model_agg[model] = {
+                        "requests": 0, "tokens": 0, "cost": 0.0,
+                        "total_response_time": 0.0, "response_count": 0
+                    }
+                model_agg[model]["requests"] += stat.requests or 0
+                model_agg[model]["tokens"] += int(stat.tokens or 0)
+                model_agg[model]["cost"] += float(stat.cost or 0)
+                if stat.avg_response_time is not None:
+                    model_agg[model]["total_response_time"] += float(stat.avg_response_time) * (stat.requests or 0)
+                    model_agg[model]["response_count"] += stat.requests or 0
+
+                # 今日 breakdown
+                daily_breakdown.setdefault(today_str, []).append({
+                    "model": model,
+                    "requests": stat.requests or 0,
+                    "tokens": int(stat.tokens or 0),
+                    "cost": float(stat.cost or 0),
+                })
+
+            # 构建 model_summary
+            model_summary = []
+            for model, agg in model_agg.items():
+                avg_rt = (agg["total_response_time"] / agg["response_count"] / 1000.0
+                          if agg["response_count"] > 0 else 0)
+                model_summary.append({
+                    "model": model,
+                    "requests": agg["requests"],
+                    "tokens": agg["tokens"],
+                    "cost": agg["cost"],
+                    "avg_response_time": avg_rt,
+                    "cost_per_request": agg["cost"] / max(agg["requests"], 1),
+                    "tokens_per_request": agg["tokens"] / max(agg["requests"], 1),
+                })
+            model_summary.sort(key=lambda x: x["cost"], reverse=True)
+
+            # 填充 model_breakdown
+            for item in formatted:
+                item["model_breakdown"] = daily_breakdown.get(item["date"], [])
+
+        else:
+            # 普通用户：实时查询（数据量较小）
+            model_query = db.query(Usage).filter(
+                and_(
+                    Usage.user_id == user.id,
+                    Usage.created_at >= start_date,
+                    Usage.created_at <= end_date
+                )
+            )
+
+            model_stats = (
+                model_query.with_entities(
+                    Usage.model,
+                    func.count(Usage.id).label("requests"),
+                    func.sum(Usage.total_tokens).label("tokens"),
+                    func.sum(Usage.total_cost_usd).label("cost"),
+                    func.avg(Usage.response_time_ms).label("avg_response_time"),
+                )
+                .group_by(Usage.model)
+                .order_by(func.sum(Usage.total_cost_usd).desc())
+                .all()
+            )
+
+            model_summary = [
                 {
                     "model": stat.model,
                     "requests": stat.requests or 0,
                     "tokens": int(stat.tokens or 0),
                     "cost": float(stat.cost or 0),
+                    "avg_response_time": (
+                        float(stat.avg_response_time or 0) / 1000.0 if stat.avg_response_time else 0
+                    ),
+                    "cost_per_request": float(stat.cost or 0) / max(stat.requests or 1, 1),
+                    "tokens_per_request": int(stat.tokens or 0) / max(stat.requests or 1, 1),
                 }
+                for stat in model_stats
+            ]
+
+            daily_model_stats = (
+                model_query.with_entities(
+                    func.date(Usage.created_at).label("date"),
+                    Usage.model,
+                    func.count(Usage.id).label("requests"),
+                    func.sum(Usage.total_tokens).label("tokens"),
+                    func.sum(Usage.total_cost_usd).label("cost"),
+                )
+                .group_by(func.date(Usage.created_at), Usage.model)
+                .order_by(func.date(Usage.created_at).desc(), func.sum(Usage.total_cost_usd).desc())
+                .all()
             )
 
-        for item in formatted:
-            item["model_breakdown"] = breakdown.get(item["date"], [])
+            breakdown = {}
+            for stat in daily_model_stats:
+                date_str = stat.date.isoformat()
+                breakdown.setdefault(date_str, []).append(
+                    {
+                        "model": stat.model,
+                        "requests": stat.requests or 0,
+                        "tokens": int(stat.tokens or 0),
+                        "cost": float(stat.cost or 0),
+                    }
+                )
+
+            for item in formatted:
+                item["model_breakdown"] = breakdown.get(item["date"], [])
 
         return {
             "daily_stats": formatted,

src/api/handlers/base/base_handler.py

@@ -28,7 +28,7 @@
 from __future__ import annotations
 
 import time
-from typing import Any, Callable, Dict, Optional, Protocol, runtime_checkable
+from typing import TYPE_CHECKING, Any, Callable, Dict, Optional, Protocol, runtime_checkable
 
 from fastapi import Request
 from fastapi.responses import JSONResponse, StreamingResponse
@@ -43,6 +43,9 @@ from src.services.provider.format import normalize_api_format
 from src.services.system.audit import audit_service
 from src.services.usage.service import UsageService
 
+if TYPE_CHECKING:
+    from src.api.handlers.base.stream_context import StreamContext
+
 
 
 class MessageTelemetry:
@@ -100,6 +103,8 @@ class MessageTelemetry:
         cache_read_tokens: int = 0,
         is_stream: bool = False,
         provider_request_headers: Optional[Dict[str, Any]] = None,
+        # 时间指标
+        first_byte_time_ms: Optional[int] = None,  # 首字时间/TTFB
         # Provider 侧追踪信息（用于记录真实成本）
         provider_id: Optional[str] = None,
         provider_endpoint_id: Optional[str] = None,
@@ -133,6 +138,7 @@ class MessageTelemetry:
             api_format=api_format,
             is_stream=is_stream,
             response_time_ms=response_time_ms,
+            first_byte_time_ms=first_byte_time_ms,  # 传递首字时间
             status_code=status_code,
             request_headers=request_headers,
             request_body=request_body,
@@ -395,3 +401,60 @@ class BaseMessageHandler:
 
         # 创建后台任务，不阻塞当前流
         asyncio.create_task(_do_update())
+
+    def _update_usage_to_streaming_with_ctx(self, ctx: "StreamContext") -> None:
+        """更新 Usage 状态为 streaming，同时更新 provider 和 target_model
+
+        使用 asyncio 后台任务执行数据库更新，避免阻塞流式传输
+
+        Args:
+            ctx: 流式上下文，包含 provider_name 和 mapped_model
+        """
+        import asyncio
+        from src.database.database import get_db
+
+        target_request_id = self.request_id
+        provider = ctx.provider_name
+        target_model = ctx.mapped_model
+
+        async def _do_update() -> None:
+            try:
+                db_gen = get_db()
+                db = next(db_gen)
+                try:
+                    UsageService.update_usage_status(
+                        db=db,
+                        request_id=target_request_id,
+                        status="streaming",
+                        provider=provider,
+                        target_model=target_model,
+                    )
+                finally:
+                    db.close()
+            except Exception as e:
+                logger.warning(f"[{target_request_id}] 更新 Usage 状态为 streaming 失败: {e}")
+
+        # 创建后台任务，不阻塞当前流
+        asyncio.create_task(_do_update())
+
+    def _log_request_error(self, message: str, error: Exception) -> None:
+        """记录请求错误日志，对业务异常不打印堆栈
+
+        Args:
+            message: 错误消息前缀
+            error: 异常对象
+        """
+        from src.core.exceptions import (
+            ProviderException,
+            QuotaExceededException,
+            RateLimitException,
+            ModelNotSupportedException,
+            UpstreamClientException,
+        )
+
+        if isinstance(error, (ProviderException, QuotaExceededException, RateLimitException, ModelNotSupportedException, UpstreamClientException)):
+            # 业务异常：简洁日志，不打印堆栈
+            logger.error(f"{message}: [{type(error).__name__}] {error}")
+        else:
+            # 未知异常：完整堆栈
+            logger.exception(f"{message}: {error}")

src/api/handlers/base/chat_adapter_base.py

@@ -19,8 +19,9 @@ Chat Adapter 通用基类
 import time
 import traceback
 from abc import abstractmethod
-from typing import Any, Dict, Optional, Type
+from typing import Any, Dict, Optional, Tuple, Type
 
+import httpx
 from fastapi import HTTPException
 from fastapi.responses import JSONResponse
 
@@ -64,18 +65,6 @@ class ChatAdapterBase(ApiAdapter):
 
     def __init__(self, allowed_api_formats: Optional[list[str]] = None):
         self.allowed_api_formats = allowed_api_formats or [self.FORMAT_ID]
-        self.response_normalizer = None
-        # 可选启用响应规范化
-        self._init_response_normalizer()
-
-    def _init_response_normalizer(self):
-        """初始化响应规范化器 - 子类可覆盖"""
-        try:
-            from src.services.provider.response_normalizer import ResponseNormalizer
-
-            self.response_normalizer = ResponseNormalizer()
-        except ImportError:
-            pass
 
     async def handle(self, context: ApiRequestContext):
         """处理 Chat API 请求"""
@@ -228,8 +217,6 @@ class ChatAdapterBase(ApiAdapter):
             user_agent=user_agent,
             start_time=start_time,
             allowed_api_formats=self.allowed_api_formats,
-            response_normalizer=self.response_normalizer,
-            enable_response_normalization=self.response_normalizer is not None,
             adapter_detector=self.detect_capability_requirements,
         )
 
@@ -634,6 +621,39 @@ class ChatAdapterBase(ApiAdapter):
         # 如果所有阶梯都有上限且都超过了，返回最后一个阶梯
         return tiers[-1] if tiers else None
 
+    # =========================================================================
+    # 模型列表查询 - 子类应覆盖此方法
+    # =========================================================================
+
+    @classmethod
+    async def fetch_models(
+        cls,
+        client: httpx.AsyncClient,
+        base_url: str,
+        api_key: str,
+        extra_headers: Optional[Dict[str, str]] = None,
+    ) -> Tuple[list, Optional[str]]:
+        """
+        查询上游 API 支持的模型列表
+
+        这是 Aether 内部发起的请求（非用户透传），用于：
+        - 管理后台查询提供商支持的模型
+        - 自动发现可用模型
+
+        Args:
+            client: httpx 异步客户端
+            base_url: API 基础 URL
+            api_key: API 密钥（已解密）
+            extra_headers: 端点配置的额外请求头
+
+        Returns:
+            (models, error): 模型列表和错误信息
+            - models: 模型信息列表，每个模型至少包含 id 字段
+            - error: 错误信息，成功时为 None
+        """
+        # 默认实现返回空列表，子类应覆盖
+        return [], f"{cls.FORMAT_ID} adapter does not implement fetch_models"
+
 
 # =========================================================================
 # Adapter 注册表 - 用于根据 API format 获取 Adapter 实例

src/api/handlers/base/chat_handler_base.py

@@ -34,6 +34,7 @@ from src.api.handlers.base.response_parser import ResponseParser
 from src.api.handlers.base.stream_context import StreamContext
 from src.api.handlers.base.stream_processor import StreamProcessor
 from src.api.handlers.base.stream_telemetry import StreamTelemetryRecorder
+from src.api.handlers.base.utils import build_sse_headers
 from src.config.settings import config
 from src.core.exceptions import (
     EmbeddedErrorException,
@@ -87,8 +88,6 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
         user_agent: str,
         start_time: float,
         allowed_api_formats: Optional[list] = None,
-        response_normalizer: Optional[Any] = None,
-        enable_response_normalization: bool = False,
         adapter_detector: Optional[Callable[[Dict[str, str], Optional[Dict[str, Any]]], Dict[str, bool]]] = None,
     ):
         allowed = allowed_api_formats or [self.FORMAT_ID]
@@ -105,8 +104,6 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
         )
         self._parser: Optional[ResponseParser] = None
         self._request_builder = PassthroughRequestBuilder()
-        self.response_normalizer = response_normalizer
-        self.enable_response_normalization = enable_response_normalization
 
     @property
     def parser(self) -> ResponseParser:
@@ -263,7 +260,13 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
         mapping = await mapper.get_mapping(source_model, provider_id)
 
         if mapping and mapping.model:
-            mapped_name = str(mapping.model.provider_model_name)
+            # 使用 select_provider_model_name 支持映射功能
+            # 传入 api_key.id 作为 affinity_key，实现相同用户稳定选择同一映射
+            # 传入 api_format 用于过滤适用的映射作用域
+            affinity_key = self.api_key.id if self.api_key else None
+            mapped_name = mapping.model.select_provider_model_name(
+                affinity_key, api_format=self.FORMAT_ID
+            )
             logger.debug(f"[Chat] 模型映射: {source_model} -> {mapped_name}")
             return mapped_name
 
@@ -290,11 +293,15 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
         # 创建类型安全的流式上下文
         ctx = StreamContext(model=model, api_format=api_format)
 
+        # 创建更新状态的回调闭包（可以访问 ctx）
+        def update_streaming_status() -> None:
+            self._update_usage_to_streaming_with_ctx(ctx)
+
         # 创建流处理器
         stream_processor = StreamProcessor(
             request_id=self.request_id,
             default_parser=self.parser,
-            on_streaming_start=self._update_usage_to_streaming,
+            on_streaming_start=update_streaming_status,
         )
 
         # 定义请求函数
@@ -362,7 +369,7 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
                 ctx,
                 original_headers,
                 original_request_body,
-                self.elapsed_ms(),
+                self.start_time,  # 传入开始时间，让 telemetry 在流结束后计算响应时间
             )
 
             # 创建监控流
@@ -375,11 +382,12 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
             return StreamingResponse(
                 monitored_stream,
                 media_type="text/event-stream",
+                headers=build_sse_headers(),
                 background=background_tasks,
             )
 
         except Exception as e:
-            logger.exception(f"流式请求失败: {e}")
+            self._log_request_error("流式请求失败", e)
             await self._record_stream_failure(ctx, e, original_headers, original_request_body)
             raise
 
@@ -458,7 +466,13 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
             pool=config.http_pool_timeout,
         )
 
-        http_client = httpx.AsyncClient(timeout=timeout_config, follow_redirects=True)
+        # 创建 HTTP 客户端（支持代理配置）
+        from src.clients.http_client import HTTPClientPool
+
+        http_client = HTTPClientPool.create_client_with_proxy(
+            proxy_config=endpoint.proxy,
+            timeout=timeout_config,
+        )
         try:
             response_ctx = http_client.stream(
                 "POST", url, json=provider_payload, headers=provider_headers
@@ -470,12 +484,13 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
 
             stream_response.raise_for_status()
 
-            # 创建行迭代器
-            line_iterator = stream_response.aiter_lines()
+            # 使用字节流迭代器（避免 aiter_lines 的性能问题）
+            # aiter_raw() 返回原始数据块，无缓冲，实现真正的流式传输
+            byte_iterator = stream_response.aiter_raw()
 
             # 预读检测嵌套错误
-            prefetched_lines = await stream_processor.prefetch_and_check_error(
-                line_iterator,
+            prefetched_chunks = await stream_processor.prefetch_and_check_error(
+                byte_iterator,
                 provider,
                 endpoint,
                 ctx,
@@ -500,13 +515,14 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
             await http_client.aclose()
             raise
 
-        # 创建流生成器
+        # 创建流生成器（传入字节流迭代器）
         return stream_processor.create_response_stream(
             ctx,
-            line_iterator,
+            byte_iterator,
             response_ctx,
             http_client,
-            prefetched_lines,
+            prefetched_chunks,
+            start_time=self.start_time,
         )
 
     async def _record_stream_failure(
@@ -623,11 +639,17 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
 
             logger.info(f"  [{self.request_id}] 发送非流式请求: Provider={provider.name}, "
                 f"模型={model} -> {mapped_model or '无映射'}")
+            logger.debug(f"  [{self.request_id}] 请求URL: {url}")
+            logger.debug(f"  [{self.request_id}] 请求体stream字段: {provider_payload.get('stream', 'N/A')}")
 
-            async with httpx.AsyncClient(
-                timeout=float(endpoint.timeout),
-                follow_redirects=True,
-            ) as http_client:
+            # 创建 HTTP 客户端（支持代理配置）
+            from src.clients.http_client import HTTPClientPool
+
+            http_client = HTTPClientPool.create_client_with_proxy(
+                proxy_config=endpoint.proxy,
+                timeout=httpx.Timeout(float(endpoint.timeout)),
+            )
+            async with http_client:
                 resp = await http_client.post(url, json=provider_payload, headers=provider_hdrs)
 
                 status_code = resp.status_code
@@ -642,10 +664,32 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
                         response_headers=response_headers,
                     )
                 elif resp.status_code >= 500:
-                    raise ProviderNotAvailableException(f"提供商服务不可用: {provider.name}")
-                elif resp.status_code != 200:
+                    # 记录响应体以便调试
+                    error_body = ""
+                    try:
+                        error_body = resp.text[:1000]
+                        logger.error(f"  [{self.request_id}] 上游返回5xx错误: status={resp.status_code}, body={error_body[:500]}")
+                    except Exception:
+                        pass
                     raise ProviderNotAvailableException(
-                        f"提供商返回错误: {provider.name}, 状态: {resp.status_code}"
+                        f"提供商服务不可用: {provider.name}",
+                        provider_name=str(provider.name),
+                        upstream_status=resp.status_code,
+                        upstream_response=error_body,
+                    )
+                elif resp.status_code != 200:
+                    # 记录非200响应以便调试
+                    error_body = ""
+                    try:
+                        error_body = resp.text[:1000]
+                        logger.warning(f"  [{self.request_id}] 上游返回非200: status={resp.status_code}, body={error_body[:500]}")
+                    except Exception:
+                        pass
+                    raise ProviderNotAvailableException(
+                        f"提供商返回错误: {provider.name}, 状态: {resp.status_code}",
+                        provider_name=str(provider.name),
+                        upstream_status=resp.status_code,
+                        upstream_response=error_body,
                     )
 
                 response_json = resp.json()

src/api/handlers/base/cli_adapter_base.py

@@ -17,8 +17,9 @@ CLI Adapter 通用基类
 
 import time
 import traceback
-from typing import Any, Dict, Optional, Type
+from typing import Any, Dict, Optional, Tuple, Type
 
+import httpx
 from fastapi import HTTPException
 from fastapi.responses import JSONResponse
 
@@ -580,6 +581,39 @@ class CliAdapterBase(ApiAdapter):
 
         return tiers[-1] if tiers else None
 
+    # =========================================================================
+    # 模型列表查询 - 子类应覆盖此方法
+    # =========================================================================
+
+    @classmethod
+    async def fetch_models(
+        cls,
+        client: httpx.AsyncClient,
+        base_url: str,
+        api_key: str,
+        extra_headers: Optional[Dict[str, str]] = None,
+    ) -> Tuple[list, Optional[str]]:
+        """
+        查询上游 API 支持的模型列表
+
+        这是 Aether 内部发起的请求（非用户透传），用于：
+        - 管理后台查询提供商支持的模型
+        - 自动发现可用模型
+
+        Args:
+            client: httpx 异步客户端
+            base_url: API 基础 URL
+            api_key: API 密钥（已解密）
+            extra_headers: 端点配置的额外请求头
+
+        Returns:
+            (models, error): 模型列表和错误信息
+            - models: 模型信息列表，每个模型至少包含 id 字段
+            - error: 错误信息，成功时为 None
+        """
+        # 默认实现返回空列表，子类应覆盖
+        return [], f"{cls.FORMAT_ID} adapter does not implement fetch_models"
+
 
 # =========================================================================
 # CLI Adapter 注册表 - 用于根据 API format 获取 CLI Adapter 实例

src/api/handlers/base/cli_handler_base.py

@@ -11,17 +11,15 @@ CLI Message Handler 通用基类
 """
 
 import asyncio
+import codecs
 import json
 import time
-from abc import abstractmethod
-from dataclasses import dataclass, field
 from typing import (
     Any,
     AsyncGenerator,
     Callable,
     Dict,
     Optional,
-    Tuple,
 )
 
 import httpx
@@ -35,6 +33,8 @@ from src.api.handlers.base.base_handler import (
 )
 from src.api.handlers.base.parsers import get_parser_for_format
 from src.api.handlers.base.request_builder import PassthroughRequestBuilder
+from src.api.handlers.base.stream_context import StreamContext
+from src.api.handlers.base.utils import build_sse_headers
 
 # 直接从具体模块导入，避免循环依赖
 from src.api.handlers.base.response_parser import (
@@ -61,63 +61,6 @@ from src.services.provider.transport import build_provider_url
 from src.utils.sse_parser import SSEEventParser
 
 
-
-@dataclass
-class StreamContext:
-    """流式请求的上下文信息"""
-
-    # 请求信息
-    model: str = "unknown"  # 用户请求的原始模型名
-    mapped_model: Optional[str] = None  # 映射后的目标模型名（如果发生了映射）
-    api_format: str = ""
-    request_id: str = ""
-
-    # 用户信息（提前提取避免 Session detached）
-    user_id: int = 0
-    api_key_id: int = 0
-
-    # 统计信息
-    input_tokens: int = 0
-    output_tokens: int = 0
-    cached_tokens: int = 0  # cache_read_input_tokens
-    cache_creation_tokens: int = 0  # cache_creation_input_tokens
-    collected_text: str = ""
-    response_id: Optional[str] = None
-    final_usage: Optional[Dict[str, Any]] = None
-    final_response: Optional[Dict[str, Any]] = None
-    parsed_chunks: list = field(default_factory=list)
-
-    # 流状态
-    start_time: float = field(default_factory=time.time)
-    chunk_count: int = 0
-    data_count: int = 0
-    has_completion: bool = False
-
-    # 响应信息
-    status_code: int = 200
-    response_headers: Dict[str, str] = field(default_factory=dict)
-
-    # 请求信息（发送给 Provider 的）
-    provider_request_headers: Dict[str, str] = field(default_factory=dict)
-    provider_request_body: Optional[Dict[str, Any]] = None  # 实际发送的请求体
-
-    # Provider 信息
-    provider_name: Optional[str] = None
-    provider_id: Optional[str] = None  # Provider ID（用于记录真实成本）
-    endpoint_id: Optional[str] = None
-    key_id: Optional[str] = None
-    attempt_id: Optional[str] = None
-    attempt_synced: bool = False
-    error_message: Optional[str] = None
-
-    # 格式转换信息
-    provider_api_format: str = ""  # Provider 的 API 格式（用于响应转换）
-    client_api_format: str = ""  # 客户端请求的 API 格式
-
-    # Provider 响应元数据（存储 provider 返回的额外信息，如 Gemini 的 modelVersion）
-    response_metadata: Dict[str, Any] = field(default_factory=dict)
-
-
 class CliMessageHandlerBase(BaseMessageHandler):
     """
     CLI 格式消息处理器基类
@@ -190,14 +133,17 @@ class CliMessageHandlerBase(BaseMessageHandler):
         """
         获取模型映射后的实际模型名
 
-        按优先级查找：映射 → 别名 → 直接匹配 GlobalModel
+        查找逻辑：
+        1. 直接通过 GlobalModel.name 匹配
+        2. 查找该 Provider 的 Model 实现
+        3. 使用 provider_model_name / provider_model_mappings 选择最终名称
 
         Args:
-            source_model: 用户请求的模型名（可能是别名）
+            source_model: 用户请求的模型名（必须是 GlobalModel.name）
             provider_id: Provider ID
 
         Returns:
-            映射后的 provider_model_name，如果没有找到映射则返回 None
+            映射后的 Provider 模型名，如果没有找到映射则返回 None
         """
         from src.services.model.mapper import ModelMapperMiddleware
 
@@ -207,7 +153,13 @@ class CliMessageHandlerBase(BaseMessageHandler):
         logger.debug(f"[CLI] _get_mapped_model: source={source_model}, provider={provider_id[:8]}..., mapping={mapping}")
 
         if mapping and mapping.model:
-            mapped_name = str(mapping.model.provider_model_name)
+            # 使用 select_provider_model_name 支持模型映射功能
+            # 传入 api_key.id 作为 affinity_key，实现相同用户稳定选择同一映射
+            # 传入 api_format 用于过滤适用的映射作用域
+            affinity_key = self.api_key.id if self.api_key else None
+            mapped_name = mapping.model.select_provider_model_name(
+                affinity_key, api_format=self.FORMAT_ID
+            )
             logger.debug(f"[CLI] 模型映射: {source_model} -> {mapped_name} (provider={provider_id[:8]}...)")
             return mapped_name
 
@@ -403,24 +355,12 @@ class CliMessageHandlerBase(BaseMessageHandler):
             return StreamingResponse(
                 monitored_stream,
                 media_type="text/event-stream",
+                headers=build_sse_headers(),
                 background=background_tasks,
             )
 
         except Exception as e:
-            # 对于已知的业务异常，只记录简洁的错误信息，不输出完整堆栈
-            from src.core.exceptions import (
-                ProviderException,
-                QuotaExceededException,
-                RateLimitException,
-                ModelNotSupportedException,
-            )
-
-            if isinstance(e, (ProviderException, QuotaExceededException, RateLimitException, ModelNotSupportedException)):
-                # 业务异常：简洁日志
-                logger.error(f"流式请求失败: [{type(e).__name__}] {e}")
-            else:
-                # 未知异常：完整堆栈
-                logger.exception(f"流式请求失败: {e}")
+            self._log_request_error("流式请求失败", e)
             await self._record_stream_failure(ctx, e, original_headers, original_request_body)
             raise
 
@@ -440,7 +380,7 @@ class CliMessageHandlerBase(BaseMessageHandler):
         ctx.chunk_count = 0
         ctx.data_count = 0
         ctx.has_completion = False
-        ctx.collected_text = ""
+        ctx._collected_text_parts = []  # 重置文本收集
         ctx.input_tokens = 0
         ctx.output_tokens = 0
         ctx.cached_tokens = 0
@@ -460,7 +400,7 @@ class CliMessageHandlerBase(BaseMessageHandler):
         ctx.provider_api_format = str(endpoint.api_format) if endpoint.api_format else ""
         ctx.client_api_format = ctx.api_format  # 已在 process_stream 中设置
 
-        # 获取模型映射（别名/映射 → 实际模型名）
+        # 获取模型映射（映射名称 → 实际模型名）
         mapped_model = await self._get_mapped_model(
             source_model=ctx.model,
             provider_id=str(provider.id),
@@ -514,7 +454,13 @@ class CliMessageHandlerBase(BaseMessageHandler):
             f"Key=***{key.api_key[-4:]}, "
             f"原始模型={ctx.model}, 映射后={mapped_model or '无映射'}, URL模型={url_model}")
 
-        http_client = httpx.AsyncClient(timeout=timeout_config, follow_redirects=True)
+        # 创建 HTTP 客户端（支持代理配置）
+        from src.clients.http_client import HTTPClientPool
+
+        http_client = HTTPClientPool.create_client_with_proxy(
+            proxy_config=endpoint.proxy,
+            timeout=timeout_config,
+        )
         try:
             response_ctx = http_client.stream(
                 "POST", url, json=provider_payload, headers=provider_headers
@@ -528,12 +474,12 @@ class CliMessageHandlerBase(BaseMessageHandler):
 
             stream_response.raise_for_status()
 
-            # 创建行迭代器（只创建一次，后续会继续使用）
-            line_iterator = stream_response.aiter_lines()
+            # 使用字节流迭代器（避免 aiter_lines 的性能问题）
+            byte_iterator = stream_response.aiter_raw()
 
             # 预读第一个数据块，检测嵌套错误（HTTP 200 但响应体包含错误）
-            prefetched_lines = await self._prefetch_and_check_embedded_error(
-                line_iterator, provider, endpoint, ctx
+            prefetched_chunks = await self._prefetch_and_check_embedded_error(
+                byte_iterator, provider, endpoint, ctx
             )
 
         except httpx.HTTPStatusError as e:
@@ -558,10 +504,10 @@ class CliMessageHandlerBase(BaseMessageHandler):
         # 创建流生成器（带预读数据，使用同一个迭代器）
         return self._create_response_stream_with_prefetch(
             ctx,
-            line_iterator,
+            byte_iterator,
             response_ctx,
             http_client,
-            prefetched_lines,
+            prefetched_chunks,
         )
 
     async def _create_response_stream(
@@ -571,58 +517,75 @@ class CliMessageHandlerBase(BaseMessageHandler):
         response_ctx: Any,
         http_client: httpx.AsyncClient,
     ) -> AsyncGenerator[bytes, None]:
-        """创建响应流生成器"""
+        """创建响应流生成器（使用字节流）"""
         try:
             sse_parser = SSEEventParser()
             last_data_time = time.time()
             streaming_status_updated = False
+            buffer = b""
+            # 使用增量解码器处理跨 chunk 的 UTF-8 字符
+            decoder = codecs.getincrementaldecoder("utf-8")(errors="replace")
 
             # 检查是否需要格式转换
             needs_conversion = self._needs_format_conversion(ctx)
 
-            async for line in stream_response.aiter_lines():
+            async for chunk in stream_response.aiter_raw():
                 # 在第一次输出数据前更新状态为 streaming
                 if not streaming_status_updated:
-                    self._update_usage_to_streaming(ctx.request_id)
+                    self._update_usage_to_streaming_with_ctx(ctx)
                     streaming_status_updated = True
 
-                normalized_line = line.rstrip("\r")
-                events = sse_parser.feed_line(normalized_line)
-
-                if normalized_line == "":
-                    for event in events:
-                        self._handle_sse_event(
-                            ctx,
-                            event.get("event"),
-                            event.get("data") or "",
+                buffer += chunk
+                # 处理缓冲区中的完整行
+                while b"\n" in buffer:
+                    line_bytes, buffer = buffer.split(b"\n", 1)
+                    try:
+                        # 使用增量解码器，可以正确处理跨 chunk 的多字节字符
+                        line = decoder.decode(line_bytes + b"\n", False).rstrip("\n")
+                    except Exception as e:
+                        logger.warning(
+                            f"[{self.request_id}] UTF-8 解码失败: {e}, "
+                            f"bytes={line_bytes[:50]!r}"
                         )
-                    yield b"\n"
-                    continue
+                        continue
 
-                ctx.chunk_count += 1
+                    normalized_line = line.rstrip("\r")
+                    events = sse_parser.feed_line(normalized_line)
 
-                # 空流检测：超过阈值且无数据，发送错误事件并结束
-                if ctx.chunk_count > self.EMPTY_CHUNK_THRESHOLD and ctx.data_count == 0:
-                    elapsed = time.time() - last_data_time
-                    if elapsed > self.DATA_TIMEOUT:
-                        logger.warning(f"提供商 '{ctx.provider_name}' 流超时且无数据")
-                        error_event = {
-                            "type": "error",
-                            "error": {
-                                "type": "empty_stream_timeout",
-                                "message": f"提供商 '{ctx.provider_name}' 流超时且未返回有效数据",
-                            },
-                        }
-                        yield f"event: error\ndata: {json.dumps(error_event)}\n\n".encode("utf-8")
-                        return  # 结束生成器
+                    if normalized_line == "":
+                        for event in events:
+                            self._handle_sse_event(
+                                ctx,
+                                event.get("event"),
+                                event.get("data") or "",
+                            )
+                        yield b"\n"
+                        continue
 
-                # 格式转换或直接透传
-                if needs_conversion:
-                    converted_line = self._convert_sse_line(ctx, line, events)
-                    if converted_line:
-                        yield (converted_line + "\n").encode("utf-8")
-                else:
-                    yield (line + "\n").encode("utf-8")
+                    ctx.chunk_count += 1
+
+                    # 空流检测：超过阈值且无数据，发送错误事件并结束
+                    if ctx.chunk_count > self.EMPTY_CHUNK_THRESHOLD and ctx.data_count == 0:
+                        elapsed = time.time() - last_data_time
+                        if elapsed > self.DATA_TIMEOUT:
+                            logger.warning(f"提供商 '{ctx.provider_name}' 流超时且无数据")
+                            error_event = {
+                                "type": "error",
+                                "error": {
+                                    "type": "empty_stream_timeout",
+                                    "message": f"提供商 '{ctx.provider_name}' 流超时且未返回有效数据",
+                                },
+                            }
+                            yield f"event: error\ndata: {json.dumps(error_event)}\n\n".encode("utf-8")
+                            return  # 结束生成器
+
+                    # 格式转换或直接透传
+                    if needs_conversion:
+                        converted_line = self._convert_sse_line(ctx, line, events)
+                        if converted_line:
+                            yield (converted_line + "\n").encode("utf-8")
+                    else:
+                        yield (line + "\n").encode("utf-8")
 
                 for event in events:
                     self._handle_sse_event(
@@ -696,7 +659,7 @@ class CliMessageHandlerBase(BaseMessageHandler):
 
     async def _prefetch_and_check_embedded_error(
         self,
-        line_iterator: Any,
+        byte_iterator: Any,
         provider: Provider,
         endpoint: ProviderEndpoint,
         ctx: StreamContext,
@@ -710,20 +673,25 @@ class CliMessageHandlerBase(BaseMessageHandler):
         同时检测 HTML 响应（通常是 base_url 配置错误导致返回网页）。
 
         Args:
-            line_iterator: 行迭代器（aiter_lines() 返回的迭代器）
+            byte_iterator: 字节流迭代器
             provider: Provider 对象
             endpoint: Endpoint 对象
             ctx: 流上下文
 
         Returns:
-            预读的行列表（需要在后续流中先输出）
+            预读的字节块列表（需要在后续流中先输出）
 
         Raises:
             EmbeddedErrorException: 如果检测到嵌套错误
             ProviderNotAvailableException: 如果检测到 HTML 响应（配置错误）
         """
-        prefetched_lines: list = []
+        prefetched_chunks: list = []
         max_prefetch_lines = 5  # 最多预读5行来检测错误
+        buffer = b""
+        line_count = 0
+        should_stop = False
+        # 使用增量解码器处理跨 chunk 的 UTF-8 字符
+        decoder = codecs.getincrementaldecoder("utf-8")(errors="replace")
 
         try:
             # 获取对应格式的解析器
@@ -736,69 +704,86 @@ class CliMessageHandlerBase(BaseMessageHandler):
             else:
                 provider_parser = self.parser
 
-            line_count = 0
-            async for line in line_iterator:
-                prefetched_lines.append(line)
-                line_count += 1
+            async for chunk in byte_iterator:
+                prefetched_chunks.append(chunk)
+                buffer += chunk
 
-                # 解析数据
-                normalized_line = line.rstrip("\r")
+                # 尝试按行解析缓冲区
+                while b"\n" in buffer:
+                    line_bytes, buffer = buffer.split(b"\n", 1)
+                    try:
+                        # 使用增量解码器，可以正确处理跨 chunk 的多字节字符
+                        line = decoder.decode(line_bytes + b"\n", False).rstrip("\n")
+                    except Exception as e:
+                        logger.warning(
+                            f"[{self.request_id}] 预读时 UTF-8 解码失败: {e}, "
+                            f"bytes={line_bytes[:50]!r}"
+                        )
+                        continue
 
-                # 检测 HTML 响应（base_url 配置错误的常见症状）
-                lower_line = normalized_line.lower()
-                if lower_line.startswith("<!doctype") or lower_line.startswith("<html"):
-                    logger.error(
-                        f"  [{self.request_id}] 检测到 HTML 响应，可能是 base_url 配置错误: "
-                        f"Provider={provider.name}, Endpoint={endpoint.id[:8]}..., "
-                        f"base_url={endpoint.base_url}"
-                    )
-                    raise ProviderNotAvailableException(
-                        f"提供商 '{provider.name}' 返回了 HTML 页面而非 API 响应，请检查 endpoint 的 base_url 配置是否正确"
-                    )
+                    line_count += 1
+                    normalized_line = line.rstrip("\r")
 
-                if not normalized_line or normalized_line.startswith(":"):
-                    # 空行或注释行，继续预读
-                    if line_count >= max_prefetch_lines:
+                    # 检测 HTML 响应（base_url 配置错误的常见症状）
+                    lower_line = normalized_line.lower()
+                    if lower_line.startswith("<!doctype") or lower_line.startswith("<html"):
+                        logger.error(
+                            f"  [{self.request_id}] 检测到 HTML 响应，可能是 base_url 配置错误: "
+                            f"Provider={provider.name}, Endpoint={endpoint.id[:8]}..., "
+                            f"base_url={endpoint.base_url}"
+                        )
+                        raise ProviderNotAvailableException(
+                            f"提供商 '{provider.name}' 返回了 HTML 页面而非 API 响应，请检查 endpoint 的 base_url 配置是否正确"
+                        )
+
+                    if not normalized_line or normalized_line.startswith(":"):
+                        # 空行或注释行，继续预读
+                        if line_count >= max_prefetch_lines:
+                            break
+                        continue
+
+                    # 尝试解析 SSE 数据
+                    data_str = normalized_line
+                    if normalized_line.startswith("data: "):
+                        data_str = normalized_line[6:]
+
+                    if data_str == "[DONE]":
+                        should_stop = True
                         break
-                    continue
 
-                # 尝试解析 SSE 数据
-                data_str = normalized_line
-                if normalized_line.startswith("data: "):
-                    data_str = normalized_line[6:]
+                    try:
+                        data = json.loads(data_str)
+                    except json.JSONDecodeError:
+                        # 不是有效 JSON，可能是部分数据，继续
+                        if line_count >= max_prefetch_lines:
+                            break
+                        continue
 
-                if data_str == "[DONE]":
+                    # 使用解析器检查是否为错误响应
+                    if isinstance(data, dict) and provider_parser.is_error_response(data):
+                        # 提取错误信息
+                        parsed = provider_parser.parse_response(data, 200)
+                        logger.warning(f"  [{self.request_id}] 检测到嵌套错误: "
+                            f"Provider={provider.name}, "
+                            f"error_type={parsed.error_type}, "
+                            f"message={parsed.error_message}")
+                        raise EmbeddedErrorException(
+                            provider_name=str(provider.name),
+                            error_code=(
+                                int(parsed.error_type)
+                                if parsed.error_type and parsed.error_type.isdigit()
+                                else None
+                            ),
+                            error_message=parsed.error_message,
+                            error_status=parsed.error_type,
+                        )
+
+                    # 预读到有效数据，没有错误，停止预读
+                    should_stop = True
                     break
 
-                try:
-                    data = json.loads(data_str)
-                except json.JSONDecodeError:
-                    # 不是有效 JSON，可能是部分数据，继续
-                    if line_count >= max_prefetch_lines:
-                        break
-                    continue
-
-                # 使用解析器检查是否为错误响应
-                if isinstance(data, dict) and provider_parser.is_error_response(data):
-                    # 提取错误信息
-                    parsed = provider_parser.parse_response(data, 200)
-                    logger.warning(f"  [{self.request_id}] 检测到嵌套错误: "
-                        f"Provider={provider.name}, "
-                        f"error_type={parsed.error_type}, "
-                        f"message={parsed.error_message}")
-                    raise EmbeddedErrorException(
-                        provider_name=str(provider.name),
-                        error_code=(
-                            int(parsed.error_type)
-                            if parsed.error_type and parsed.error_type.isdigit()
-                            else None
-                        ),
-                        error_message=parsed.error_message,
-                        error_status=parsed.error_type,
-                    )
-
-                # 预读到有效数据，没有错误，停止预读
-                break
+                if should_stop or line_count >= max_prefetch_lines:
+                    break
 
         except EmbeddedErrorException:
             # 重新抛出嵌套错误
@@ -807,112 +792,168 @@ class CliMessageHandlerBase(BaseMessageHandler):
             # 其他异常（如网络错误）在预读阶段发生，记录日志但不中断
             logger.debug(f"  [{self.request_id}] 预读流时发生异常: {e}")
 
-        return prefetched_lines
+        return prefetched_chunks
 
     async def _create_response_stream_with_prefetch(
         self,
         ctx: StreamContext,
-        line_iterator: Any,
+        byte_iterator: Any,
         response_ctx: Any,
         http_client: httpx.AsyncClient,
-        prefetched_lines: list,
+        prefetched_chunks: list,
     ) -> AsyncGenerator[bytes, None]:
-        """创建响应流生成器（带预读数据）"""
+        """创建响应流生成器（带预读数据，使用字节流）"""
         try:
             sse_parser = SSEEventParser()
             last_data_time = time.time()
+            buffer = b""
+            first_yield = True  # 标记是否是第一次 yield
+            # 使用增量解码器处理跨 chunk 的 UTF-8 字符
+            decoder = codecs.getincrementaldecoder("utf-8")(errors="replace")
 
             # 检查是否需要格式转换
             needs_conversion = self._needs_format_conversion(ctx)
 
             # 在第一次输出数据前更新状态为 streaming
-            if prefetched_lines:
-                self._update_usage_to_streaming(ctx.request_id)
+            if prefetched_chunks:
+                self._update_usage_to_streaming_with_ctx(ctx)
 
-            # 先处理预读的数据
-            for line in prefetched_lines:
-                normalized_line = line.rstrip("\r")
-                events = sse_parser.feed_line(normalized_line)
+            # 先处理预读的字节块
+            for chunk in prefetched_chunks:
+                buffer += chunk
+                # 处理缓冲区中的完整行
+                while b"\n" in buffer:
+                    line_bytes, buffer = buffer.split(b"\n", 1)
+                    try:
+                        # 使用增量解码器，可以正确处理跨 chunk 的多字节字符
+                        line = decoder.decode(line_bytes + b"\n", False).rstrip("\n")
+                    except Exception as e:
+                        logger.warning(
+                            f"[{self.request_id}] UTF-8 解码失败: {e}, "
+                            f"bytes={line_bytes[:50]!r}"
+                        )
+                        continue
+
+                    normalized_line = line.rstrip("\r")
+                    events = sse_parser.feed_line(normalized_line)
+
+                    if normalized_line == "":
+                        for event in events:
+                            self._handle_sse_event(
+                                ctx,
+                                event.get("event"),
+                                event.get("data") or "",
+                            )
+                        # 记录首字时间 (第一次 yield)
+                        if first_yield:
+                            ctx.record_first_byte_time(self.start_time)
+                            first_yield = False
+                        yield b"\n"
+                        continue
+
+                    ctx.chunk_count += 1
+
+                    # 格式转换或直接透传
+                    if needs_conversion:
+                        converted_line = self._convert_sse_line(ctx, line, events)
+                        if converted_line:
+                            # 记录首字时间 (第一次 yield)
+                            if first_yield:
+                                ctx.record_first_byte_time(self.start_time)
+                                first_yield = False
+                            yield (converted_line + "\n").encode("utf-8")
+                    else:
+                        # 记录首字时间 (第一次 yield)
+                        if first_yield:
+                            ctx.record_first_byte_time(self.start_time)
+                            first_yield = False
+                        yield (line + "\n").encode("utf-8")
 
-                if normalized_line == "":
                     for event in events:
                         self._handle_sse_event(
                             ctx,
                             event.get("event"),
                             event.get("data") or "",
                         )
-                    yield b"\n"
-                    continue
 
-                ctx.chunk_count += 1
-
-                # 格式转换或直接透传
-                if needs_conversion:
-                    converted_line = self._convert_sse_line(ctx, line, events)
-                    if converted_line:
-                        yield (converted_line + "\n").encode("utf-8")
-                else:
-                    yield (line + "\n").encode("utf-8")
-
-                for event in events:
-                    self._handle_sse_event(
-                        ctx,
-                        event.get("event"),
-                        event.get("data") or "",
-                    )
-
-                if ctx.data_count > 0:
-                    last_data_time = time.time()
+                    if ctx.data_count > 0:
+                        last_data_time = time.time()
 
             # 继续处理剩余的流数据（使用同一个迭代器）
-            async for line in line_iterator:
-                normalized_line = line.rstrip("\r")
-                events = sse_parser.feed_line(normalized_line)
+            async for chunk in byte_iterator:
+                buffer += chunk
+                # 处理缓冲区中的完整行
+                while b"\n" in buffer:
+                    line_bytes, buffer = buffer.split(b"\n", 1)
+                    try:
+                        # 使用增量解码器，可以正确处理跨 chunk 的多字节字符
+                        line = decoder.decode(line_bytes + b"\n", False).rstrip("\n")
+                    except Exception as e:
+                        logger.warning(
+                            f"[{self.request_id}] UTF-8 解码失败: {e}, "
+                            f"bytes={line_bytes[:50]!r}"
+                        )
+                        continue
+
+                    normalized_line = line.rstrip("\r")
+                    events = sse_parser.feed_line(normalized_line)
+
+                    if normalized_line == "":
+                        for event in events:
+                            self._handle_sse_event(
+                                ctx,
+                                event.get("event"),
+                                event.get("data") or "",
+                            )
+                        # 记录首字时间 (第一次 yield) - 如果预读数据为空
+                        if first_yield:
+                            ctx.record_first_byte_time(self.start_time)
+                            first_yield = False
+                        yield b"\n"
+                        continue
+
+                    ctx.chunk_count += 1
+
+                    # 空流检测：超过阈值且无数据，发送错误事件并结束
+                    if ctx.chunk_count > self.EMPTY_CHUNK_THRESHOLD and ctx.data_count == 0:
+                        elapsed = time.time() - last_data_time
+                        if elapsed > self.DATA_TIMEOUT:
+                            logger.warning(f"提供商 '{ctx.provider_name}' 流超时且无数据")
+                            error_event = {
+                                "type": "error",
+                                "error": {
+                                    "type": "empty_stream_timeout",
+                                    "message": f"提供商 '{ctx.provider_name}' 流超时且未返回有效数据",
+                                },
+                            }
+                            yield f"event: error\ndata: {json.dumps(error_event)}\n\n".encode("utf-8")
+                            return
+
+                    # 格式转换或直接透传
+                    if needs_conversion:
+                        converted_line = self._convert_sse_line(ctx, line, events)
+                        if converted_line:
+                            # 记录首字时间 (第一次 yield) - 如果预读数据为空
+                            if first_yield:
+                                ctx.record_first_byte_time(self.start_time)
+                                first_yield = False
+                            yield (converted_line + "\n").encode("utf-8")
+                    else:
+                        # 记录首字时间 (第一次 yield) - 如果预读数据为空
+                        if first_yield:
+                            ctx.record_first_byte_time(self.start_time)
+                            first_yield = False
+                        yield (line + "\n").encode("utf-8")
 
-                if normalized_line == "":
                     for event in events:
                         self._handle_sse_event(
                             ctx,
                             event.get("event"),
                             event.get("data") or "",
                         )
-                    yield b"\n"
-                    continue
 
-                ctx.chunk_count += 1
-
-                # 空流检测：超过阈值且无数据，发送错误事件并结束
-                if ctx.chunk_count > self.EMPTY_CHUNK_THRESHOLD and ctx.data_count == 0:
-                    elapsed = time.time() - last_data_time
-                    if elapsed > self.DATA_TIMEOUT:
-                        logger.warning(f"提供商 '{ctx.provider_name}' 流超时且无数据")
-                        error_event = {
-                            "type": "error",
-                            "error": {
-                                "type": "empty_stream_timeout",
-                                "message": f"提供商 '{ctx.provider_name}' 流超时且未返回有效数据",
-                            },
-                        }
-                        yield f"event: error\ndata: {json.dumps(error_event)}\n\n".encode("utf-8")
-                        return
-
-                # 格式转换或直接透传
-                if needs_conversion:
-                    converted_line = self._convert_sse_line(ctx, line, events)
-                    if converted_line:
-                        yield (converted_line + "\n").encode("utf-8")
-                else:
-                    yield (line + "\n").encode("utf-8")
-
-                for event in events:
-                    self._handle_sse_event(
-                        ctx,
-                        event.get("event"),
-                        event.get("data") or "",
-                    )
-
-                if ctx.data_count > 0:
-                    last_data_time = time.time()
+                    if ctx.data_count > 0:
+                        last_data_time = time.time()
 
             # 处理剩余事件
             flushed_events = sse_parser.flush()
@@ -1041,7 +1082,7 @@ class CliMessageHandlerBase(BaseMessageHandler):
         # 提取文本内容
         text = self.parser.extract_text_content(data)
         if text:
-            ctx.collected_text += text
+            ctx.append_text(text)
 
         # 检查完成事件
         if event_type in ("response.completed", "message_stop"):
@@ -1073,8 +1114,10 @@ class CliMessageHandlerBase(BaseMessageHandler):
             async for chunk in stream_generator:
                 yield chunk
         except asyncio.CancelledError:
-            ctx.status_code = 499
-            ctx.error_message = "Client disconnected"
+            # 如果响应已完成，不标记为失败
+            if not ctx.has_completion:
+                ctx.status_code = 499
+                ctx.error_message = "Client disconnected"
             raise
         except httpx.TimeoutException as e:
             ctx.status_code = 504
@@ -1093,9 +1136,11 @@ class CliMessageHandlerBase(BaseMessageHandler):
     ) -> None:
         """在流完成后记录统计信息"""
         try:
-            await asyncio.sleep(0.1)
+            # 使用 self.start_time 作为时间基准，与首字时间保持一致
+            # 注意：不要把统计延迟算进响应时间里
+            response_time_ms = int((time.time() - self.start_time) * 1000)
 
-            response_time_ms = int((time.time() - ctx.start_time) * 1000)
+            await asyncio.sleep(0.1)
 
             if not ctx.provider_name:
                 logger.warning(f"[{ctx.request_id}] 流式请求失败，未选中提供商")
@@ -1175,6 +1220,7 @@ class CliMessageHandlerBase(BaseMessageHandler):
                         input_tokens=actual_input_tokens,
                         output_tokens=ctx.output_tokens,
                         response_time_ms=response_time_ms,
+                        first_byte_time_ms=ctx.first_byte_time_ms,  # 传递首字时间
                         status_code=ctx.status_code,
                         request_headers=original_headers,
                         request_body=actual_request_body,
@@ -1195,9 +1241,18 @@ class CliMessageHandlerBase(BaseMessageHandler):
                         response_metadata=ctx.response_metadata if ctx.response_metadata else None,
                     )
                     logger.debug(f"{self.FORMAT_ID} 流式响应完成")
-                    # 简洁的请求完成摘要
-                    logger.info(f"[OK] {self.request_id[:8]} | {ctx.model} | {ctx.provider_name} | {response_time_ms}ms | "
-                        f"in:{ctx.input_tokens or 0} out:{ctx.output_tokens or 0}")
+                    # 简洁的请求完成摘要（两行格式）
+                    line1 = (
+                        f"[OK] {self.request_id[:8]} | {ctx.model} | {ctx.provider_name}"
+                    )
+                    if ctx.first_byte_time_ms:
+                        line1 += f" | TTFB: {ctx.first_byte_time_ms}ms"
+
+                    line2 = (
+                        f"      Total: {response_time_ms}ms | "
+                        f"in:{ctx.input_tokens or 0} out:{ctx.output_tokens or 0}"
+                    )
+                    logger.info(f"{line1}\n{line2}")
 
                 # 更新候选记录的最终状态和延迟时间
                 # 注意：RequestExecutor 会在流开始时过早地标记成功（只记录了连接建立的时间）
@@ -1249,7 +1304,8 @@ class CliMessageHandlerBase(BaseMessageHandler):
         original_request_body: Dict[str, Any],
     ) -> None:
         """记录流式请求失败"""
-        response_time_ms = int((time.time() - ctx.start_time) * 1000)
+        # 使用 self.start_time 作为时间基准，与首字时间保持一致
+        response_time_ms = int((time.time() - self.start_time) * 1000)
 
         status_code = 503
         if isinstance(error, ProviderAuthException):
@@ -1326,7 +1382,7 @@ class CliMessageHandlerBase(BaseMessageHandler):
             provider_name = str(provider.name)
             provider_api_format = str(endpoint.api_format) if endpoint.api_format else ""
 
-            # 获取模型映射（别名/映射 → 实际模型名）
+            # 获取模型映射（映射名称 → 实际模型名）
             mapped_model = await self._get_mapped_model(
                 source_model=model,
                 provider_id=str(provider.id),
@@ -1371,10 +1427,14 @@ class CliMessageHandlerBase(BaseMessageHandler):
                 f"Key=***{key.api_key[-4:]}, "
                 f"原始模型={model}, 映射后={mapped_model or '无映射'}, URL模型={url_model}")
 
-            async with httpx.AsyncClient(
-                timeout=float(endpoint.timeout),
-                follow_redirects=True,
-            ) as http_client:
+            # 创建 HTTP 客户端（支持代理配置）
+            from src.clients.http_client import HTTPClientPool
+
+            http_client = HTTPClientPool.create_client_with_proxy(
+                proxy_config=endpoint.proxy,
+                timeout=httpx.Timeout(float(endpoint.timeout)),
+            )
+            async with http_client:
                 resp = await http_client.post(url, json=provider_payload, headers=provider_headers)
 
                 status_code = resp.status_code

src/api/handlers/base/content_extractors.py

@@ -0,0 +1,274 @@
+"""
+流式内容提取器 - 策略模式实现
+
+为不同 API 格式（OpenAI、Claude、Gemini）提供内容提取和 chunk 构造的抽象。
+StreamSmoother 使用这些提取器来处理不同格式的 SSE 事件。
+"""
+
+import copy
+import json
+from abc import ABC, abstractmethod
+from typing import Optional
+
+
+class ContentExtractor(ABC):
+    """
+    流式内容提取器抽象基类
+
+    定义从 SSE 事件中提取文本内容和构造新 chunk 的接口。
+    每种 API 格式（OpenAI、Claude、Gemini）需要实现自己的提取器。
+    """
+
+    @abstractmethod
+    def extract_content(self, data: dict) -> Optional[str]:
+        """
+        从 SSE 数据中提取可拆分的文本内容
+
+        Args:
+            data: 解析后的 JSON 数据
+
+        Returns:
+            提取的文本内容，如果无法提取则返回 None
+        """
+        pass
+
+    @abstractmethod
+    def create_chunk(
+        self,
+        original_data: dict,
+        new_content: str,
+        event_type: str = "",
+        is_first: bool = False,
+    ) -> bytes:
+        """
+        使用新内容构造 SSE chunk
+
+        Args:
+            original_data: 原始 JSON 数据
+            new_content: 新的文本内容
+            event_type: SSE 事件类型（某些格式需要）
+            is_first: 是否是第一个 chunk（用于保留 role 等字段）
+
+        Returns:
+            编码后的 SSE 字节数据
+        """
+        pass
+
+
+class OpenAIContentExtractor(ContentExtractor):
+    """
+    OpenAI 格式内容提取器
+
+    处理 OpenAI Chat Completions API 的流式响应格式：
+    - 数据结构: choices[0].delta.content
+    - 只在 delta 仅包含 role/content 时允许拆分，避免破坏 tool_calls 等结构
+    """
+
+    def extract_content(self, data: dict) -> Optional[str]:
+        if not isinstance(data, dict):
+            return None
+
+        choices = data.get("choices")
+        if not isinstance(choices, list) or len(choices) != 1:
+            return None
+
+        first_choice = choices[0]
+        if not isinstance(first_choice, dict):
+            return None
+
+        delta = first_choice.get("delta")
+        if not isinstance(delta, dict):
+            return None
+
+        content = delta.get("content")
+        if not isinstance(content, str):
+            return None
+
+        # 只有 delta 仅包含 role/content 时才允许拆分
+        # 避免破坏 tool_calls、function_call 等复杂结构
+        allowed_keys = {"role", "content"}
+        if not all(key in allowed_keys for key in delta.keys()):
+            return None
+
+        return content
+
+    def create_chunk(
+        self,
+        original_data: dict,
+        new_content: str,
+        event_type: str = "",
+        is_first: bool = False,
+    ) -> bytes:
+        new_data = original_data.copy()
+
+        if "choices" in new_data and new_data["choices"]:
+            new_choices = []
+            for choice in new_data["choices"]:
+                new_choice = choice.copy()
+                if "delta" in new_choice:
+                    new_delta = {}
+                    # 只有第一个 chunk 保留 role
+                    if is_first and "role" in new_choice["delta"]:
+                        new_delta["role"] = new_choice["delta"]["role"]
+                    new_delta["content"] = new_content
+                    new_choice["delta"] = new_delta
+                new_choices.append(new_choice)
+            new_data["choices"] = new_choices
+
+        return f"data: {json.dumps(new_data, ensure_ascii=False)}\n\n".encode("utf-8")
+
+
+class ClaudeContentExtractor(ContentExtractor):
+    """
+    Claude 格式内容提取器
+
+    处理 Claude Messages API 的流式响应格式：
+    - 事件类型: content_block_delta
+    - 数据结构: delta.type=text_delta, delta.text
+    """
+
+    def extract_content(self, data: dict) -> Optional[str]:
+        if not isinstance(data, dict):
+            return None
+
+        # 检查事件类型
+        if data.get("type") != "content_block_delta":
+            return None
+
+        delta = data.get("delta", {})
+        if not isinstance(delta, dict):
+            return None
+
+        # 检查 delta 类型
+        if delta.get("type") != "text_delta":
+            return None
+
+        text = delta.get("text")
+        if not isinstance(text, str):
+            return None
+
+        return text
+
+    def create_chunk(
+        self,
+        original_data: dict,
+        new_content: str,
+        event_type: str = "",
+        is_first: bool = False,
+    ) -> bytes:
+        new_data = original_data.copy()
+
+        if "delta" in new_data:
+            new_delta = new_data["delta"].copy()
+            new_delta["text"] = new_content
+            new_data["delta"] = new_delta
+
+        # Claude 格式需要 event: 前缀
+        event_name = event_type or "content_block_delta"
+        return f"event: {event_name}\ndata: {json.dumps(new_data, ensure_ascii=False)}\n\n".encode(
+            "utf-8"
+        )
+
+
+class GeminiContentExtractor(ContentExtractor):
+    """
+    Gemini 格式内容提取器
+
+    处理 Gemini API 的流式响应格式：
+    - 数据结构: candidates[0].content.parts[0].text
+    - 只有纯文本块才拆分
+    """
+
+    def extract_content(self, data: dict) -> Optional[str]:
+        if not isinstance(data, dict):
+            return None
+
+        candidates = data.get("candidates")
+        if not isinstance(candidates, list) or len(candidates) != 1:
+            return None
+
+        first_candidate = candidates[0]
+        if not isinstance(first_candidate, dict):
+            return None
+
+        content = first_candidate.get("content", {})
+        if not isinstance(content, dict):
+            return None
+
+        parts = content.get("parts", [])
+        if not isinstance(parts, list) or len(parts) != 1:
+            return None
+
+        first_part = parts[0]
+        if not isinstance(first_part, dict):
+            return None
+
+        text = first_part.get("text")
+        # 只有纯文本块（只有 text 字段）才拆分
+        if not isinstance(text, str) or len(first_part) != 1:
+            return None
+
+        return text
+
+    def create_chunk(
+        self,
+        original_data: dict,
+        new_content: str,
+        event_type: str = "",
+        is_first: bool = False,
+    ) -> bytes:
+        new_data = copy.deepcopy(original_data)
+
+        if "candidates" in new_data and new_data["candidates"]:
+            first_candidate = new_data["candidates"][0]
+            if "content" in first_candidate:
+                content = first_candidate["content"]
+                if "parts" in content and content["parts"]:
+                    content["parts"][0]["text"] = new_content
+
+        return f"data: {json.dumps(new_data, ensure_ascii=False)}\n\n".encode("utf-8")
+
+
+# 提取器注册表
+_EXTRACTORS: dict[str, type[ContentExtractor]] = {
+    "openai": OpenAIContentExtractor,
+    "claude": ClaudeContentExtractor,
+    "gemini": GeminiContentExtractor,
+}
+
+
+def get_extractor(format_name: str) -> Optional[ContentExtractor]:
+    """
+    根据格式名获取对应的内容提取器实例
+
+    Args:
+        format_name: 格式名称（openai, claude, gemini）
+
+    Returns:
+        对应的提取器实例，如果格式不支持则返回 None
+    """
+    extractor_class = _EXTRACTORS.get(format_name.lower())
+    if extractor_class:
+        return extractor_class()
+    return None
+
+
+def register_extractor(format_name: str, extractor_class: type[ContentExtractor]) -> None:
+    """
+    注册新的内容提取器
+
+    Args:
+        format_name: 格式名称
+        extractor_class: 提取器类
+    """
+    _EXTRACTORS[format_name.lower()] = extractor_class
+
+
+def get_extractor_formats() -> list[str]:
+    """
+    获取所有已注册的格式名称列表
+
+    Returns:
+        格式名称列表
+    """
+    return list(_EXTRACTORS.keys())

src/api/handlers/base/parsers.py

@@ -5,7 +5,7 @@
 不再经过 Protocol 抽象层。
 """
 
-from typing import Any, Dict, Optional, Tuple
+from typing import Any, Dict, Optional, Tuple, Type
 
 from src.api.handlers.base.response_parser import (
     ParsedChunk,
@@ -13,6 +13,7 @@ from src.api.handlers.base.response_parser import (
     ResponseParser,
     StreamStats,
 )
+from src.api.handlers.base.utils import extract_cache_creation_tokens
 
 
 def _check_nested_error(response: Dict[str, Any]) -> Tuple[bool, Optional[Dict[str, Any]]]:
@@ -60,7 +61,7 @@ def _check_nested_error(response: Dict[str, Any]) -> Tuple[bool, Optional[Dict[s
 class OpenAIResponseParser(ResponseParser):
     """OpenAI 格式响应解析器"""
 
-    def __init__(self):
+    def __init__(self) -> None:
         from src.api.handlers.openai.stream_parser import OpenAIStreamParser
 
         self._parser = OpenAIStreamParser()
@@ -146,7 +147,7 @@ class OpenAIResponseParser(ResponseParser):
         if choices:
             message = choices[0].get("message", {})
             content = message.get("content")
-            if content:
+            if isinstance(content, str):
                 return content
         return ""
 
@@ -158,7 +159,7 @@ class OpenAIResponseParser(ResponseParser):
 class OpenAICliResponseParser(OpenAIResponseParser):
     """OpenAI CLI 格式响应解析器"""
 
-    def __init__(self):
+    def __init__(self) -> None:
         super().__init__()
         self.name = "OPENAI_CLI"
         self.api_format = "OPENAI_CLI"
@@ -167,7 +168,7 @@ class OpenAICliResponseParser(OpenAIResponseParser):
 class ClaudeResponseParser(ResponseParser):
     """Claude 格式响应解析器"""
 
-    def __init__(self):
+    def __init__(self) -> None:
         from src.api.handlers.claude.stream_parser import ClaudeStreamParser
 
         self._parser = ClaudeStreamParser()
@@ -252,7 +253,7 @@ class ClaudeResponseParser(ResponseParser):
         usage = response.get("usage", {})
         result.input_tokens = usage.get("input_tokens", 0)
         result.output_tokens = usage.get("output_tokens", 0)
-        result.cache_creation_tokens = usage.get("cache_creation_input_tokens", 0)
+        result.cache_creation_tokens = extract_cache_creation_tokens(usage)
         result.cache_read_tokens = usage.get("cache_read_input_tokens", 0)
 
         # 检查错误（支持嵌套错误格式）
@@ -265,11 +266,16 @@ class ClaudeResponseParser(ResponseParser):
         return result
 
     def extract_usage_from_response(self, response: Dict[str, Any]) -> Dict[str, int]:
+        # 对于 message_start 事件，usage 在 message.usage 路径下
+        # 对于其他响应，usage 在顶层
         usage = response.get("usage", {})
+        if not usage and "message" in response:
+            usage = response.get("message", {}).get("usage", {})
+
         return {
             "input_tokens": usage.get("input_tokens", 0),
             "output_tokens": usage.get("output_tokens", 0),
-            "cache_creation_tokens": usage.get("cache_creation_input_tokens", 0),
+            "cache_creation_tokens": extract_cache_creation_tokens(usage),
             "cache_read_tokens": usage.get("cache_read_input_tokens", 0),
         }
 
@@ -291,7 +297,7 @@ class ClaudeResponseParser(ResponseParser):
 class ClaudeCliResponseParser(ClaudeResponseParser):
     """Claude CLI 格式响应解析器"""
 
-    def __init__(self):
+    def __init__(self) -> None:
         super().__init__()
         self.name = "CLAUDE_CLI"
         self.api_format = "CLAUDE_CLI"
@@ -300,7 +306,7 @@ class ClaudeCliResponseParser(ClaudeResponseParser):
 class GeminiResponseParser(ResponseParser):
     """Gemini 格式响应解析器"""
 
-    def __init__(self):
+    def __init__(self) -> None:
         from src.api.handlers.gemini.stream_parser import GeminiStreamParser
 
         self._parser = GeminiStreamParser()
@@ -443,20 +449,20 @@ class GeminiResponseParser(ResponseParser):
 
         使用增强的错误检测逻辑，支持嵌套在 chunks 中的错误
         """
-        return self._parser.is_error_event(response)
+        return bool(self._parser.is_error_event(response))
 
 
 class GeminiCliResponseParser(GeminiResponseParser):
     """Gemini CLI 格式响应解析器"""
 
-    def __init__(self):
+    def __init__(self) -> None:
         super().__init__()
         self.name = "GEMINI_CLI"
         self.api_format = "GEMINI_CLI"
 
 
 # 解析器注册表
-_PARSERS = {
+_PARSERS: Dict[str, Type[ResponseParser]] = {
     "CLAUDE": ClaudeResponseParser,
     "CLAUDE_CLI": ClaudeCliResponseParser,
     "OPENAI": OpenAIResponseParser,
@@ -498,6 +504,5 @@ __all__ = [
     "GeminiResponseParser",
     "GeminiCliResponseParser",
     "get_parser_for_format",
-    "get_parser_from_protocol",
     "is_cli_format",
 ]

src/api/handlers/base/stream_context.py

@@ -8,6 +8,7 @@
 - 请求/响应数据
 """
 
+import time
 from dataclasses import dataclass, field
 from typing import Any, Dict, List, Optional
 
@@ -25,12 +26,18 @@ class StreamContext:
     model: str
     api_format: str
 
+    # 请求标识信息（CLI handler 需要）
+    request_id: str = ""
+    user_id: int = 0
+    api_key_id: int = 0
+
     # Provider 信息（在请求执行时填充）
     provider_name: Optional[str] = None
     provider_id: Optional[str] = None
     endpoint_id: Optional[str] = None
     key_id: Optional[str] = None
     attempt_id: Optional[str] = None
+    attempt_synced: bool = False
     provider_api_format: Optional[str] = None  # Provider 的响应格式
 
     # 模型映射
@@ -43,7 +50,14 @@ class StreamContext:
     cache_creation_tokens: int = 0
 
     # 响应内容
-    collected_text: str = ""
+    _collected_text_parts: List[str] = field(default_factory=list, repr=False)
+    response_id: Optional[str] = None
+    final_usage: Optional[Dict[str, Any]] = None
+    final_response: Optional[Dict[str, Any]] = None
+
+    # 时间指标
+    first_byte_time_ms: Optional[int] = None  # 首字时间 (TTFB - Time To First Byte)
+    start_time: float = field(default_factory=time.time)
 
     # 响应状态
     status_code: int = 200
@@ -55,6 +69,12 @@ class StreamContext:
     provider_request_headers: Dict[str, str] = field(default_factory=dict)
     provider_request_body: Optional[Dict[str, Any]] = None
 
+    # 格式转换信息（CLI handler 需要）
+    client_api_format: str = ""
+
+    # Provider 响应元数据（CLI handler 需要）
+    response_metadata: Dict[str, Any] = field(default_factory=dict)
+
     # 流式处理统计
     data_count: int = 0
     chunk_count: int = 0
@@ -71,16 +91,30 @@ class StreamContext:
         self.chunk_count = 0
         self.data_count = 0
         self.has_completion = False
-        self.collected_text = ""
+        self._collected_text_parts = []
         self.input_tokens = 0
         self.output_tokens = 0
         self.cached_tokens = 0
         self.cache_creation_tokens = 0
         self.error_message = None
         self.status_code = 200
+        self.first_byte_time_ms = None
         self.response_headers = {}
         self.provider_request_headers = {}
         self.provider_request_body = None
+        self.response_id = None
+        self.final_usage = None
+        self.final_response = None
+
+    @property
+    def collected_text(self) -> str:
+        """已收集的文本内容（按需拼接，避免在流式过程中频繁做字符串拷贝）"""
+        return "".join(self._collected_text_parts)
+
+    def append_text(self, text: str) -> None:
+        """追加文本内容（仅在需要收集文本时调用）"""
+        if text:
+            self._collected_text_parts.append(text)
 
     def update_provider_info(
         self,
@@ -104,14 +138,40 @@ class StreamContext:
         cached_tokens: Optional[int] = None,
         cache_creation_tokens: Optional[int] = None,
     ) -> None:
-        """更新 Token 使用统计"""
-        if input_tokens is not None:
+        """
+        更新 Token 使用统计
+
+        采用防御性更新策略：只有当新值 > 0 或当前值为 0 时才更新，避免用 0 覆盖已有的正确值。
+
+        设计原理：
+        - 在流式响应中，某些事件可能不包含完整的 usage 信息（字段为 0 或不存在）
+        - 后续事件可能会提供完整的统计数据
+        - 通过这种策略，确保一旦获得非零值就保留它，不会被后续的 0 值覆盖
+
+        示例场景：
+        - message_start 事件：input_tokens=100, output_tokens=0
+        - message_delta 事件：input_tokens=0, output_tokens=50
+        - 最终结果：input_tokens=100, output_tokens=50
+
+        注意事项：
+        - 此策略假设初始值为 0 是正确的默认状态
+        - 如果需要将已有值重置为 0，请直接修改实例属性（不使用此方法）
+
+        Args:
+            input_tokens: 输入 tokens 数量
+            output_tokens: 输出 tokens 数量
+            cached_tokens: 缓存命中 tokens 数量
+            cache_creation_tokens: 缓存创建 tokens 数量
+        """
+        if input_tokens is not None and (input_tokens > 0 or self.input_tokens == 0):
             self.input_tokens = input_tokens
-        if output_tokens is not None:
+        if output_tokens is not None and (output_tokens > 0 or self.output_tokens == 0):
             self.output_tokens = output_tokens
-        if cached_tokens is not None:
+        if cached_tokens is not None and (cached_tokens > 0 or self.cached_tokens == 0):
             self.cached_tokens = cached_tokens
-        if cache_creation_tokens is not None:
+        if cache_creation_tokens is not None and (
+            cache_creation_tokens > 0 or self.cache_creation_tokens == 0
+        ):
             self.cache_creation_tokens = cache_creation_tokens
 
     def mark_failed(self, status_code: int, error_message: str) -> None:
@@ -119,6 +179,19 @@ class StreamContext:
         self.status_code = status_code
         self.error_message = error_message
 
+    def record_first_byte_time(self, start_time: float) -> None:
+        """
+        记录首字时间 (TTFB - Time To First Byte)
+
+        应在第一次向客户端发送数据时调用。
+        如果已记录过,则不会覆盖(避免重试时重复记录)。
+
+        Args:
+            start_time: 请求开始时间 (time.time())
+        """
+        if self.first_byte_time_ms is None:
+            self.first_byte_time_ms = int((time.time() - start_time) * 1000)
+
     def is_success(self) -> bool:
         """检查请求是否成功"""
         return self.status_code < 400
@@ -145,10 +218,22 @@ class StreamContext:
         获取日志摘要
 
         用于请求完成/失败时的日志输出。
+        包含首字时间 (TTFB) 和总响应时间,分两行显示。
         """
         status = "OK" if self.is_success() else "FAIL"
-        return (
+
+        # 第一行:基本信息 + 首字时间
+        line1 = (
             f"[{status}] {request_id[:8]} | {self.model} | "
-            f"{self.provider_name or 'unknown'} | {response_time_ms}ms | "
+            f"{self.provider_name or 'unknown'}"
+        )
+        if self.first_byte_time_ms is not None:
+            line1 += f" | TTFB: {self.first_byte_time_ms}ms"
+
+        # 第二行:总响应时间 + tokens
+        line2 = (
+            f"      Total: {response_time_ms}ms | "
             f"in:{self.input_tokens} out:{self.output_tokens}"
         )
+
+        return f"{line1}\n{line2}"

src/api/handlers/base/stream_processor.py

@@ -6,14 +6,22 @@
 2. 响应流生成
 3. 预读和嵌套错误检测
 4. 客户端断开检测
+5. 流式平滑输出
 """
 
 import asyncio
+import codecs
 import json
+from dataclasses import dataclass
 from typing import Any, AsyncGenerator, Callable, Optional
 
 import httpx
 
+from src.api.handlers.base.content_extractors import (
+    ContentExtractor,
+    get_extractor,
+    get_extractor_formats,
+)
 from src.api.handlers.base.parsers import get_parser_for_format
 from src.api.handlers.base.response_parser import ResponseParser
 from src.api.handlers.base.stream_context import StreamContext
@@ -23,11 +31,20 @@ from src.models.database import Provider, ProviderEndpoint
 from src.utils.sse_parser import SSEEventParser
 
 
+@dataclass
+class StreamSmoothingConfig:
+    """流式平滑输出配置"""
+
+    enabled: bool = False
+    chunk_size: int = 20
+    delay_ms: int = 8
+
+
 class StreamProcessor:
     """
     流式响应处理器
 
-    负责处理 SSE 流的解析、错误检测和响应生成。
+    负责处理 SSE 流的解析、错误检测、响应生成和平滑输出。
     从 ChatHandlerBase 中提取，使其职责更加单一。
     """
 
@@ -36,6 +53,9 @@ class StreamProcessor:
         request_id: str,
         default_parser: ResponseParser,
         on_streaming_start: Optional[Callable[[], None]] = None,
+        *,
+        collect_text: bool = False,
+        smoothing_config: Optional[StreamSmoothingConfig] = None,
     ):
         """
         初始化流处理器
@@ -44,10 +64,17 @@ class StreamProcessor:
             request_id: 请求 ID（用于日志）
             default_parser: 默认响应解析器
             on_streaming_start: 流开始时的回调（用于更新状态）
+            collect_text: 是否收集文本内容
+            smoothing_config: 流式平滑输出配置
         """
         self.request_id = request_id
         self.default_parser = default_parser
         self.on_streaming_start = on_streaming_start
+        self.collect_text = collect_text
+        self.smoothing_config = smoothing_config or StreamSmoothingConfig()
+
+        # 内容提取器缓存
+        self._extractors: dict[str, ContentExtractor] = {}
 
     def get_parser_for_provider(self, ctx: StreamContext) -> ResponseParser:
         """
@@ -112,18 +139,26 @@ class StreamProcessor:
             )
 
         # 提取文本
-        text = parser.extract_text_content(data)
-        if text:
-            ctx.collected_text += text
+        if self.collect_text:
+            text = parser.extract_text_content(data)
+            if text:
+                ctx.append_text(text)
 
         # 检查完成
         event_type = event_name or data.get("type", "")
         if event_type in ("response.completed", "message_stop"):
             ctx.has_completion = True
 
+        # 检查 OpenAI 格式的 finish_reason
+        choices = data.get("choices", [])
+        if choices and isinstance(choices, list) and len(choices) > 0:
+            finish_reason = choices[0].get("finish_reason")
+            if finish_reason is not None:
+                ctx.has_completion = True
+
     async def prefetch_and_check_error(
         self,
-        line_iterator: Any,
+        byte_iterator: Any,
         provider: Provider,
         endpoint: ProviderEndpoint,
         ctx: StreamContext,
@@ -136,97 +171,126 @@ class StreamProcessor:
         这种情况需要在流开始输出之前检测，以便触发重试逻辑。
 
         Args:
-            line_iterator: 行迭代器
+            byte_iterator: 字节流迭代器
             provider: Provider 对象
             endpoint: Endpoint 对象
             ctx: 流式上下文
             max_prefetch_lines: 最多预读行数
 
         Returns:
-            预读的行列表
+            预读的字节块列表
 
         Raises:
             EmbeddedErrorException: 如果检测到嵌套错误
         """
-        prefetched_lines: list = []
+        prefetched_chunks: list = []
         parser = self.get_parser_for_provider(ctx)
+        buffer = b""
+        line_count = 0
+        should_stop = False
+        # 使用增量解码器处理跨 chunk 的 UTF-8 字符
+        decoder = codecs.getincrementaldecoder("utf-8")(errors="replace")
 
         try:
-            line_count = 0
-            async for line in line_iterator:
-                prefetched_lines.append(line)
-                line_count += 1
+            async for chunk in byte_iterator:
+                prefetched_chunks.append(chunk)
+                buffer += chunk
 
-                normalized_line = line.rstrip("\r")
-                if not normalized_line or normalized_line.startswith(":"):
-                    if line_count >= max_prefetch_lines:
+                # 尝试按行解析缓冲区
+                while b"\n" in buffer:
+                    line_bytes, buffer = buffer.split(b"\n", 1)
+                    try:
+                        # 使用增量解码器，可以正确处理跨 chunk 的多字节字符
+                        line = decoder.decode(line_bytes + b"\n", False).rstrip("\r\n")
+                    except Exception as e:
+                        logger.warning(
+                            f"[{self.request_id}] 预读时 UTF-8 解码失败: {e}, "
+                            f"bytes={line_bytes[:50]!r}"
+                        )
+                        continue
+
+                    line_count += 1
+
+                    # 跳过空行和注释行
+                    if not line or line.startswith(":"):
+                        if line_count >= max_prefetch_lines:
+                            should_stop = True
+                            break
+                        continue
+
+                    # 尝试解析 SSE 数据
+                    data_str = line
+                    if line.startswith("data: "):
+                        data_str = line[6:]
+
+                    if data_str == "[DONE]":
+                        should_stop = True
                         break
-                    continue
 
-                # 尝试解析 SSE 数据
-                data_str = normalized_line
-                if normalized_line.startswith("data: "):
-                    data_str = normalized_line[6:]
+                    try:
+                        data = json.loads(data_str)
+                    except json.JSONDecodeError:
+                        if line_count >= max_prefetch_lines:
+                            should_stop = True
+                            break
+                        continue
 
-                if data_str == "[DONE]":
+                    # 使用解析器检查是否为错误响应
+                    if isinstance(data, dict) and parser.is_error_response(data):
+                        parsed = parser.parse_response(data, 200)
+                        logger.warning(
+                            f"  [{self.request_id}] 检测到嵌套错误: "
+                            f"Provider={provider.name}, "
+                            f"error_type={parsed.error_type}, "
+                            f"message={parsed.error_message}"
+                        )
+                        raise EmbeddedErrorException(
+                            provider_name=str(provider.name),
+                            error_code=(
+                                int(parsed.error_type)
+                                if parsed.error_type and parsed.error_type.isdigit()
+                                else None
+                            ),
+                            error_message=parsed.error_message,
+                            error_status=parsed.error_type,
+                        )
+
+                    # 预读到有效数据，没有错误，停止预读
+                    should_stop = True
                     break
 
-                try:
-                    data = json.loads(data_str)
-                except json.JSONDecodeError:
-                    if line_count >= max_prefetch_lines:
-                        break
-                    continue
-
-                # 使用解析器检查是否为错误响应
-                if isinstance(data, dict) and parser.is_error_response(data):
-                    parsed = parser.parse_response(data, 200)
-                    logger.warning(
-                        f"  [{self.request_id}] 检测到嵌套错误: "
-                        f"Provider={provider.name}, "
-                        f"error_type={parsed.error_type}, "
-                        f"message={parsed.error_message}"
-                    )
-                    raise EmbeddedErrorException(
-                        provider_name=str(provider.name),
-                        error_code=(
-                            int(parsed.error_type)
-                            if parsed.error_type and parsed.error_type.isdigit()
-                            else None
-                        ),
-                        error_message=parsed.error_message,
-                        error_status=parsed.error_type,
-                    )
-
-                # 预读到有效数据，没有错误，停止预读
-                break
+                if should_stop or line_count >= max_prefetch_lines:
+                    break
 
         except EmbeddedErrorException:
             raise
         except Exception as e:
             logger.debug(f"  [{self.request_id}] 预读流时发生异常: {e}")
 
-        return prefetched_lines
+        return prefetched_chunks
 
     async def create_response_stream(
         self,
         ctx: StreamContext,
-        line_iterator: Any,
+        byte_iterator: Any,
         response_ctx: Any,
         http_client: httpx.AsyncClient,
-        prefetched_lines: Optional[list] = None,
+        prefetched_chunks: Optional[list] = None,
+        *,
+        start_time: Optional[float] = None,
     ) -> AsyncGenerator[bytes, None]:
         """
         创建响应流生成器
 
-        统一的流生成器，支持带预读数据和不带预读数据两种情况。
+        从字节流中解析 SSE 数据并转发，支持预读数据。
 
         Args:
             ctx: 流式上下文
-            line_iterator: 行迭代器
+            byte_iterator: 字节流迭代器
             response_ctx: HTTP 响应上下文管理器
             http_client: HTTP 客户端
-            prefetched_lines: 预读的行列表（可选）
+            prefetched_chunks: 预读的字节块列表（可选）
+            start_time: 请求开始时间,用于计算 TTFB（可选）
 
         Yields:
             编码后的响应数据块
@@ -234,25 +298,82 @@ class StreamProcessor:
         try:
             sse_parser = SSEEventParser()
             streaming_started = False
+            buffer = b""
+            # 使用增量解码器处理跨 chunk 的 UTF-8 字符
+            decoder = codecs.getincrementaldecoder("utf-8")(errors="replace")
 
             # 处理预读数据
-            if prefetched_lines:
+            if prefetched_chunks:
                 if not streaming_started and self.on_streaming_start:
                     self.on_streaming_start()
                     streaming_started = True
 
-                for line in prefetched_lines:
-                    for chunk in self._process_line(ctx, sse_parser, line):
-                        yield chunk
+                for chunk in prefetched_chunks:
+                    # 记录首字时间 (TTFB) - 在 yield 之前记录
+                    if start_time is not None:
+                        ctx.record_first_byte_time(start_time)
+                        start_time = None  # 只记录一次
+
+                    # 把原始数据转发给客户端
+                    yield chunk
+
+                    buffer += chunk
+                    # 处理缓冲区中的完整行
+                    while b"\n" in buffer:
+                        line_bytes, buffer = buffer.split(b"\n", 1)
+                        try:
+                            # 使用增量解码器，可以正确处理跨 chunk 的多字节字符
+                            line = decoder.decode(line_bytes + b"\n", False)
+                            self._process_line(ctx, sse_parser, line)
+                        except Exception as e:
+                            # 解码失败，记录警告但继续处理
+                            logger.warning(
+                                f"[{self.request_id}] UTF-8 解码失败: {e}, "
+                                f"bytes={line_bytes[:50]!r}"
+                            )
+                            continue
 
             # 处理剩余的流数据
-            async for line in line_iterator:
+            async for chunk in byte_iterator:
                 if not streaming_started and self.on_streaming_start:
                     self.on_streaming_start()
                     streaming_started = True
 
-                for chunk in self._process_line(ctx, sse_parser, line):
-                    yield chunk
+                # 记录首字时间 (TTFB) - 在 yield 之前记录（如果预读数据为空）
+                if start_time is not None:
+                    ctx.record_first_byte_time(start_time)
+                    start_time = None  # 只记录一次
+
+                # 原始数据透传
+                yield chunk
+
+                buffer += chunk
+                # 处理缓冲区中的完整行
+                while b"\n" in buffer:
+                    line_bytes, buffer = buffer.split(b"\n", 1)
+                    try:
+                        # 使用增量解码器，可以正确处理跨 chunk 的多字节字符
+                        line = decoder.decode(line_bytes + b"\n", False)
+                        self._process_line(ctx, sse_parser, line)
+                    except Exception as e:
+                        # 解码失败，记录警告但继续处理
+                        logger.warning(
+                            f"[{self.request_id}] UTF-8 解码失败: {e}, "
+                            f"bytes={line_bytes[:50]!r}"
+                        )
+                        continue
+
+            # 处理剩余的缓冲区数据（如果有未完成的行）
+            if buffer:
+                try:
+                    # 使用 final=True 处理最后的不完整字符
+                    line = decoder.decode(buffer, True)
+                    self._process_line(ctx, sse_parser, line)
+                except Exception as e:
+                    logger.warning(
+                        f"[{self.request_id}] 处理剩余缓冲区失败: {e}, "
+                        f"bytes={buffer[:50]!r}"
+                    )
 
             # 处理剩余事件
             for event in sse_parser.flush():
@@ -268,7 +389,7 @@ class StreamProcessor:
         ctx: StreamContext,
         sse_parser: SSEEventParser,
         line: str,
-    ) -> list[bytes]:
+    ) -> None:
         """
         处理单行数据
 
@@ -276,26 +397,17 @@ class StreamProcessor:
             ctx: 流式上下文
             sse_parser: SSE 解析器
             line: 原始行数据
-
-        Returns:
-            要发送的数据块列表
         """
-        result: list[bytes] = []
-        normalized_line = line.rstrip("\r")
+        # SSEEventParser 以"去掉换行符"的单行文本作为输入；这里统一剔除 CR/LF，
+        # 避免把空行误判成 "\n" 并导致事件边界解析错误。
+        normalized_line = line.rstrip("\r\n")
         events = sse_parser.feed_line(normalized_line)
 
-        if normalized_line == "":
-            for event in events:
-                self.handle_sse_event(ctx, event.get("event"), event.get("data") or "")
-            result.append(b"\n")
-        else:
+        if normalized_line != "":
             ctx.chunk_count += 1
-            result.append((line + "\n").encode("utf-8"))
 
-            for event in events:
-                self.handle_sse_event(ctx, event.get("event"), event.get("data") or "")
-
-        return result
+        for event in events:
+            self.handle_sse_event(ctx, event.get("event"), event.get("data") or "")
 
     async def create_monitored_stream(
         self,
@@ -317,22 +429,201 @@ class StreamProcessor:
             响应数据块
         """
         try:
-            async for chunk in stream_generator:
-                if await is_disconnected():
-                    logger.warning(f"ID:{self.request_id} | Client disconnected")
-                    ctx.status_code = 499  # Client Closed Request
-                    ctx.error_message = "client_disconnected"
-                    break
-                yield chunk
+            # 使用后台任务检查断连，完全不阻塞流式传输
+            disconnected = False
+
+            async def check_disconnect_background() -> None:
+                nonlocal disconnected
+                while not disconnected and not ctx.has_completion:
+                    await asyncio.sleep(0.5)
+                    if await is_disconnected():
+                        disconnected = True
+                        break
+
+            # 启动后台检查任务
+            check_task = asyncio.create_task(check_disconnect_background())
+
+            try:
+                async for chunk in stream_generator:
+                    if disconnected:
+                        # 如果响应已完成，客户端断开不算失败
+                        if ctx.has_completion:
+                            logger.info(
+                                f"ID:{self.request_id} | Client disconnected after completion"
+                            )
+                        else:
+                            logger.warning(f"ID:{self.request_id} | Client disconnected")
+                            ctx.status_code = 499
+                            ctx.error_message = "client_disconnected"
+                        break
+                    yield chunk
+            finally:
+                check_task.cancel()
+                try:
+                    await check_task
+                except asyncio.CancelledError:
+                    pass
         except asyncio.CancelledError:
-            ctx.status_code = 499
-            ctx.error_message = "client_disconnected"
+            # 如果响应已完成，不标记为失败
+            if not ctx.has_completion:
+                ctx.status_code = 499
+                ctx.error_message = "client_disconnected"
             raise
         except Exception as e:
             ctx.status_code = 500
             ctx.error_message = str(e)
             raise
 
+    async def create_smoothed_stream(
+        self,
+        stream_generator: AsyncGenerator[bytes, None],
+    ) -> AsyncGenerator[bytes, None]:
+        """
+        创建平滑输出的流生成器
+
+        如果启用了平滑输出，将大 chunk 拆分成小块并添加微小延迟。
+        否则直接透传原始流。
+
+        Args:
+            stream_generator: 原始流生成器
+
+        Yields:
+            平滑处理后的响应数据块
+        """
+        if not self.smoothing_config.enabled:
+            # 未启用平滑输出，直接透传
+            async for chunk in stream_generator:
+                yield chunk
+            return
+
+        # 启用平滑输出
+        buffer = b""
+        is_first_content = True
+
+        async for chunk in stream_generator:
+            buffer += chunk
+
+            # 按双换行分割 SSE 事件（标准 SSE 格式）
+            while b"\n\n" in buffer:
+                event_block, buffer = buffer.split(b"\n\n", 1)
+                event_str = event_block.decode("utf-8", errors="replace")
+
+                # 解析事件块
+                lines = event_str.strip().split("\n")
+                data_str = None
+                event_type = ""
+
+                for line in lines:
+                    line = line.rstrip("\r")
+                    if line.startswith("event: "):
+                        event_type = line[7:].strip()
+                    elif line.startswith("data: "):
+                        data_str = line[6:]
+
+                # 没有 data 行，直接透传
+                if data_str is None:
+                    yield event_block + b"\n\n"
+                    continue
+
+                # [DONE] 直接透传
+                if data_str.strip() == "[DONE]":
+                    yield event_block + b"\n\n"
+                    continue
+
+                # 尝试解析 JSON
+                try:
+                    data = json.loads(data_str)
+                except json.JSONDecodeError:
+                    yield event_block + b"\n\n"
+                    continue
+
+                # 检测格式并提取内容
+                content, extractor = self._detect_format_and_extract(data)
+
+                # 只有内容长度大于 1 才需要平滑处理
+                if content and len(content) > 1 and extractor:
+                    # 获取配置的延迟
+                    delay_seconds = self._calculate_delay()
+
+                    # 拆分内容
+                    content_chunks = self._split_content(content)
+
+                    for i, sub_content in enumerate(content_chunks):
+                        is_first = is_first_content and i == 0
+
+                        # 使用提取器创建新 chunk
+                        sse_chunk = extractor.create_chunk(
+                            data,
+                            sub_content,
+                            event_type=event_type,
+                            is_first=is_first,
+                        )
+
+                        yield sse_chunk
+
+                        # 除了最后一个块，其他块之间加延迟
+                        if i < len(content_chunks) - 1:
+                            await asyncio.sleep(delay_seconds)
+
+                    is_first_content = False
+                else:
+                    # 不需要拆分，直接透传
+                    yield event_block + b"\n\n"
+                    if content:
+                        is_first_content = False
+
+        # 处理剩余数据
+        if buffer:
+            yield buffer
+
+    def _get_extractor(self, format_name: str) -> Optional[ContentExtractor]:
+        """获取或创建格式对应的提取器（带缓存）"""
+        if format_name not in self._extractors:
+            extractor = get_extractor(format_name)
+            if extractor:
+                self._extractors[format_name] = extractor
+        return self._extractors.get(format_name)
+
+    def _detect_format_and_extract(
+        self, data: dict
+    ) -> tuple[Optional[str], Optional[ContentExtractor]]:
+        """
+        检测数据格式并提取内容
+
+        依次尝试各格式的提取器，返回第一个成功提取内容的结果。
+
+        Returns:
+            (content, extractor): 提取的内容和对应的提取器
+        """
+        for format_name in get_extractor_formats():
+            extractor = self._get_extractor(format_name)
+            if extractor:
+                content = extractor.extract_content(data)
+                if content is not None:
+                    return content, extractor
+
+        return None, None
+
+    def _calculate_delay(self) -> float:
+        """获取配置的延迟（秒）"""
+        return self.smoothing_config.delay_ms / 1000.0
+
+    def _split_content(self, content: str) -> list[str]:
+        """
+        按块拆分文本
+        """
+        chunk_size = self.smoothing_config.chunk_size
+        text_length = len(content)
+
+        if text_length <= chunk_size:
+            return [content]
+
+        # 按块拆分
+        chunks = []
+        for i in range(0, text_length, chunk_size):
+            chunks.append(content[i : i + chunk_size])
+        return chunks
+
     async def _cleanup(
         self,
         response_ctx: Any,
@@ -347,3 +638,128 @@ class StreamProcessor:
             await http_client.aclose()
         except Exception:
             pass
+
+
+async def create_smoothed_stream(
+    stream_generator: AsyncGenerator[bytes, None],
+    chunk_size: int = 20,
+    delay_ms: int = 8,
+) -> AsyncGenerator[bytes, None]:
+    """
+    独立的平滑流生成函数
+
+    供 CLI handler 等场景使用，无需创建完整的 StreamProcessor 实例。
+
+    Args:
+        stream_generator: 原始流生成器
+        chunk_size: 每块字符数
+        delay_ms: 每块之间的延迟毫秒数
+
+    Yields:
+        平滑处理后的响应数据块
+    """
+    processor = _LightweightSmoother(chunk_size=chunk_size, delay_ms=delay_ms)
+    async for chunk in processor.smooth(stream_generator):
+        yield chunk
+
+
+class _LightweightSmoother:
+    """
+    轻量级平滑处理器
+
+    只包含平滑输出所需的最小逻辑，不依赖 StreamProcessor 的其他功能。
+    """
+
+    def __init__(self, chunk_size: int = 20, delay_ms: int = 8) -> None:
+        self.chunk_size = chunk_size
+        self.delay_ms = delay_ms
+        self._extractors: dict[str, ContentExtractor] = {}
+
+    def _get_extractor(self, format_name: str) -> Optional[ContentExtractor]:
+        if format_name not in self._extractors:
+            extractor = get_extractor(format_name)
+            if extractor:
+                self._extractors[format_name] = extractor
+        return self._extractors.get(format_name)
+
+    def _detect_format_and_extract(
+        self, data: dict
+    ) -> tuple[Optional[str], Optional[ContentExtractor]]:
+        for format_name in get_extractor_formats():
+            extractor = self._get_extractor(format_name)
+            if extractor:
+                content = extractor.extract_content(data)
+                if content is not None:
+                    return content, extractor
+        return None, None
+
+    def _calculate_delay(self) -> float:
+        return self.delay_ms / 1000.0
+
+    def _split_content(self, content: str) -> list[str]:
+        text_length = len(content)
+        if text_length <= self.chunk_size:
+            return [content]
+        return [content[i : i + self.chunk_size] for i in range(0, text_length, self.chunk_size)]
+
+    async def smooth(
+        self, stream_generator: AsyncGenerator[bytes, None]
+    ) -> AsyncGenerator[bytes, None]:
+        buffer = b""
+        is_first_content = True
+
+        async for chunk in stream_generator:
+            buffer += chunk
+
+            while b"\n\n" in buffer:
+                event_block, buffer = buffer.split(b"\n\n", 1)
+                event_str = event_block.decode("utf-8", errors="replace")
+
+                lines = event_str.strip().split("\n")
+                data_str = None
+                event_type = ""
+
+                for line in lines:
+                    line = line.rstrip("\r")
+                    if line.startswith("event: "):
+                        event_type = line[7:].strip()
+                    elif line.startswith("data: "):
+                        data_str = line[6:]
+
+                if data_str is None:
+                    yield event_block + b"\n\n"
+                    continue
+
+                if data_str.strip() == "[DONE]":
+                    yield event_block + b"\n\n"
+                    continue
+
+                try:
+                    data = json.loads(data_str)
+                except json.JSONDecodeError:
+                    yield event_block + b"\n\n"
+                    continue
+
+                content, extractor = self._detect_format_and_extract(data)
+
+                if content and len(content) > 1 and extractor:
+                    delay_seconds = self._calculate_delay()
+                    content_chunks = self._split_content(content)
+
+                    for i, sub_content in enumerate(content_chunks):
+                        is_first = is_first_content and i == 0
+                        sse_chunk = extractor.create_chunk(
+                            data, sub_content, event_type=event_type, is_first=is_first
+                        )
+                        yield sse_chunk
+                        if i < len(content_chunks) - 1:
+                            await asyncio.sleep(delay_seconds)
+
+                    is_first_content = False
+                else:
+                    yield event_block + b"\n\n"
+                    if content:
+                        is_first_content = False
+
+        if buffer:
+            yield buffer

src/api/handlers/base/stream_telemetry.py

@@ -8,6 +8,7 @@
 """
 
 import asyncio
+import time
 from typing import Any, Dict, Optional
 
 from sqlalchemy.orm import Session
@@ -57,7 +58,7 @@ class StreamTelemetryRecorder:
         ctx: StreamContext,
         original_headers: Dict[str, str],
         original_request_body: Dict[str, Any],
-        response_time_ms: int,
+        start_time: float,
     ) -> None:
         """
         记录流式统计信息
@@ -66,11 +67,15 @@ class StreamTelemetryRecorder:
             ctx: 流式上下文
             original_headers: 原始请求头
             original_request_body: 原始请求体
-            response_time_ms: 响应时间（毫秒）
+            start_time: 请求开始时间 (time.time())
         """
         bg_db = None
 
         try:
+            # 在流结束后计算响应时间，与首字时间使用相同的时间基准
+            # 注意：不要把统计延迟（stream_stats_delay）算进响应时间里
+            response_time_ms = int((time.time() - start_time) * 1000)
+
             await asyncio.sleep(config.stream_stats_delay)  # 等待流完全关闭
 
             if not ctx.provider_name:
@@ -155,6 +160,7 @@ class StreamTelemetryRecorder:
             input_tokens=ctx.input_tokens,
             output_tokens=ctx.output_tokens,
             response_time_ms=response_time_ms,
+            first_byte_time_ms=ctx.first_byte_time_ms,  # 传递首字时间
             status_code=ctx.status_code,
             request_headers=original_headers,
             request_body=actual_request_body,