refactor(ui): 重构批量模型管理对话框为单列勾选模式

- BatchAssignModelsDialog: 从左右双栏改为单列勾选，支持统一保存
- KeyAllowedModelsEditDialog: 添加分组折叠功能，优化布局
- Dialog: 调整移动端内边距适配
- ProviderDetailDrawer: 密钥列表改为两行显示
- KeyFormDialog: 缩小对话框尺寸
- ModelsTab: 调整表格列宽

.github/workflows/docker-publish.yml

@@ -146,10 +146,33 @@ jobs:
             type=semver,pattern={{major}}.{{minor}}
             type=sha,prefix=
 
+      - name: Extract version from tag
+        id: version
+        run: |
+          # 从 tag 提取版本号，如 v0.2.5 -> 0.2.5
+          VERSION="${GITHUB_REF#refs/tags/v}"
+          if [ "$VERSION" = "$GITHUB_REF" ]; then
+            # 不是 tag 触发，使用 git describe
+            VERSION=$(git describe --tags --always | sed 's/^v//')
+          fi
+          echo "version=$VERSION" >> $GITHUB_OUTPUT
+          echo "Extracted version: $VERSION"
+
       - name: Update Dockerfile.app to use registry base image
         run: |
           sed -i "s|FROM aether-base:latest AS builder|FROM ${{ env.REGISTRY }}/${{ env.BASE_IMAGE_NAME }}:latest AS builder|g" Dockerfile.app
 
+      - name: Generate version file
+        run: |
+          # 生成 _version.py 文件
+          cat > src/_version.py << EOF
+          # Auto-generated by CI
+          __version__ = '${{ steps.version.outputs.version }}'
+          __version_tuple__ = tuple(int(x) for x in '${{ steps.version.outputs.version }}'.split('.') if x.isdigit())
+          version = __version__
+          version_tuple = __version_tuple__
+          EOF
+
       - name: Build and push app image
         uses: docker/build-push-action@v5
         with:

.gitignore

@@ -224,3 +224,6 @@ extracted_*.ts
 .deps-hash
 .code-hash
 .migration-hash
+
+# Version file (auto-generated by hatch-vcs)
+src/_version.py

Dockerfile.app

@@ -58,6 +58,15 @@ RUN printf '%s\n' \
 '    index index.html;' \
 '    client_max_body_size 100M;' \
 '' \
+'    # gzip 压缩配置（对 base64 图片等非流式响应有效）' \
+'    gzip on;' \
+'    gzip_min_length 256;' \
+'    gzip_comp_level 5;' \
+'    gzip_vary on;' \
+'    gzip_proxied any;' \
+'    gzip_types application/json text/plain text/css text/javascript application/javascript application/octet-stream;' \
+'    gzip_disable "msie6";' \
+'' \
 '    location ~* \.(js|css|png|jpg|jpeg|gif|ico|svg|woff|woff2|ttf|eot)$ {' \
 '        expires 1y;' \
 '        add_header Cache-Control "public, no-transform";' \
@@ -73,6 +82,15 @@ RUN printf '%s\n' \
 '        try_files $uri $uri/ /index.html;' \
 '    }' \
 '' \
+'    location ~ ^/(docs|redoc|openapi\\.json)$ {' \
+'        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 $real_ip;' \
+'        proxy_set_header X-Forwarded-For $forwarded_for;' \
+'        proxy_set_header X-Forwarded-Proto $scheme;' \
+'    }' \
+'' \
 '    location / {' \
 '        try_files $uri $uri/ @backend;' \
 '    }' \
@@ -109,14 +127,14 @@ RUN printf '%s\n' \
 'pidfile=/var/run/supervisord.pid' \
 '' \
 '[program:nginx]' \
-'command=/bin/bash -c "sed \"s/PORT_PLACEHOLDER/${PORT:-8084}/g\" /etc/nginx/sites-available/default.template > /etc/nginx/sites-available/default && /usr/sbin/nginx -g \"daemon off;\""' \
+'command=/bin/bash -c "sed \"s/PORT_PLACEHOLDER/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 --preload -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' \
+'command=gunicorn src.main:app --preload -w %(ENV_GUNICORN_WORKERS)s -k uvicorn.workers.UvicornWorker --bind 127.0.0.1:8084 --timeout 120 --access-logfile - --error-logfile - --log-level info' \
 'directory=/app' \
 'autostart=true' \
 'autorestart=true' \
@@ -129,17 +147,23 @@ RUN printf '%s\n' \
 # 创建目录
 RUN mkdir -p /var/log/supervisor /app/logs /app/data
 
+# 入口脚本（启动前执行迁移）
+COPY entrypoint.sh /entrypoint.sh
+RUN chmod +x /entrypoint.sh
+
 # 环境变量
 ENV PYTHONUNBUFFERED=1 \
     PYTHONDONTWRITEBYTECODE=1 \
     PYTHONIOENCODING=utf-8 \
     LANG=C.UTF-8 \
     LC_ALL=C.UTF-8 \
-    PORT=8084
+    PORT=8084 \
+    GUNICORN_WORKERS=4
 
 EXPOSE 80
 
 HEALTHCHECK --interval=30s --timeout=10s --start-period=5s --retries=3 \
     CMD curl -f http://localhost/health || exit 1
 
+ENTRYPOINT ["/entrypoint.sh"]
 CMD ["/usr/bin/supervisord", "-c", "/etc/supervisor/conf.d/supervisord.conf"]

Dockerfile.app.local

@@ -59,6 +59,15 @@ RUN printf '%s\n' \
 '    index index.html;' \
 '    client_max_body_size 100M;' \
 '' \
+'    # gzip 压缩配置（对 base64 图片等非流式响应有效）' \
+'    gzip on;' \
+'    gzip_min_length 256;' \
+'    gzip_comp_level 5;' \
+'    gzip_vary on;' \
+'    gzip_proxied any;' \
+'    gzip_types application/json text/plain text/css text/javascript application/javascript application/octet-stream;' \
+'    gzip_disable "msie6";' \
+'' \
 '    location ~* \.(js|css|png|jpg|jpeg|gif|ico|svg|woff|woff2|ttf|eot)$ {' \
 '        expires 1y;' \
 '        add_header Cache-Control "public, no-transform";' \
@@ -130,6 +139,10 @@ RUN printf '%s\n' \
 # 创建目录
 RUN mkdir -p /var/log/supervisor /app/logs /app/data
 
+# 入口脚本（启动前执行迁移）
+COPY entrypoint.sh /entrypoint.sh
+RUN chmod +x /entrypoint.sh
+
 # 环境变量
 ENV PYTHONUNBUFFERED=1 \
     PYTHONDONTWRITEBYTECODE=1 \
@@ -143,4 +156,5 @@ EXPOSE 80
 HEALTHCHECK --interval=30s --timeout=10s --start-period=5s --retries=3 \
     CMD curl -f http://localhost/health || exit 1
 
+ENTRYPOINT ["/entrypoint.sh"]
 CMD ["/usr/bin/supervisord", "-c", "/etc/supervisor/conf.d/supervisord.conf"]

README.md

@@ -57,14 +57,8 @@ cd Aether
 cp .env.example .env
 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. 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
+# 3. 部署 / 更新（自动执行数据库迁移）
+docker compose pull && docker compose up -d
 ```
 
 ### Docker Compose（本地构建镜像）

alembic/versions/20260106_1524_ad55f1d008b7_add_management_tokens_table.py

@@ -0,0 +1,131 @@
+"""add_management_tokens_table
+
+Revision ID: ad55f1d008b7
+Revises: c3d4e5f6g7h8
+Create Date: 2026-01-06 15:24:10.660394+00:00
+
+"""
+from alembic import op
+import sqlalchemy as sa
+from sqlalchemy import inspect
+
+# revision identifiers, used by Alembic.
+revision = 'ad55f1d008b7'
+down_revision = 'c3d4e5f6g7h8'
+branch_labels = None
+depends_on = None
+
+
+def table_exists(table_name: str) -> bool:
+    """检查表是否存在"""
+    conn = op.get_bind()
+    inspector = inspect(conn)
+    return table_name in inspector.get_table_names()
+
+
+def index_exists(table_name: str, index_name: str) -> bool:
+    """检查索引是否存在"""
+    conn = op.get_bind()
+    inspector = inspect(conn)
+    try:
+        indexes = inspector.get_indexes(table_name)
+        return any(idx["name"] == index_name for idx in indexes)
+    except Exception:
+        return False
+
+
+def constraint_exists(table_name: str, constraint_name: str) -> bool:
+    """检查约束是否存在"""
+    conn = op.get_bind()
+    inspector = inspect(conn)
+    try:
+        constraints = inspector.get_unique_constraints(table_name)
+        if any(c["name"] == constraint_name for c in constraints):
+            return True
+        # 也检查 check 约束
+        check_constraints = inspector.get_check_constraints(table_name)
+        if any(c["name"] == constraint_name for c in check_constraints):
+            return True
+        return False
+    except Exception:
+        return False
+
+
+def upgrade() -> None:
+    """应用迁移：创建 management_tokens 表"""
+    # 幂等性检查
+    if table_exists("management_tokens"):
+        # 表已存在，检查是否需要添加约束
+        if not constraint_exists("management_tokens", "uq_management_tokens_user_name"):
+            op.create_unique_constraint(
+                "uq_management_tokens_user_name",
+                "management_tokens",
+                ["user_id", "name"],
+            )
+        # 添加 IP 白名单非空检查约束
+        if not constraint_exists("management_tokens", "check_allowed_ips_not_empty"):
+            op.create_check_constraint(
+                "check_allowed_ips_not_empty",
+                "management_tokens",
+                "allowed_ips IS NULL OR allowed_ips::text = 'null' OR json_array_length(allowed_ips) > 0",
+            )
+        return
+
+    op.create_table('management_tokens',
+        sa.Column('id', sa.String(length=36), nullable=False),
+        sa.Column('user_id', sa.String(length=36), nullable=False),
+        sa.Column('token_hash', sa.String(length=64), nullable=False),
+        sa.Column('token_prefix', sa.String(length=12), nullable=True),
+        sa.Column('name', sa.String(length=100), nullable=False),
+        sa.Column('description', sa.Text(), nullable=True),
+        sa.Column('allowed_ips', sa.JSON(), nullable=True),
+        sa.Column('expires_at', sa.DateTime(timezone=True), nullable=True),
+        sa.Column('last_used_at', sa.DateTime(timezone=True), nullable=True),
+        sa.Column('last_used_ip', sa.String(length=45), nullable=True),
+        sa.Column('usage_count', sa.Integer(), server_default='0', nullable=False),
+        sa.Column('is_active', sa.Boolean(), server_default='true', nullable=False),
+        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.ForeignKeyConstraint(['user_id'], ['users.id'], ondelete='CASCADE'),
+        sa.PrimaryKeyConstraint('id')
+    )
+    op.create_index('idx_management_tokens_is_active', 'management_tokens', ['is_active'], unique=False)
+    op.create_index('idx_management_tokens_user_id', 'management_tokens', ['user_id'], unique=False)
+    op.create_index(op.f('ix_management_tokens_token_hash'), 'management_tokens', ['token_hash'], unique=True)
+    # 添加用户名称唯一约束
+    op.create_unique_constraint(
+        "uq_management_tokens_user_name",
+        "management_tokens",
+        ["user_id", "name"],
+    )
+    # 添加 IP 白名单非空检查约束
+    # 注意：JSON 类型的 NULL 可能被序列化为 JSON 'null'，需要同时处理
+    op.create_check_constraint(
+        "check_allowed_ips_not_empty",
+        "management_tokens",
+        "allowed_ips IS NULL OR allowed_ips::text = 'null' OR json_array_length(allowed_ips) > 0",
+    )
+
+
+def downgrade() -> None:
+    """回滚迁移：删除 management_tokens 表"""
+    # 幂等性检查
+    if not table_exists("management_tokens"):
+        return
+
+    # 删除约束
+    if constraint_exists("management_tokens", "check_allowed_ips_not_empty"):
+        op.drop_constraint("check_allowed_ips_not_empty", "management_tokens", type_="check")
+    if constraint_exists("management_tokens", "uq_management_tokens_user_name"):
+        op.drop_constraint("uq_management_tokens_user_name", "management_tokens", type_="unique")
+
+    # 删除索引
+    if index_exists("management_tokens", "ix_management_tokens_token_hash"):
+        op.drop_index(op.f('ix_management_tokens_token_hash'), table_name='management_tokens')
+    if index_exists("management_tokens", "idx_management_tokens_user_id"):
+        op.drop_index('idx_management_tokens_user_id', table_name='management_tokens')
+    if index_exists("management_tokens", "idx_management_tokens_is_active"):
+        op.drop_index('idx_management_tokens_is_active', table_name='management_tokens')
+
+    # 删除表
+    op.drop_table('management_tokens')

alembic/versions/20260107_1120_02a45b66b7c4_cleanup_allowed_endpoints_fields.py

@@ -0,0 +1,73 @@
+"""cleanup ambiguous database fields
+
+Revision ID: 02a45b66b7c4
+Revises: ad55f1d008b7
+Create Date: 2026-01-07 11:20:12.684426+00:00
+
+变更内容：
+1. users 表：重命名 allowed_endpoints 为 allowed_api_formats（修正历史命名错误）
+2. api_keys 表：删除 allowed_endpoints 字段（未使用的功能）
+3. providers 表：删除 rate_limit 字段（与 rpm_limit 功能重复，且未使用）
+4. usage 表：重命名 provider 为 provider_name（避免与 provider_id 外键混淆）
+"""
+from alembic import op
+import sqlalchemy as sa
+from sqlalchemy import inspect
+
+
+# revision identifiers, used by Alembic.
+revision = '02a45b66b7c4'
+down_revision = 'ad55f1d008b7'
+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 upgrade() -> None:
+    """
+    1. users.allowed_endpoints -> allowed_api_formats（重命名）
+    2. api_keys.allowed_endpoints 删除
+    3. providers.rate_limit 删除（与 rpm_limit 重复）
+    4. usage.provider -> provider_name（重命名）
+    """
+    # 1. users 表：重命名 allowed_endpoints 为 allowed_api_formats
+    if _column_exists('users', 'allowed_endpoints'):
+        op.alter_column('users', 'allowed_endpoints', new_column_name='allowed_api_formats')
+
+    # 2. api_keys 表：删除 allowed_endpoints 字段
+    if _column_exists('api_keys', 'allowed_endpoints'):
+        op.drop_column('api_keys', 'allowed_endpoints')
+
+    # 3. providers 表：删除 rate_limit 字段（与 rpm_limit 功能重复）
+    if _column_exists('providers', 'rate_limit'):
+        op.drop_column('providers', 'rate_limit')
+
+    # 4. usage 表：重命名 provider 为 provider_name
+    if _column_exists('usage', 'provider'):
+        op.alter_column('usage', 'provider', new_column_name='provider_name')
+
+
+def downgrade() -> None:
+    """回滚：恢复原字段"""
+    # 4. usage 表：将 provider_name 改回 provider
+    if _column_exists('usage', 'provider_name'):
+        op.alter_column('usage', 'provider_name', new_column_name='provider')
+
+    # 3. providers 表：恢复 rate_limit 字段
+    if not _column_exists('providers', 'rate_limit'):
+        op.add_column('providers', sa.Column('rate_limit', sa.Integer(), nullable=True))
+
+    # 2. api_keys 表：恢复 allowed_endpoints 字段
+    if not _column_exists('api_keys', 'allowed_endpoints'):
+        op.add_column('api_keys', sa.Column('allowed_endpoints', sa.JSON(), nullable=True))
+
+    # 1. users 表：将 allowed_api_formats 改回 allowed_endpoints
+    if _column_exists('users', 'allowed_api_formats'):
+        op.alter_column('users', 'allowed_api_formats', new_column_name='allowed_endpoints')

alembic/versions/20260110_2000_consolidated_schema_updates.py

@@ -0,0 +1,530 @@
+"""consolidated schema updates
+
+Revision ID: m4n5o6p7q8r9
+Revises: 02a45b66b7c4
+Create Date: 2026-01-10 20:00:00.000000
+
+This migration consolidates all schema changes from 2026-01-08 to 2026-01-10:
+
+1. provider_api_keys: Key 直接关联 Provider (provider_id, api_formats)
+2. provider_api_keys: 添加 rate_multipliers JSON 字段（按格式费率）
+3. models: global_model_id 改为可空（支持独立 ProviderModel）
+4. providers: 添加 timeout, max_retries, proxy（从 endpoint 迁移）
+5. providers: display_name 重命名为 name，删除原 name
+6. provider_api_keys: max_concurrent -> rpm_limit（并发改 RPM）
+7. provider_api_keys: 健康度改为按格式存储（health_by_format, circuit_breaker_by_format）
+8. provider_endpoints: 删除废弃的 rate_limit 列
+9. usage: 添加 client_response_headers 字段
+10. provider_api_keys: 删除 endpoint_id（Key 不再与 Endpoint 绑定）
+11. provider_endpoints: 删除废弃的 max_concurrent 列
+12. providers: 删除废弃的 rpm_limit, rpm_used, rpm_reset_at 列
+"""
+
+import logging
+from alembic import op
+import sqlalchemy as sa
+from sqlalchemy.dialects import postgresql
+from sqlalchemy import inspect
+
+
+# 配置日志
+alembic_logger = logging.getLogger("alembic.runtime.migration")
+
+revision = "m4n5o6p7q8r9"
+down_revision = "02a45b66b7c4"
+branch_labels = None
+depends_on = None
+
+
+def _column_exists(table_name: str, column_name: str) -> bool:
+    """Check if a column exists in the table"""
+    bind = op.get_bind()
+    inspector = inspect(bind)
+    columns = [col["name"] for col in inspector.get_columns(table_name)]
+    return column_name in columns
+
+
+def _constraint_exists(table_name: str, constraint_name: str) -> bool:
+    """Check if a constraint exists"""
+    bind = op.get_bind()
+    inspector = inspect(bind)
+    fks = inspector.get_foreign_keys(table_name)
+    return any(fk.get("name") == constraint_name for fk in fks)
+
+
+def _index_exists(table_name: str, index_name: str) -> bool:
+    """Check if an index exists"""
+    bind = op.get_bind()
+    inspector = inspect(bind)
+    indexes = inspector.get_indexes(table_name)
+    return any(idx.get("name") == index_name for idx in indexes)
+
+
+def upgrade() -> None:
+    """Apply all consolidated schema changes"""
+    bind = op.get_bind()
+
+    # ========== 1. provider_api_keys: 添加 provider_id 和 api_formats ==========
+    if not _column_exists("provider_api_keys", "provider_id"):
+        op.add_column("provider_api_keys", sa.Column("provider_id", sa.String(36), nullable=True))
+
+        # 数据迁移：从 endpoint 获取 provider_id
+        op.execute("""
+            UPDATE provider_api_keys k
+            SET provider_id = e.provider_id
+            FROM provider_endpoints e
+            WHERE k.endpoint_id = e.id AND k.provider_id IS NULL
+        """)
+
+        # 检查无法关联的孤儿 Key
+        result = bind.execute(sa.text(
+            "SELECT COUNT(*) FROM provider_api_keys WHERE provider_id IS NULL"
+        ))
+        orphan_count = result.scalar() or 0
+        if orphan_count > 0:
+            # 使用 logger 记录更明显的告警
+            alembic_logger.warning("=" * 60)
+            alembic_logger.warning(f"[MIGRATION WARNING] 发现 {orphan_count} 个无法关联 Provider 的孤儿 Key")
+            alembic_logger.warning("=" * 60)
+            alembic_logger.info("正在备份孤儿 Key 到 _orphan_api_keys_backup 表...")
+
+            # 先备份孤儿数据到临时表，避免数据丢失
+            op.execute("""
+                CREATE TABLE IF NOT EXISTS _orphan_api_keys_backup AS
+                SELECT *, NOW() as backup_at
+                FROM provider_api_keys
+                WHERE provider_id IS NULL
+            """)
+
+            # 记录备份的 Key ID
+            orphan_ids = bind.execute(sa.text(
+                "SELECT id, name FROM provider_api_keys WHERE provider_id IS NULL"
+            )).fetchall()
+            alembic_logger.info("备份的孤儿 Key 列表：")
+            for key_id, key_name in orphan_ids:
+                alembic_logger.info(f"  - Key: {key_name} (ID: {key_id})")
+
+            # 删除孤儿数据
+            op.execute("DELETE FROM provider_api_keys WHERE provider_id IS NULL")
+            alembic_logger.info(f"已备份并删除 {orphan_count} 个孤儿 Key")
+
+            # 提供恢复指南
+            alembic_logger.warning("-" * 60)
+            alembic_logger.warning("[恢复指南] 如需恢复孤儿 Key：")
+            alembic_logger.warning("  1. 查询备份表: SELECT * FROM _orphan_api_keys_backup;")
+            alembic_logger.warning("  2. 确定正确的 provider_id")
+            alembic_logger.warning("  3. 执行恢复:")
+            alembic_logger.warning("     INSERT INTO provider_api_keys (...)")
+            alembic_logger.warning("     SELECT ... FROM _orphan_api_keys_backup WHERE ...;")
+            alembic_logger.warning("-" * 60)
+
+        # 设置 NOT NULL 并创建外键
+        op.alter_column("provider_api_keys", "provider_id", nullable=False)
+
+        if not _constraint_exists("provider_api_keys", "fk_provider_api_keys_provider"):
+            op.create_foreign_key(
+                "fk_provider_api_keys_provider",
+                "provider_api_keys",
+                "providers",
+                ["provider_id"],
+                ["id"],
+                ondelete="CASCADE",
+            )
+
+        if not _index_exists("provider_api_keys", "idx_provider_api_keys_provider_id"):
+            op.create_index("idx_provider_api_keys_provider_id", "provider_api_keys", ["provider_id"])
+
+    if not _column_exists("provider_api_keys", "api_formats"):
+        op.add_column("provider_api_keys", sa.Column("api_formats", sa.JSON(), nullable=True))
+
+        # 数据迁移：从 endpoint 获取 api_format
+        op.execute("""
+            UPDATE provider_api_keys k
+            SET api_formats = json_build_array(e.api_format)
+            FROM provider_endpoints e
+            WHERE k.endpoint_id = e.id AND k.api_formats IS NULL
+        """)
+
+        op.alter_column("provider_api_keys", "api_formats", nullable=False, server_default="[]")
+
+    # 修改 endpoint_id 为可空，外键改为 SET NULL
+    if _constraint_exists("provider_api_keys", "provider_api_keys_endpoint_id_fkey"):
+        op.drop_constraint("provider_api_keys_endpoint_id_fkey", "provider_api_keys", type_="foreignkey")
+        op.alter_column("provider_api_keys", "endpoint_id", nullable=True)
+        # 不再重建外键，因为后面会删除这个字段
+
+    # ========== 2. provider_api_keys: 添加 rate_multipliers ==========
+    if not _column_exists("provider_api_keys", "rate_multipliers"):
+        op.add_column(
+            "provider_api_keys",
+            sa.Column("rate_multipliers", postgresql.JSON(astext_type=sa.Text()), nullable=True),
+        )
+
+        # 数据迁移：将 rate_multiplier 按 api_formats 转换
+        op.execute("""
+            UPDATE provider_api_keys
+            SET rate_multipliers = (
+                SELECT jsonb_object_agg(elem, rate_multiplier)
+                FROM jsonb_array_elements_text(api_formats::jsonb) AS elem
+            )
+            WHERE api_formats IS NOT NULL
+              AND api_formats::text != '[]'
+              AND api_formats::text != 'null'
+              AND rate_multipliers IS NULL
+        """)
+
+    # ========== 3. models: global_model_id 改为可空 ==========
+    op.alter_column("models", "global_model_id", existing_type=sa.String(36), nullable=True)
+
+    # ========== 4. providers: 添加 timeout, max_retries, proxy ==========
+    if not _column_exists("providers", "timeout"):
+        op.add_column(
+            "providers",
+            sa.Column("timeout", sa.Integer(), nullable=True, comment="请求超时（秒）"),
+        )
+
+    if not _column_exists("providers", "max_retries"):
+        op.add_column(
+            "providers",
+            sa.Column("max_retries", sa.Integer(), nullable=True, comment="最大重试次数"),
+        )
+
+    if not _column_exists("providers", "proxy"):
+        op.add_column(
+            "providers",
+            sa.Column("proxy", postgresql.JSONB(), nullable=True, comment="代理配置"),
+        )
+
+    # 从端点迁移数据到 provider
+    op.execute("""
+        UPDATE providers p
+        SET
+            timeout = COALESCE(
+                p.timeout,
+                (SELECT MAX(e.timeout) FROM provider_endpoints e WHERE e.provider_id = p.id AND e.timeout IS NOT NULL),
+                300
+            ),
+            max_retries = COALESCE(
+                p.max_retries,
+                (SELECT MAX(e.max_retries) FROM provider_endpoints e WHERE e.provider_id = p.id AND e.max_retries IS NOT NULL),
+                2
+            ),
+            proxy = COALESCE(
+                p.proxy,
+                (SELECT e.proxy FROM provider_endpoints e WHERE e.provider_id = p.id AND e.proxy IS NOT NULL ORDER BY e.created_at LIMIT 1)
+            )
+        WHERE p.timeout IS NULL OR p.max_retries IS NULL
+    """)
+
+    # ========== 5. providers: display_name -> name ==========
+    # 注意：这里假设 display_name 已经被重命名为 name
+    # 如果 display_name 仍然存在，则需要执行重命名
+    if _column_exists("providers", "display_name"):
+        # 删除旧的 name 索引
+        if _index_exists("providers", "ix_providers_name"):
+            op.drop_index("ix_providers_name", table_name="providers")
+
+        # 如果存在旧的 name 列，先删除
+        if _column_exists("providers", "name"):
+            op.drop_column("providers", "name")
+
+        # 重命名 display_name 为 name
+        op.alter_column("providers", "display_name", new_column_name="name")
+
+        # 创建新索引
+        op.create_index("ix_providers_name", "providers", ["name"], unique=True)
+
+    # ========== 6. provider_api_keys: max_concurrent -> rpm_limit ==========
+    if _column_exists("provider_api_keys", "max_concurrent"):
+        op.alter_column("provider_api_keys", "max_concurrent", new_column_name="rpm_limit")
+
+    if _column_exists("provider_api_keys", "learned_max_concurrent"):
+        op.alter_column("provider_api_keys", "learned_max_concurrent", new_column_name="learned_rpm_limit")
+
+    if _column_exists("provider_api_keys", "last_concurrent_peak"):
+        op.alter_column("provider_api_keys", "last_concurrent_peak", new_column_name="last_rpm_peak")
+
+    # 删除废弃字段
+    for col in ["rate_limit", "daily_limit", "monthly_limit"]:
+        if _column_exists("provider_api_keys", col):
+            op.drop_column("provider_api_keys", col)
+
+    # ========== 7. provider_api_keys: 健康度改为按格式存储 ==========
+    if not _column_exists("provider_api_keys", "health_by_format"):
+        op.add_column(
+            "provider_api_keys",
+            sa.Column(
+                "health_by_format",
+                postgresql.JSONB(astext_type=sa.Text()),
+                nullable=True,
+                comment="按API格式存储的健康度数据",
+            ),
+        )
+
+    if not _column_exists("provider_api_keys", "circuit_breaker_by_format"):
+        op.add_column(
+            "provider_api_keys",
+            sa.Column(
+                "circuit_breaker_by_format",
+                postgresql.JSONB(astext_type=sa.Text()),
+                nullable=True,
+                comment="按API格式存储的熔断器状态",
+            ),
+        )
+
+    # 数据迁移：如果存在旧字段，迁移数据到新结构
+    if _column_exists("provider_api_keys", "health_score"):
+        op.execute("""
+            UPDATE provider_api_keys
+            SET health_by_format = (
+                SELECT jsonb_object_agg(
+                    elem,
+                    jsonb_build_object(
+                        'health_score', COALESCE(health_score, 1.0),
+                        'consecutive_failures', COALESCE(consecutive_failures, 0),
+                        'last_failure_at', last_failure_at,
+                        'request_results_window', COALESCE(request_results_window::jsonb, '[]'::jsonb)
+                    )
+                )
+                FROM jsonb_array_elements_text(api_formats::jsonb) AS elem
+            )
+            WHERE api_formats IS NOT NULL
+              AND api_formats::text != '[]'
+              AND health_by_format IS NULL
+        """)
+
+    # Circuit Breaker 迁移策略：
+    # 不复制旧的 circuit_breaker_open 状态到所有 format，而是全部重置为 closed
+    # 原因：旧的单一 circuit breaker 状态可能因某一个 format 失败而打开，
+    #       如果复制到所有 format，会导致其他正常工作的 format 被错误标记为不可用
+    if _column_exists("provider_api_keys", "circuit_breaker_open"):
+        op.execute("""
+            UPDATE provider_api_keys
+            SET circuit_breaker_by_format = (
+                SELECT jsonb_object_agg(
+                    elem,
+                    jsonb_build_object(
+                        'open', false,
+                        'open_at', NULL,
+                        'next_probe_at', NULL,
+                        'half_open_until', NULL,
+                        'half_open_successes', 0,
+                        'half_open_failures', 0
+                    )
+                )
+                FROM jsonb_array_elements_text(api_formats::jsonb) AS elem
+            )
+            WHERE api_formats IS NOT NULL
+              AND api_formats::text != '[]'
+              AND circuit_breaker_by_format IS NULL
+        """)
+
+    # 设置默认空对象
+    op.execute("""
+        UPDATE provider_api_keys
+        SET health_by_format = '{}'::jsonb
+        WHERE health_by_format IS NULL
+    """)
+    op.execute("""
+        UPDATE provider_api_keys
+        SET circuit_breaker_by_format = '{}'::jsonb
+        WHERE circuit_breaker_by_format IS NULL
+    """)
+
+    # 创建 GIN 索引
+    if not _index_exists("provider_api_keys", "ix_provider_api_keys_health_by_format"):
+        op.create_index(
+            "ix_provider_api_keys_health_by_format",
+            "provider_api_keys",
+            ["health_by_format"],
+            postgresql_using="gin",
+        )
+    if not _index_exists("provider_api_keys", "ix_provider_api_keys_circuit_breaker_by_format"):
+        op.create_index(
+            "ix_provider_api_keys_circuit_breaker_by_format",
+            "provider_api_keys",
+            ["circuit_breaker_by_format"],
+            postgresql_using="gin",
+        )
+
+    # 删除旧字段
+    old_health_columns = [
+        "health_score",
+        "consecutive_failures",
+        "last_failure_at",
+        "request_results_window",
+        "circuit_breaker_open",
+        "circuit_breaker_open_at",
+        "next_probe_at",
+        "half_open_until",
+        "half_open_successes",
+        "half_open_failures",
+    ]
+    for col in old_health_columns:
+        if _column_exists("provider_api_keys", col):
+            op.drop_column("provider_api_keys", col)
+
+    # ========== 8. provider_endpoints: 删除废弃的 rate_limit 列 ==========
+    if _column_exists("provider_endpoints", "rate_limit"):
+        op.drop_column("provider_endpoints", "rate_limit")
+
+    # ========== 9. usage: 添加 client_response_headers ==========
+    if not _column_exists("usage", "client_response_headers"):
+        op.add_column(
+            "usage",
+            sa.Column("client_response_headers", sa.JSON(), nullable=True),
+        )
+
+    # ========== 10. provider_api_keys: 删除 endpoint_id ==========
+    # Key 不再与 Endpoint 绑定，通过 provider_id + api_formats 关联
+    if _column_exists("provider_api_keys", "endpoint_id"):
+        # 确保外键已删除（前面可能已经删除）
+        try:
+            bind = op.get_bind()
+            inspector = inspect(bind)
+            for fk in inspector.get_foreign_keys("provider_api_keys"):
+                constrained = fk.get("constrained_columns") or []
+                if "endpoint_id" in constrained:
+                    name = fk.get("name")
+                    if name:
+                        op.drop_constraint(name, "provider_api_keys", type_="foreignkey")
+        except Exception:
+            pass  # 外键可能已经不存在
+        op.drop_column("provider_api_keys", "endpoint_id")
+
+    # ========== 11. provider_endpoints: 删除废弃的 max_concurrent 列 ==========
+    if _column_exists("provider_endpoints", "max_concurrent"):
+        op.drop_column("provider_endpoints", "max_concurrent")
+
+    # ========== 12. providers: 删除废弃的 RPM 相关字段 ==========
+    if _column_exists("providers", "rpm_limit"):
+        op.drop_column("providers", "rpm_limit")
+    if _column_exists("providers", "rpm_used"):
+        op.drop_column("providers", "rpm_used")
+    if _column_exists("providers", "rpm_reset_at"):
+        op.drop_column("providers", "rpm_reset_at")
+
+    alembic_logger.info("[OK] Consolidated migration completed successfully")
+
+
+def downgrade() -> None:
+    """
+    Downgrade is complex due to data migrations.
+    For safety, this only removes new columns without restoring old structure.
+    Manual intervention may be required for full rollback.
+    """
+    bind = op.get_bind()
+
+    # 12. 恢复 providers RPM 相关字段
+    if not _column_exists("providers", "rpm_limit"):
+        op.add_column("providers", sa.Column("rpm_limit", sa.Integer(), nullable=True))
+    if not _column_exists("providers", "rpm_used"):
+        op.add_column(
+            "providers",
+            sa.Column("rpm_used", sa.Integer(), server_default="0", nullable=True),
+        )
+    if not _column_exists("providers", "rpm_reset_at"):
+        op.add_column(
+            "providers",
+            sa.Column("rpm_reset_at", sa.DateTime(timezone=True), nullable=True),
+        )
+
+    # 11. 恢复 provider_endpoints.max_concurrent
+    if not _column_exists("provider_endpoints", "max_concurrent"):
+        op.add_column("provider_endpoints", sa.Column("max_concurrent", sa.Integer(), nullable=True))
+
+    # 10. 恢复 endpoint_id
+    if not _column_exists("provider_api_keys", "endpoint_id"):
+        op.add_column("provider_api_keys", sa.Column("endpoint_id", sa.String(36), nullable=True))
+
+    # 9. 删除 client_response_headers
+    if _column_exists("usage", "client_response_headers"):
+        op.drop_column("usage", "client_response_headers")
+
+    # 8. 恢复 provider_endpoints.rate_limit（如果需要）
+    if not _column_exists("provider_endpoints", "rate_limit"):
+        op.add_column("provider_endpoints", sa.Column("rate_limit", sa.Integer(), nullable=True))
+
+    # 7. 删除健康度 JSON 字段
+    bind.execute(sa.text("DROP INDEX IF EXISTS ix_provider_api_keys_health_by_format"))
+    bind.execute(sa.text("DROP INDEX IF EXISTS ix_provider_api_keys_circuit_breaker_by_format"))
+    if _column_exists("provider_api_keys", "health_by_format"):
+        op.drop_column("provider_api_keys", "health_by_format")
+    if _column_exists("provider_api_keys", "circuit_breaker_by_format"):
+        op.drop_column("provider_api_keys", "circuit_breaker_by_format")
+
+    # 6. rpm_limit -> max_concurrent（简化版：仅重命名）
+    if _column_exists("provider_api_keys", "rpm_limit"):
+        op.alter_column("provider_api_keys", "rpm_limit", new_column_name="max_concurrent")
+    if _column_exists("provider_api_keys", "learned_rpm_limit"):
+        op.alter_column("provider_api_keys", "learned_rpm_limit", new_column_name="learned_max_concurrent")
+    if _column_exists("provider_api_keys", "last_rpm_peak"):
+        op.alter_column("provider_api_keys", "last_rpm_peak", new_column_name="last_concurrent_peak")
+
+    # 恢复已删除的字段
+    if not _column_exists("provider_api_keys", "rate_limit"):
+        op.add_column("provider_api_keys", sa.Column("rate_limit", sa.Integer(), nullable=True))
+    if not _column_exists("provider_api_keys", "daily_limit"):
+        op.add_column("provider_api_keys", sa.Column("daily_limit", sa.Integer(), nullable=True))
+    if not _column_exists("provider_api_keys", "monthly_limit"):
+        op.add_column("provider_api_keys", sa.Column("monthly_limit", sa.Integer(), nullable=True))
+
+    # 5. name -> display_name (需要先删除索引)
+    if _index_exists("providers", "ix_providers_name"):
+        op.drop_index("ix_providers_name", table_name="providers")
+    op.alter_column("providers", "name", new_column_name="display_name")
+    # 重新添加原 name 字段
+    op.add_column("providers", sa.Column("name", sa.String(100), nullable=True))
+    op.execute("""
+        UPDATE providers
+        SET name = LOWER(REPLACE(REPLACE(display_name, ' ', '_'), '-', '_'))
+    """)
+    op.alter_column("providers", "name", nullable=False)
+    op.create_index("ix_providers_name", "providers", ["name"], unique=True)
+
+    # 4. 删除 providers 的 timeout, max_retries, proxy
+    if _column_exists("providers", "proxy"):
+        op.drop_column("providers", "proxy")
+    if _column_exists("providers", "max_retries"):
+        op.drop_column("providers", "max_retries")
+    if _column_exists("providers", "timeout"):
+        op.drop_column("providers", "timeout")
+
+    # 3. models: global_model_id 改回 NOT NULL
+    result = bind.execute(sa.text(
+        "SELECT COUNT(*) FROM models WHERE global_model_id IS NULL"
+    ))
+    orphan_model_count = result.scalar() or 0
+    if orphan_model_count > 0:
+        alembic_logger.warning(f"[WARN] 发现 {orphan_model_count} 个无 global_model_id 的独立模型，将被删除")
+        op.execute("DELETE FROM models WHERE global_model_id IS NULL")
+        alembic_logger.info(f"已删除 {orphan_model_count} 个独立模型")
+    op.alter_column("models", "global_model_id", nullable=False)
+
+    # 2. 删除 rate_multipliers
+    if _column_exists("provider_api_keys", "rate_multipliers"):
+        op.drop_column("provider_api_keys", "rate_multipliers")
+
+    # 1. 删除 provider_id 和 api_formats
+    if _index_exists("provider_api_keys", "idx_provider_api_keys_provider_id"):
+        op.drop_index("idx_provider_api_keys_provider_id", table_name="provider_api_keys")
+    if _constraint_exists("provider_api_keys", "fk_provider_api_keys_provider"):
+        op.drop_constraint("fk_provider_api_keys_provider", "provider_api_keys", type_="foreignkey")
+    if _column_exists("provider_api_keys", "api_formats"):
+        op.drop_column("provider_api_keys", "api_formats")
+    if _column_exists("provider_api_keys", "provider_id"):
+        op.drop_column("provider_api_keys", "provider_id")
+
+    # 恢复 endpoint_id 外键（简化版：仅创建外键，不强制 NOT NULL）
+    if _column_exists("provider_api_keys", "endpoint_id"):
+        if not _constraint_exists("provider_api_keys", "provider_api_keys_endpoint_id_fkey"):
+            op.create_foreign_key(
+                "provider_api_keys_endpoint_id_fkey",
+                "provider_api_keys",
+                "provider_endpoints",
+                ["endpoint_id"],
+                ["id"],
+                ondelete="SET NULL",
+            )
+
+    alembic_logger.info("[OK] Downgrade completed (simplified version)")

deploy.sh

@@ -88,9 +88,28 @@ build_base() {
     save_deps_hash
 }
 
+# 生成版本文件
+generate_version_file() {
+    # 从 git 获取版本号
+    local version
+    version=$(git describe --tags --always 2>/dev/null | sed 's/^v//')
+    if [ -z "$version" ]; then
+        version="unknown"
+    fi
+    echo ">>> Generating version file: $version"
+    cat > src/_version.py << EOF
+# Auto-generated by deploy.sh - do not edit
+__version__ = '$version'
+__version_tuple__ = tuple(int(x) for x in '$version'.split('-')[0].split('.') if x.isdigit())
+version = __version__
+version_tuple = __version_tuple__
+EOF
+}
+
 # 构建应用镜像
 build_app() {
     echo ">>> Building app image (code only)..."
+    generate_version_file
     docker build -f Dockerfile.app.local -t aether-app:latest .
     save_code_hash
 }

docker-compose.build.yml

@@ -17,7 +17,7 @@ services:
     ports:
       - "${DB_PORT:-5432}:5432"
     healthcheck:
-      test: ["CMD-SHELL", "pg_isready -U postgres"]
+      test: [ "CMD-SHELL", "pg_isready -U postgres" ]
       interval: 5s
       timeout: 5s
       retries: 5
@@ -32,7 +32,7 @@ services:
     ports:
       - "${REDIS_PORT:-6379}:6379"
     healthcheck:
-      test: ["CMD", "redis-cli", "--raw", "incr", "ping"]
+      test: [ "CMD", "redis-cli", "--raw", "incr", "ping" ]
       interval: 5s
       timeout: 3s
       retries: 5
@@ -44,20 +44,15 @@ services:
       dockerfile: Dockerfile.app.local
     image: aether-app:latest
     container_name: aether-app
+    env_file:
+      - .env
     environment:
+      # 需要组合的变量
       DATABASE_URL: postgresql://postgres:${DB_PASSWORD}@postgres:5432/aether
       REDIS_URL: redis://:${REDIS_PASSWORD}@redis:6379/0
-      PORT: 8084
-      JWT_SECRET_KEY: ${JWT_SECRET_KEY}
-      ENCRYPTION_KEY: ${ENCRYPTION_KEY}
-      JWT_ALGORITHM: HS256
-      JWT_EXPIRATION_DELTA: 86400
-      LOG_LEVEL: ${LOG_LEVEL:-INFO}
-      ADMIN_EMAIL: ${ADMIN_EMAIL}
-      ADMIN_USERNAME: ${ADMIN_USERNAME}
-      ADMIN_PASSWORD: ${ADMIN_PASSWORD}
-      API_KEY_PREFIX: ${API_KEY_PREFIX:-sk}
+      # Supervisor 需要的变量
       GUNICORN_WORKERS: ${GUNICORN_WORKERS:-4}
+      # 容器级别设置
       TZ: Asia/Shanghai
       PYTHONIOENCODING: utf-8
       LANG: C.UTF-8

docker-compose.yml

@@ -13,7 +13,7 @@ services:
     volumes:
       - postgres_data:/var/lib/postgresql/data
     healthcheck:
-      test: ["CMD-SHELL", "pg_isready -U postgres"]
+      test: [ "CMD-SHELL", "pg_isready -U postgres" ]
       interval: 5s
       timeout: 5s
       retries: 5
@@ -26,7 +26,7 @@ services:
     volumes:
       - redis_data:/data
     healthcheck:
-      test: ["CMD", "redis-cli", "--raw", "incr", "ping"]
+      test: [ "CMD", "redis-cli", "--raw", "incr", "ping" ]
       interval: 5s
       timeout: 3s
       retries: 5
@@ -35,20 +35,15 @@ services:
   app:
     image: ghcr.io/fawney19/aether:latest
     container_name: aether-app
+    env_file:
+      - .env
     environment:
+      # 需要组合的变量
       DATABASE_URL: postgresql://postgres:${DB_PASSWORD}@postgres:5432/aether
       REDIS_URL: redis://:${REDIS_PASSWORD}@redis:6379/0
-      PORT: 8084
-      JWT_SECRET_KEY: ${JWT_SECRET_KEY}
-      ENCRYPTION_KEY: ${ENCRYPTION_KEY}
-      JWT_ALGORITHM: HS256
-      JWT_EXPIRATION_DELTA: 86400
-      LOG_LEVEL: ${LOG_LEVEL:-INFO}
-      ADMIN_EMAIL: ${ADMIN_EMAIL}
-      ADMIN_USERNAME: ${ADMIN_USERNAME}
-      ADMIN_PASSWORD: ${ADMIN_PASSWORD}
-      API_KEY_PREFIX: ${API_KEY_PREFIX:-sk}
+      # Supervisor 需要的变量
       GUNICORN_WORKERS: ${GUNICORN_WORKERS:-4}
+      # 容器级别设置
       TZ: Asia/Shanghai
       PYTHONIOENCODING: utf-8
       LANG: C.UTF-8

entrypoint.sh

@@ -0,0 +1,8 @@
+#!/bin/bash
+set -e
+
+echo "Running database migrations..."
+alembic upgrade head
+
+echo "Starting application..."
+exec "$@"

frontend/src/api/admin.ts

@@ -22,7 +22,7 @@ export interface UserExport {
   password_hash: string
   role: string
   allowed_providers?: string[] | null
-  allowed_endpoints?: string[] | null
+  allowed_api_formats?: string[] | null
   allowed_models?: string[] | null
   model_capability_settings?: any
   quota_usd?: number | null
@@ -40,7 +40,6 @@ export interface UserApiKeyExport {
   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 | null  // null = 无限制
@@ -68,7 +67,6 @@ export interface GlobalModelExport {
 
 export interface ProviderExport {
   name: string
-  display_name: string
   description?: string | null
   website?: string | null
   billing_type?: string | null
@@ -77,10 +75,13 @@ export interface ProviderExport {
   rpm_limit?: number | null
   provider_priority?: number
   is_active: boolean
-  rate_limit?: number | null
   concurrent_limit?: number | null
+  timeout?: number | null
+  max_retries?: number | null
+  proxy?: any
   config?: any
   endpoints: EndpointExport[]
+  api_keys: ProviderKeyExport[]
   models: ModelExport[]
 }
 
@@ -90,27 +91,26 @@ export interface EndpointExport {
   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[]
+  proxy?: any
 }
 
-export interface KeyExport {
+export interface ProviderKeyExport {
   api_key: string
   name?: string | null
   note?: string | null
+  api_formats: string[]
   rate_multiplier?: number
+  rate_multipliers?: Record<string, number> | null
   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
+  rpm_limit?: number | null
+  allowed_models?: any
   capabilities?: any
+  cache_ttl_minutes?: number
+  max_probe_interval_minutes?: number
   is_active: boolean
 }
 
@@ -160,6 +160,15 @@ export interface EmailTemplateResetResponse {
   }
 }
 
+// 检查更新响应
+export interface CheckUpdateResponse {
+  current_version: string
+  latest_version: string | null
+  has_update: boolean
+  release_url: string | null
+  error: string | null
+}
+
 // LDAP 配置响应
 export interface LdapConfigResponse {
   server_url: string | null
@@ -527,6 +536,14 @@ export const adminApi = {
     return response.data
   },
 
+  // 检查系统更新
+  async checkUpdate(): Promise<CheckUpdateResponse> {
+    const response = await apiClient.get<CheckUpdateResponse>(
+      '/api/admin/system/check-update'
+    )
+    return response.data
+  },
+
   // LDAP 配置相关
   // 获取 LDAP 配置
   async getLdapConfig(): Promise<LdapConfigResponse> {

frontend/src/api/auth.ts

@@ -98,7 +98,7 @@ export interface User {
   used_usd?: number
   total_usd?: number
   allowed_providers?: string[] | null  // 允许使用的提供商 ID 列表
-  allowed_endpoints?: string[] | null  // 允许使用的端点 ID 列表
+  allowed_api_formats?: string[] | null  // 允许使用的 API 格式列表
   allowed_models?: string[] | null  // 允许使用的模型名称列表
   created_at: string
   last_login_at?: string

frontend/src/api/dashboard.ts

@@ -155,6 +155,7 @@ export interface RequestDetail {
   request_body?: Record<string, any>
   provider_request_headers?: Record<string, any>
   response_headers?: Record<string, any>
+  client_response_headers?: Record<string, any>
   response_body?: Record<string, any>
   metadata?: Record<string, any>
   // 阶梯计费信息

frontend/src/api/endpoints/adaptive.ts

@@ -14,7 +14,7 @@ export async function toggleAdaptiveMode(
   message: string
   key_id: string
   is_adaptive: boolean
-  max_concurrent: number | null
+  rpm_limit: number | null
   effective_limit: number | null
 }> {
   const response = await client.patch(`/api/admin/adaptive/keys/${keyId}/mode`, data)
@@ -22,16 +22,16 @@ export async function toggleAdaptiveMode(
 }
 
 /**
- * 设置 Key 的固定并发限制
+ * 设置 Key 的固定 RPM 限制
  */
-export async function setConcurrentLimit(
+export async function setRpmLimit(
   keyId: string,
   limit: number
 ): Promise<{
   message: string
   key_id: string
   is_adaptive: boolean
-  max_concurrent: number
+  rpm_limit: number
   previous_mode: string
 }> {
   const response = await client.patch(`/api/admin/adaptive/keys/${keyId}/limit`, null, {

frontend/src/api/endpoints/endpoints.ts

@@ -27,15 +27,9 @@ export async function createEndpoint(
     api_format: string
     base_url: string
     custom_path?: string
-    auth_type?: string
-    auth_header?: string
     headers?: Record<string, string>
     timeout?: number
     max_retries?: number
-    priority?: number
-    weight?: number
-    max_concurrent?: number
-    rate_limit?: number
     is_active?: boolean
     config?: Record<string, any>
     proxy?: ProxyConfig | null
@@ -52,16 +46,10 @@ export async function updateEndpoint(
   endpointId: string,
   data: Partial<{
     base_url: string
-    custom_path: string
-    auth_type: string
-    auth_header: string
+    custom_path: string | null
     headers: Record<string, string>
     timeout: number
     max_retries: number
-    priority: number
-    weight: number
-    max_concurrent: number
-    rate_limit: number
     is_active: boolean
     config: Record<string, any>
     proxy: ProxyConfig | null
@@ -74,7 +62,7 @@ export async function updateEndpoint(
 /**
  * 删除 Endpoint
  */
-export async function deleteEndpoint(endpointId: string): Promise<{ message: string; deleted_keys_count: number }> {
+export async function deleteEndpoint(endpointId: string): Promise<{ message: string; affected_keys_count: number }> {
   const response = await client.delete(`/api/admin/endpoints/${endpointId}`)
   return response.data
 }

frontend/src/api/endpoints/health.ts

@@ -32,16 +32,21 @@ export async function getKeyHealth(keyId: string): Promise<HealthStatus> {
 
 /**
  * 恢复Key健康状态（一键恢复：重置健康度 + 关闭熔断器 + 取消自动禁用）
+ * @param keyId Key ID
+ * @param apiFormat 可选，指定 API 格式（如 CLAUDE、OPENAI），不指定则恢复所有格式
  */
-export async function recoverKeyHealth(keyId: string): Promise<{
+export async function recoverKeyHealth(keyId: string, apiFormat?: string): Promise<{
   message: string
   details: {
+    api_format?: string
     health_score: number
     circuit_breaker_open: boolean
     is_active: boolean
   }
 }> {
-  const response = await client.patch(`/api/admin/endpoints/health/keys/${keyId}`)
+  const response = await client.patch(`/api/admin/endpoints/health/keys/${keyId}`, null, {
+    params: apiFormat ? { api_format: apiFormat } : undefined
+  })
   return response.data
 }
 

frontend/src/api/endpoints/keys.ts

@@ -1,5 +1,5 @@
 import client from '../client'
-import type { EndpointAPIKey } from './types'
+import type { EndpointAPIKey, AllowedModels } from './types'
 
 /**
  * 能力定义类型
@@ -49,67 +49,6 @@ export async function getModelCapabilities(modelName: string): Promise<ModelCapa
   return response.data
 }
 
-/**
- * 获取 Endpoint 的所有 Keys
- */
-export async function getEndpointKeys(endpointId: string): Promise<EndpointAPIKey[]> {
-  const response = await client.get(`/api/admin/endpoints/${endpointId}/keys`)
-  return response.data
-}
-
-/**
- * 为 Endpoint 添加 Key
- */
-export async function addEndpointKey(
-  endpointId: string,
-  data: {
-    endpoint_id: string
-    api_key: string
-    name: string  // 密钥名称（必填）
-    rate_multiplier?: number  // 成本倍率（默认 1.0）
-    internal_priority?: number  // Endpoint 内部优先级（数字越小越优先）
-    max_concurrent?: number  // 最大并发数（留空=自适应模式）
-    rate_limit?: number
-    daily_limit?: number
-    monthly_limit?: number
-    cache_ttl_minutes?: number  // 缓存 TTL（分钟），0=禁用
-    max_probe_interval_minutes?: number  // 熔断探测间隔（分钟）
-    allowed_models?: string[]  // 允许使用的模型列表
-    capabilities?: Record<string, boolean>  // 能力标签配置
-    note?: string  // 备注说明（可选）
-  }
-): Promise<EndpointAPIKey> {
-  const response = await client.post(`/api/admin/endpoints/${endpointId}/keys`, data)
-  return response.data
-}
-
-/**
- * 更新 Endpoint Key
- */
-export async function updateEndpointKey(
-  keyId: string,
-  data: Partial<{
-    api_key: string
-    name: string  // 密钥名称
-    rate_multiplier: number  // 成本倍率
-    internal_priority: number  // Endpoint 内部优先级（提供商优先模式，数字越小越优先）
-    global_priority: number  // 全局 Key 优先级（全局 Key 优先模式，数字越小越优先）
-    max_concurrent: number  // 最大并发数（留空=自适应模式）
-    rate_limit: number
-    daily_limit: number
-    monthly_limit: number
-    cache_ttl_minutes: number  // 缓存 TTL（分钟），0=禁用
-    max_probe_interval_minutes: number  // 熔断探测间隔（分钟）
-    allowed_models: string[] | null  // 允许使用的模型列表，null 表示允许所有
-    capabilities: Record<string, boolean> | null  // 能力标签配置
-    is_active: boolean
-    note: string  // 备注说明
-  }>
-): Promise<EndpointAPIKey> {
-  const response = await client.put(`/api/admin/endpoints/keys/${keyId}`, data)
-  return response.data
-}
-
 /**
  * 获取完整的 API Key（用于查看和复制）
  */
@@ -119,22 +58,71 @@ export async function revealEndpointKey(keyId: string): Promise<{ api_key: strin
 }
 
 /**
- * 删除 Endpoint Key
+ * 删除 Key
  */
 export async function deleteEndpointKey(keyId: string): Promise<{ message: string }> {
   const response = await client.delete(`/api/admin/endpoints/keys/${keyId}`)
   return response.data
 }
 
+
+// ========== Provider 级别的 Keys API ==========
+
+
 /**
- * 批量更新 Endpoint Keys 的优先级（用于拖动排序）
+ * 获取 Provider 的所有 Keys
  */
-export async function batchUpdateKeyPriority(
-  endpointId: string,
-  priorities: Array<{ key_id: string; internal_priority: number }>
-): Promise<{ message: string; updated_count: number }> {
-  const response = await client.put(`/api/admin/endpoints/${endpointId}/keys/batch-priority`, {
-    priorities
-  })
+export async function getProviderKeys(providerId: string): Promise<EndpointAPIKey[]> {
+  const response = await client.get(`/api/admin/endpoints/providers/${providerId}/keys`)
+  return response.data
+}
+
+/**
+ * 为 Provider 添加 Key
+ */
+export async function addProviderKey(
+  providerId: string,
+  data: {
+    api_formats: string[]  // 支持的 API 格式列表（必填）
+    api_key: string
+    name: string
+    rate_multiplier?: number  // 默认成本倍率
+    rate_multipliers?: Record<string, number> | null  // 按 API 格式的成本倍率
+    internal_priority?: number
+    rpm_limit?: number | null  // RPM 限制（留空=自适应模式）
+    cache_ttl_minutes?: number
+    max_probe_interval_minutes?: number
+    allowed_models?: AllowedModels
+    capabilities?: Record<string, boolean>
+    note?: string
+  }
+): Promise<EndpointAPIKey> {
+  const response = await client.post(`/api/admin/endpoints/providers/${providerId}/keys`, data)
+  return response.data
+}
+
+/**
+ * 更新 Key
+ */
+export async function updateProviderKey(
+  keyId: string,
+  data: Partial<{
+    api_formats: string[]  // 支持的 API 格式列表
+    api_key: string
+    name: string
+    rate_multiplier: number  // 默认成本倍率
+    rate_multipliers: Record<string, number> | null  // 按 API 格式的成本倍率
+    internal_priority: number
+    global_priority: number | null
+    rpm_limit: number | null  // RPM 限制（留空=自适应模式）
+    cache_ttl_minutes: number
+    max_probe_interval_minutes: number
+    allowed_models: AllowedModels
+    capabilities: Record<string, boolean> | null
+    is_active: boolean
+    note: string
+  }>
+): Promise<EndpointAPIKey> {
+  const response = await client.put(`/api/admin/endpoints/keys/${keyId}`, data)
   return response.data
 }

frontend/src/api/endpoints/models.ts

@@ -147,14 +147,26 @@ export async function queryProviderUpstreamModels(
 
 /**
  * 从上游提供商导入模型
+ * @param providerId 提供商 ID
+ * @param modelIds 模型 ID 列表
+ * @param options 可选配置
+ * @param options.tiered_pricing 阶梯计费配置
+ * @param options.price_per_request 按次计费价格
  */
 export async function importModelsFromUpstream(
   providerId: string,
-  modelIds: string[]
+  modelIds: string[],
+  options?: {
+    tiered_pricing?: object
+    price_per_request?: number
+  }
 ): Promise<ImportFromUpstreamResponse> {
   const response = await client.post(
     `/api/admin/providers/${providerId}/import-from-upstream`,
-    { model_ids: modelIds }
+    {
+      model_ids: modelIds,
+      ...options
+    }
   )
   return response.data
 }

frontend/src/api/endpoints/providers.ts

@@ -1,5 +1,5 @@
 import client from '../client'
-import type { ProviderWithEndpointsSummary } from './types'
+import type { ProviderWithEndpointsSummary, ProxyConfig } from './types'
 
 /**
  * 获取 Providers 摘要（包含 Endpoints 统计）
@@ -23,7 +23,7 @@ export async function getProvider(providerId: string): Promise<ProviderWithEndpo
 export async function updateProvider(
   providerId: string,
   data: Partial<{
-    display_name: string
+    name: string
     description: string
     website: string
     provider_priority: number
@@ -33,6 +33,10 @@ export async function updateProvider(
     quota_last_reset_at: string  // 周期开始时间
     quota_expires_at: string
     rpm_limit: number | null
+    // 请求配置（从 Endpoint 迁移）
+    timeout: number
+    max_retries: number
+    proxy: ProxyConfig | null
     cache_ttl_minutes: number  // 0表示不支持缓存，>0表示支持缓存并设置TTL(分钟)
     max_probe_interval_minutes: number
     is_active: boolean
@@ -83,7 +87,6 @@ export interface TestModelResponse {
   provider?: {
     id: string
     name: string
-    display_name: string
   }
   model?: string
 }
@@ -92,4 +95,3 @@ export async function testModel(data: TestModelRequest): Promise<TestModelRespon
   const response = await client.post('/api/admin/provider-query/test-model', data)
   return response.data
 }
-

frontend/src/api/endpoints/types.ts

@@ -20,6 +20,38 @@ export const API_FORMAT_LABELS: Record<string, string> = {
   [API_FORMATS.GEMINI_CLI]: 'Gemini CLI',
 }
 
+// API 格式缩写映射（用于空间紧凑的显示场景）
+export const API_FORMAT_SHORT: Record<string, string> = {
+  [API_FORMATS.OPENAI]: 'O',
+  [API_FORMATS.OPENAI_CLI]: 'OC',
+  [API_FORMATS.CLAUDE]: 'C',
+  [API_FORMATS.CLAUDE_CLI]: 'CC',
+  [API_FORMATS.GEMINI]: 'G',
+  [API_FORMATS.GEMINI_CLI]: 'GC',
+}
+
+// API 格式排序顺序（统一的显示顺序）
+export const API_FORMAT_ORDER: string[] = [
+  API_FORMATS.OPENAI,
+  API_FORMATS.OPENAI_CLI,
+  API_FORMATS.CLAUDE,
+  API_FORMATS.CLAUDE_CLI,
+  API_FORMATS.GEMINI,
+  API_FORMATS.GEMINI_CLI,
+]
+
+// 工具函数：按标准顺序排序 API 格式数组
+export function sortApiFormats(formats: string[]): string[] {
+  return [...formats].sort((a, b) => {
+    const aIdx = API_FORMAT_ORDER.indexOf(a)
+    const bIdx = API_FORMAT_ORDER.indexOf(b)
+    if (aIdx === -1 && bIdx === -1) return 0
+    if (aIdx === -1) return 1
+    if (bIdx === -1) return -1
+    return aIdx - bIdx
+  })
+}
+
 /**
  * 代理配置类型
  */
@@ -37,18 +69,9 @@ export interface ProviderEndpoint {
   api_format: string
   base_url: string
   custom_path?: string  // 自定义请求路径（可选，为空则使用 API 格式默认路径）
-  auth_type: string
-  auth_header?: string
   headers?: Record<string, string>
   timeout: number
   max_retries: number
-  priority: number
-  weight: number
-  max_concurrent?: number
-  rate_limit?: number
-  health_score: number
-  consecutive_failures: number
-  last_failure_at?: string
   is_active: boolean
   config?: Record<string, any>
   proxy?: ProxyConfig | null
@@ -58,25 +81,55 @@ export interface ProviderEndpoint {
   updated_at: string
 }
 
+/**
+ * 模型权限配置类型（支持简单列表和按格式字典两种模式）
+ *
+ * 使用示例：
+ * 1. 不限制（允许所有模型）: null
+ * 2. 简单列表模式（所有 API 格式共享同一个白名单）: ["gpt-4", "claude-3-opus"]
+ * 3. 按格式字典模式（不同 API 格式使用不同的白名单）:
+ *    { "OPENAI": ["gpt-4"], "CLAUDE": ["claude-3-opus"] }
+ */
+export type AllowedModels = string[] | Record<string, string[]> | null
+
+// AllowedModels 类型守卫函数
+export function isAllowedModelsList(value: AllowedModels): value is string[] {
+  return Array.isArray(value)
+}
+
+export function isAllowedModelsDict(value: AllowedModels): value is Record<string, string[]> {
+  if (value === null || typeof value !== 'object' || Array.isArray(value)) {
+    return false
+  }
+  // 验证所有值都是字符串数组
+  return Object.values(value).every(
+    (v) => Array.isArray(v) && v.every((item) => typeof item === 'string')
+  )
+}
+
 export interface EndpointAPIKey {
   id: string
-  endpoint_id: string
+  provider_id: string
+  api_formats: string[]  // 支持的 API 格式列表
   api_key_masked: string
   api_key_plain?: string | null
   name: string  // 密钥名称（必填，用于识别）
-  rate_multiplier: number  // 成本倍率（真实成本 = 表面成本 × 倍率）
-  internal_priority: number  // Endpoint 内部优先级
+  rate_multiplier: number  // 默认成本倍率（真实成本 = 表面成本 × 倍率）
+  rate_multipliers?: Record<string, number> | null  // 按 API 格式的成本倍率，如 {"CLAUDE": 1.0, "OPENAI": 0.8}
+  internal_priority: number  // Key 内部优先级
   global_priority?: number | null  // 全局 Key 优先级
-  max_concurrent?: number
-  rate_limit?: number
-  daily_limit?: number
-  monthly_limit?: number
-  allowed_models?: string[] | null  // 允许使用的模型列表（null = 支持所有模型）
+  rpm_limit?: number | null  // RPM 速率限制 (1-10000)，null 表示自适应模式
+  allowed_models?: AllowedModels  // 允许使用的模型列表（null=不限制，列表=简单白名单，字典=按格式区分）
   capabilities?: Record<string, boolean> | null  // 能力标签配置（如 cache_1h, context_1m）
   // 缓存与熔断配置
   cache_ttl_minutes: number  // 缓存 TTL（分钟），0=禁用
   max_probe_interval_minutes: number  // 熔断探测间隔（分钟）
+  // 按格式的健康度数据
+  health_by_format?: Record<string, FormatHealthData>
+  circuit_breaker_by_format?: Record<string, FormatCircuitBreakerData>
+  // 聚合字段（从 health_by_format 计算，用于列表显示）
   health_score: number
+  circuit_breaker_open?: boolean
   consecutive_failures: number
   last_failure_at?: string
   request_count: number
@@ -89,10 +142,10 @@ export interface EndpointAPIKey {
   last_used_at?: string
   created_at: string
   updated_at: string
-  // 自适应并发字段
-  is_adaptive?: boolean  // 是否为自适应模式（max_concurrent=NULL）
-  effective_limit?: number  // 当前有效限制（自适应使用学习值，固定使用配置值）
-  learned_max_concurrent?: number
+  // 自适应 RPM 字段
+  is_adaptive?: boolean  // 是否为自适应模式（rpm_limit=NULL）
+  effective_limit?: number  // 当前有效 RPM 限制（自适应使用学习值，固定使用配置值）
+  learned_rpm_limit?: number  // 学习到的 RPM 限制
   // 滑动窗口利用率采样
   utilization_samples?: Array<{ ts: number; util: number }>  // 利用率采样窗口
   last_probe_increase_at?: string  // 上次探测性扩容时间
@@ -100,8 +153,7 @@ export interface EndpointAPIKey {
   rpm_429_count?: number
   last_429_at?: string
   last_429_type?: string
-  // 熔断器字段（滑动窗口 + 半开模式）
-  circuit_breaker_open?: boolean
+  // 单格式场景的熔断器字段
   circuit_breaker_open_at?: string
   next_probe_at?: string
   half_open_until?: string
@@ -110,17 +162,36 @@ export interface EndpointAPIKey {
   request_results_window?: Array<{ ts: number; ok: boolean }>  // 请求结果滑动窗口
 }
 
+// 按格式的健康度数据
+export interface FormatHealthData {
+  health_score: number
+  error_rate: number
+  window_size: number
+  consecutive_failures: number
+  last_failure_at?: string | null
+  circuit_breaker: FormatCircuitBreakerData
+}
+
+// 按格式的熔断器数据
+export interface FormatCircuitBreakerData {
+  open: boolean
+  open_at?: string | null
+  next_probe_at?: string | null
+  half_open_until?: string | null
+  half_open_successes: number
+  half_open_failures: number
+}
+
 export interface EndpointAPIKeyUpdate {
+  api_formats?: string[]  // 支持的 API 格式列表
   name?: string
   api_key?: string  // 仅在需要更新时提供
-  rate_multiplier?: number
+  rate_multiplier?: number  // 默认成本倍率
+  rate_multipliers?: Record<string, number> | null  // 按 API 格式的成本倍率
   internal_priority?: number
   global_priority?: number | null
-  max_concurrent?: number | null  // null 表示切换为自适应模式
-  rate_limit?: number
-  daily_limit?: number
-  monthly_limit?: number
-  allowed_models?: string[] | null
+  rpm_limit?: number | null  // RPM 速率限制 (1-10000)，null 表示切换为自适应模式
+  allowed_models?: AllowedModels
   capabilities?: Record<string, boolean> | null
   cache_ttl_minutes?: number
   max_probe_interval_minutes?: number
@@ -198,7 +269,6 @@ export interface PublicEndpointStatusMonitorResponse {
 export interface ProviderWithEndpointsSummary {
   id: string
   name: string
-  display_name: string
   description?: string
   website?: string
   provider_priority: number
@@ -208,9 +278,10 @@ export interface ProviderWithEndpointsSummary {
   quota_reset_day?: number
   quota_last_reset_at?: string  // 当前周期开始时间
   quota_expires_at?: string
-  rpm_limit?: number | null
-  rpm_used?: number
-  rpm_reset_at?: string
+  // 请求配置（从 Endpoint 迁移）
+  timeout?: number  // 请求超时（秒）
+  max_retries?: number  // 最大重试次数
+  proxy?: ProxyConfig | null  // 代理配置
   is_active: boolean
   total_endpoints: number
   active_endpoints: number
@@ -253,13 +324,10 @@ export interface HealthSummary {
   }
 }
 
-export interface ConcurrencyStatus {
-  endpoint_id?: string
-  endpoint_current_concurrency: number
-  endpoint_max_concurrent?: number
-  key_id?: string
-  key_current_concurrency: number
-  key_max_concurrent?: number
+export interface KeyRpmStatus {
+  key_id: string
+  current_rpm: number
+  rpm_limit?: number
 }
 
 export interface ProviderModelMapping {
@@ -361,7 +429,6 @@ export interface ModelPriceRange {
 export interface ModelCatalogProviderDetail {
   provider_id: string
   provider_name: string
-  provider_display_name?: string | null
   model_id?: string | null
   target_model: string
   input_price_per_1m?: number | null
@@ -534,10 +601,10 @@ export interface UpstreamModel {
  */
 export interface ImportFromUpstreamSuccessItem {
   model_id: string
-  global_model_id: string
-  global_model_name: string
   provider_model_id: string
-  created_global_model: boolean
+  global_model_id?: string  // 可选，未关联时为空字符串
+  global_model_name?: string  // 可选，未关联时为空字符串
+  created_global_model: boolean  // 始终为 false（不再自动创建 GlobalModel）
 }
 
 /**

frontend/src/api/management-tokens.ts

@@ -0,0 +1,203 @@
+/**
+ * Management Token API
+ */
+
+import apiClient from './client'
+
+// ============== 类型定义 ==============
+
+export interface ManagementToken {
+  id: string
+  user_id: string
+  name: string
+  description?: string
+  token_display: string
+  allowed_ips?: string[] | null
+  expires_at?: string | null
+  last_used_at?: string | null
+  last_used_ip?: string | null
+  usage_count: number
+  is_active: boolean
+  created_at: string
+  updated_at: string
+  user?: {
+    id: string
+    email: string
+    username: string
+    role: string
+  }
+}
+
+export interface CreateManagementTokenRequest {
+  name: string
+  description?: string
+  allowed_ips?: string[]
+  expires_at?: string | null
+}
+
+export interface CreateManagementTokenResponse {
+  message: string
+  token: string
+  data: ManagementToken
+}
+
+export interface UpdateManagementTokenRequest {
+  name?: string
+  description?: string | null
+  allowed_ips?: string[] | null
+  expires_at?: string | null
+  is_active?: boolean
+}
+
+export interface ManagementTokenListResponse {
+  items: ManagementToken[]
+  total: number
+  skip: number
+  limit: number
+  quota?: {
+    used: number
+    max: number
+  }
+}
+
+// ============== 用户自助管理 API ==============
+
+export const managementTokenApi = {
+  /**
+   * 列出当前用户的 Management Tokens
+   */
+  async listTokens(params?: {
+    is_active?: boolean
+    skip?: number
+    limit?: number
+  }): Promise<ManagementTokenListResponse> {
+    const response = await apiClient.get<ManagementTokenListResponse>(
+      '/api/me/management-tokens',
+      { params }
+    )
+    return response.data
+  },
+
+  /**
+   * 创建 Management Token
+   */
+  async createToken(
+    data: CreateManagementTokenRequest
+  ): Promise<CreateManagementTokenResponse> {
+    const response = await apiClient.post<CreateManagementTokenResponse>(
+      '/api/me/management-tokens',
+      data
+    )
+    return response.data
+  },
+
+  /**
+   * 获取 Token 详情
+   */
+  async getToken(tokenId: string): Promise<ManagementToken> {
+    const response = await apiClient.get<ManagementToken>(
+      `/api/me/management-tokens/${tokenId}`
+    )
+    return response.data
+  },
+
+  /**
+   * 更新 Token
+   */
+  async updateToken(
+    tokenId: string,
+    data: UpdateManagementTokenRequest
+  ): Promise<{ message: string; data: ManagementToken }> {
+    const response = await apiClient.put<{ message: string; data: ManagementToken }>(
+      `/api/me/management-tokens/${tokenId}`,
+      data
+    )
+    return response.data
+  },
+
+  /**
+   * 删除 Token
+   */
+  async deleteToken(tokenId: string): Promise<{ message: string }> {
+    const response = await apiClient.delete<{ message: string }>(
+      `/api/me/management-tokens/${tokenId}`
+    )
+    return response.data
+  },
+
+  /**
+   * 切换 Token 状态
+   */
+  async toggleToken(
+    tokenId: string
+  ): Promise<{ message: string; data: ManagementToken }> {
+    const response = await apiClient.patch<{ message: string; data: ManagementToken }>(
+      `/api/me/management-tokens/${tokenId}/status`
+    )
+    return response.data
+  },
+
+  /**
+   * 重新生成 Token
+   */
+  async regenerateToken(
+    tokenId: string
+  ): Promise<{ token: string; data: ManagementToken }> {
+    const response = await apiClient.post<{ token: string; data: ManagementToken }>(
+      `/api/me/management-tokens/${tokenId}/regenerate`
+    )
+    return response.data
+  }
+}
+
+// ============== 管理员 API ==============
+
+export const adminManagementTokenApi = {
+  /**
+   * 列出所有 Management Tokens（管理员）
+   */
+  async listAllTokens(params?: {
+    user_id?: string
+    is_active?: boolean
+    skip?: number
+    limit?: number
+  }): Promise<ManagementTokenListResponse> {
+    const response = await apiClient.get<ManagementTokenListResponse>(
+      '/api/admin/management-tokens',
+      { params }
+    )
+    return response.data
+  },
+
+  /**
+   * 获取 Token 详情（管理员）
+   */
+  async getToken(tokenId: string): Promise<ManagementToken> {
+    const response = await apiClient.get<ManagementToken>(
+      `/api/admin/management-tokens/${tokenId}`
+    )
+    return response.data
+  },
+
+  /**
+   * 删除任意 Token（管理员）
+   */
+  async deleteToken(tokenId: string): Promise<{ message: string }> {
+    const response = await apiClient.delete<{ message: string }>(
+      `/api/admin/management-tokens/${tokenId}`
+    )
+    return response.data
+  },
+
+  /**
+   * 切换任意 Token 状态（管理员）
+   */
+  async toggleToken(
+    tokenId: string
+  ): Promise<{ message: string; data: ManagementToken }> {
+    const response = await apiClient.patch<{ message: string; data: ManagementToken }>(
+      `/api/admin/management-tokens/${tokenId}/status`
+    )
+    return response.data
+  }
+}

frontend/src/api/users.ts

@@ -10,7 +10,7 @@ export interface User {
   used_usd: number
   total_usd: number
   allowed_providers: string[] | null  // 允许使用的提供商 ID 列表
-  allowed_endpoints: string[] | null  // 允许使用的端点 ID 列表
+  allowed_api_formats: string[] | null  // 允许使用的 API 格式列表
   allowed_models: string[] | null  // 允许使用的模型名称列表
   created_at: string
   updated_at?: string
@@ -23,7 +23,7 @@ export interface CreateUserRequest {
   role?: 'admin' | 'user'
   quota_usd?: number | null
   allowed_providers?: string[] | null
-  allowed_endpoints?: string[] | null
+  allowed_api_formats?: string[] | null
   allowed_models?: string[] | null
 }
 
@@ -34,7 +34,7 @@ export interface UpdateUserRequest {
   quota_usd?: number | null
   password?: string
   allowed_providers?: string[] | null
-  allowed_endpoints?: string[] | null
+  allowed_api_formats?: string[] | null
   allowed_models?: string[] | null
 }
 

frontend/src/components/common/UpdateDialog.vue

@@ -0,0 +1,112 @@
+<template>
+  <Dialog
+    v-model="isOpen"
+    size="md"
+    title=""
+  >
+    <div class="flex flex-col items-center text-center py-2">
+      <!-- Logo -->
+      <HeaderLogo
+        size="h-16 w-16"
+        class-name="text-primary"
+      />
+
+      <!-- Title -->
+      <h2 class="text-xl font-semibold text-foreground mt-4 mb-2">
+        发现新版本
+      </h2>
+
+      <!-- Version Info -->
+      <div class="flex items-center gap-3 mb-4">
+        <span class="px-3 py-1.5 rounded-lg bg-muted text-sm font-mono text-muted-foreground">
+          v{{ currentVersion }}
+        </span>
+        <svg
+          class="h-4 w-4 text-muted-foreground"
+          fill="none"
+          stroke="currentColor"
+          viewBox="0 0 24 24"
+        >
+          <path
+            stroke-linecap="round"
+            stroke-linejoin="round"
+            stroke-width="2"
+            d="M13 7l5 5m0 0l-5 5m5-5H6"
+          />
+        </svg>
+        <span class="px-3 py-1.5 rounded-lg bg-primary/10 text-sm font-mono font-medium text-primary">
+          v{{ latestVersion }}
+        </span>
+      </div>
+
+      <!-- Description -->
+      <p class="text-sm text-muted-foreground max-w-xs">
+        新版本已发布，建议更新以获得最新功能和安全修复
+      </p>
+    </div>
+
+    <template #footer>
+      <div class="flex w-full gap-3">
+        <Button
+          variant="outline"
+          class="flex-1"
+          @click="handleLater"
+        >
+          稍后提醒
+        </Button>
+        <Button
+          class="flex-1"
+          @click="handleViewRelease"
+        >
+          查看更新
+        </Button>
+      </div>
+    </template>
+  </Dialog>
+</template>
+
+<script setup lang="ts">
+import { ref, watch } from 'vue'
+import { Dialog } from '@/components/ui'
+import Button from '@/components/ui/button.vue'
+import HeaderLogo from '@/components/HeaderLogo.vue'
+
+const props = defineProps<{
+  modelValue: boolean
+  currentVersion: string
+  latestVersion: string
+  releaseUrl: string | null
+}>()
+
+const emit = defineEmits<{
+  'update:modelValue': [value: boolean]
+}>()
+
+const isOpen = ref(props.modelValue)
+
+watch(() => props.modelValue, (val) => {
+  isOpen.value = val
+})
+
+watch(isOpen, (val) => {
+  emit('update:modelValue', val)
+})
+
+function handleLater() {
+  // 记录忽略的版本，24小时内不再提醒
+  const ignoreKey = 'aether_update_ignore'
+  const ignoreData = {
+    version: props.latestVersion,
+    until: Date.now() + 24 * 60 * 60 * 1000 // 24小时
+  }
+  localStorage.setItem(ignoreKey, JSON.stringify(ignoreData))
+  isOpen.value = false
+}
+
+function handleViewRelease() {
+  if (props.releaseUrl) {
+    window.open(props.releaseUrl, '_blank')
+  }
+  isOpen.value = false
+}
+</script>

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

@@ -0,0 +1,15 @@
+<script setup lang="ts">
+import { CollapsibleContent, type CollapsibleContentProps } from 'radix-vue'
+import { cn } from '@/lib/utils'
+
+const props = defineProps<CollapsibleContentProps & { class?: string }>()
+</script>
+
+<template>
+  <CollapsibleContent
+    v-bind="props"
+    :class="cn('overflow-hidden data-[state=closed]:animate-collapsible-up data-[state=open]:animate-collapsible-down', props.class)"
+  >
+    <slot />
+  </CollapsibleContent>
+</template>

frontend/src/components/ui/collapsible-trigger.vue

@@ -0,0 +1,11 @@
+<script setup lang="ts">
+import { CollapsibleTrigger, type CollapsibleTriggerProps } from 'radix-vue'
+
+const props = defineProps<CollapsibleTriggerProps>()
+</script>
+
+<template>
+  <CollapsibleTrigger v-bind="props" as-child>
+    <slot />
+  </CollapsibleTrigger>
+</template>

frontend/src/components/ui/collapsible.vue

@@ -0,0 +1,15 @@
+<script setup lang="ts">
+import { CollapsibleRoot, type CollapsibleRootEmits, type CollapsibleRootProps } from 'radix-vue'
+import { useForwardPropsEmits } from 'radix-vue'
+
+const props = defineProps<CollapsibleRootProps>()
+const emits = defineEmits<CollapsibleRootEmits>()
+
+const forwarded = useForwardPropsEmits(props, emits)
+</script>
+
+<template>
+  <CollapsibleRoot v-bind="forwarded">
+    <slot />
+  </CollapsibleRoot>
+</template>

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

@@ -18,11 +18,11 @@
           v-if="isOpen"
           class="fixed inset-0 bg-black/40 backdrop-blur-sm transition-opacity pointer-events-auto"
           :style="{ zIndex: backdropZIndex }"
-          @click="handleClose"
+          @click="handleBackdropClick"
         />
       </Transition>
 
-      <div class="relative flex min-h-full items-end justify-center p-4 text-center sm:items-center sm:p-0 pointer-events-none">
+      <div class="relative flex min-h-full items-end justify-center px-3 py-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 pointer-events-auto"
+            class="relative transform rounded-lg bg-background text-left shadow-2xl transition-all w-full sm:my-8 sm:w-full border border-border pointer-events-auto"
             :style="{ zIndex: contentZIndex }"
             :class="maxWidthClass"
             @click.stop
@@ -106,6 +106,7 @@ const props = defineProps<{
   iconClass?: string // Custom icon color class
   zIndex?: number // Custom z-index for nested dialogs (default: 60)
   noPadding?: boolean // Disable default content padding
+  persistent?: boolean // Prevent closing on backdrop click
 }>()
 
 // Emits 定义
@@ -138,6 +139,13 @@ function handleClose() {
   }
 }
 
+// 处理背景点击
+function handleBackdropClick() {
+  if (!props.persistent) {
+    handleClose()
+  }
+}
+
 const maxWidthClass = computed(() => {
   const sizeValue = props.maxWidth || props.size || 'md'
   const sizes = {
@@ -162,7 +170,7 @@ const contentZIndex = computed(() => (props.zIndex || 60) + 10)
 
 // 添加 ESC 键监听
 useEscapeKey(() => {
-  if (isOpen.value) {
+  if (isOpen.value && !props.persistent) {
     handleClose()
     return true  // 阻止其他监听器（如父级抽屉的 ESC 监听器）
   }

frontend/src/components/ui/index.ts

@@ -65,3 +65,8 @@ export { default as RefreshButton } from './refresh-button.vue'
 
 // Tooltip 提示系列
 export { Tooltip, TooltipContent, TooltipProvider, TooltipTrigger } from './tooltip'
+
+// Collapsible 折叠系列
+export { default as Collapsible } from './collapsible.vue'
+export { default as CollapsibleTrigger } from './collapsible-trigger.vue'
+export { default as CollapsibleContent } from './collapsible-content.vue'

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

@@ -186,7 +186,7 @@
                     @click.stop
                     @change="toggleSelection('allowed_providers', provider.id)"
                   >
-                  <span class="text-sm">{{ provider.display_name || provider.name }}</span>
+                  <span class="text-sm">{{ provider.name }}</span>
                 </div>
                 <div
                   v-if="providers.length === 0"

frontend/src/features/auth/components/LoginDialog.vue

@@ -73,14 +73,16 @@
       >
         <button
           type="button"
-          :class="['auth-tab', authType === 'local' && 'active']"
+          class="auth-tab"
+          :class="[authType === 'local' && 'active']"
           @click="authType = 'local'"
         >
           本地登录
         </button>
         <button
           type="button"
-          :class="['auth-tab', authType === 'ldap' && 'active']"
+          class="auth-tab"
+          :class="[authType === 'ldap' && 'active']"
           @click="authType = 'ldap'"
         >
           LDAP 登录

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

@@ -460,13 +460,13 @@
                         <TableHead class="h-10 font-semibold">
                           Provider
                         </TableHead>
-                        <TableHead class="w-[120px] h-10 font-semibold">
+                        <TableHead class="w-[100px] h-10 font-semibold">
                           能力
                         </TableHead>
-                        <TableHead class="w-[180px] h-10 font-semibold">
+                        <TableHead class="w-[200px] h-10 font-semibold">
                           价格 ($/M)
                         </TableHead>
-                        <TableHead class="w-[80px] h-10 font-semibold text-center">
+                        <TableHead class="w-[100px] h-10 font-semibold text-center">
                           操作
                         </TableHead>
                       </TableRow>
@@ -484,7 +484,7 @@
                               :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>
+                            <span class="font-medium truncate">{{ provider.name }}</span>
                           </div>
                         </TableCell>
                         <TableCell class="py-3">
@@ -595,7 +595,7 @@
                             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>
+                          <span class="font-medium truncate">{{ provider.name }}</span>
                         </div>
                         <div class="flex items-center gap-1 shrink-0">
                           <Button

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

@@ -1,245 +1,200 @@
 <template>
   <Dialog
     :model-value="internalOpen"
-    :title="isEditMode ? '编辑 API 端点' : '添加 API 端点'"
-    :description="isEditMode ? `修改 ${provider?.display_name} 的端点配置` : '为提供商添加新的 API 端点'"
-    :icon="isEditMode ? SquarePen : Link"
-    size="xl"
+    title="端点管理"
+    :description="`管理 ${provider?.name} 的 API 端点`"
+    :icon="Settings"
+    size="2xl"
     @update:model-value="handleDialogUpdate"
   >
-    <form
-      class="space-y-6"
-      @submit.prevent="handleSubmit()"
-    >
-      <!-- API 配置 -->
-      <div class="space-y-4">
-        <h3
-          v-if="isEditMode"
-          class="text-sm font-medium"
-        >
-          API 配置
-        </h3>
-
-        <div class="grid grid-cols-2 gap-4">
-          <!-- API 格式 -->
-          <div class="space-y-2">
-            <Label for="api_format">API 格式 *</Label>
-            <template v-if="isEditMode">
-              <Input
-                id="api_format"
-                v-model="form.api_format"
-                disabled
-                class="bg-muted"
-              />
-              <p class="text-xs text-muted-foreground">
-                API 格式创建后不可修改
-              </p>
-            </template>
-            <template v-else>
-              <Select
-                v-model="form.api_format"
-                v-model:open="selectOpen"
-                required
-              >
-                <SelectTrigger>
-                  <SelectValue placeholder="请选择 API 格式" />
-                </SelectTrigger>
-                <SelectContent>
-                  <SelectItem
-                    v-for="format in apiFormats"
-                    :key="format.value"
-                    :value="format.value"
-                  >
-                    {{ format.label }}
-                  </SelectItem>
-                </SelectContent>
-              </Select>
-            </template>
-          </div>
-
-          <!-- API URL -->
-          <div class="space-y-2">
-            <Label for="base_url">API URL *</Label>
-            <Input
-              id="base_url"
-              v-model="form.base_url"
-              placeholder="https://api.example.com"
-              required
-            />
-          </div>
-        </div>
-
-        <!-- 自定义路径 -->
+    <div class="space-y-4">
+      <!-- 已有端点列表 -->
+      <div
+        v-if="localEndpoints.length > 0"
+        class="space-y-2"
+      >
+        <Label class="text-muted-foreground">已配置的端点</Label>
         <div class="space-y-2">
-          <Label for="custom_path">自定义请求路径（可选）</Label>
-          <Input
-            id="custom_path"
-            v-model="form.custom_path"
-            :placeholder="defaultPathPlaceholder"
-          />
-        </div>
-      </div>
-
-      <!-- 请求配置 -->
-      <div class="space-y-4">
-        <h3 class="text-sm font-medium">
-          请求配置
-        </h3>
-
-        <div class="grid grid-cols-3 gap-4">
-          <div class="space-y-2">
-            <Label for="timeout">超时（秒）</Label>
-            <Input
-              id="timeout"
-              v-model.number="form.timeout"
-              type="number"
-              placeholder="300"
-            />
-          </div>
-
-          <div class="space-y-2">
-            <Label for="max_retries">最大重试</Label>
-            <Input
-              id="max_retries"
-              v-model.number="form.max_retries"
-              type="number"
-              placeholder="3"
-            />
-          </div>
-
-          <div class="space-y-2">
-            <Label for="max_concurrent">最大并发</Label>
-            <Input
-              id="max_concurrent"
-              :model-value="form.max_concurrent ?? ''"
-              type="number"
-              placeholder="无限制"
-              @update:model-value="(v) => form.max_concurrent = parseNumberInput(v)"
-            />
-          </div>
-        </div>
-
-        <div class="grid grid-cols-2 gap-4">
-          <div class="space-y-2">
-            <Label for="rate_limit">速率限制（请求/分钟）</Label>
-            <Input
-              id="rate_limit"
-              :model-value="form.rate_limit ?? ''"
-              type="number"
-              placeholder="无限制"
-              @update:model-value="(v) => form.rate_limit = parseNumberInput(v)"
-            />
+          <div
+            v-for="endpoint in localEndpoints"
+            :key="endpoint.id"
+            class="rounded-md border px-3 py-2"
+            :class="{ 'opacity-50': !endpoint.is_active }"
+          >
+            <!-- 编辑模式 -->
+            <template v-if="editingEndpointId === endpoint.id">
+              <div class="space-y-2">
+                <div class="flex items-center gap-2">
+                  <span class="text-sm font-medium w-24 shrink-0">{{ API_FORMAT_LABELS[endpoint.api_format] || endpoint.api_format }}</span>
+                  <div class="flex items-center gap-1 ml-auto">
+                    <Button
+                      variant="ghost"
+                      size="icon"
+                      class="h-7 w-7"
+                      title="保存"
+                      :disabled="savingEndpointId === endpoint.id"
+                      @click="saveEndpointUrl(endpoint)"
+                    >
+                      <Check class="w-3.5 h-3.5" />
+                    </Button>
+                    <Button
+                      variant="ghost"
+                      size="icon"
+                      class="h-7 w-7"
+                      title="取消"
+                      @click="cancelEdit"
+                    >
+                      <X class="w-3.5 h-3.5" />
+                    </Button>
+                  </div>
+                </div>
+                <div class="grid grid-cols-2 gap-2">
+                  <div class="space-y-1">
+                    <Label class="text-xs text-muted-foreground">Base URL</Label>
+                    <Input
+                      v-model="editingUrl"
+                      class="h-8 text-sm"
+                      placeholder="https://api.example.com"
+                      @keyup.escape="cancelEdit"
+                    />
+                  </div>
+                  <div class="space-y-1">
+                    <Label class="text-xs text-muted-foreground">自定义路径 (可选)</Label>
+                    <Input
+                      v-model="editingPath"
+                      class="h-8 text-sm"
+                      :placeholder="editingDefaultPath || '留空使用默认路径'"
+                      @keyup.escape="cancelEdit"
+                    />
+                  </div>
+                </div>
+              </div>
+            </template>
+            <!-- 查看模式 -->
+            <template v-else>
+              <div class="flex items-center gap-3">
+                <div class="w-24 shrink-0">
+                  <span class="text-sm font-medium">{{ API_FORMAT_LABELS[endpoint.api_format] || endpoint.api_format }}</span>
+                </div>
+                <div class="flex-1 min-w-0">
+                  <span class="text-sm text-muted-foreground truncate block">
+                    {{ endpoint.base_url }}{{ endpoint.custom_path ? endpoint.custom_path : '' }}
+                  </span>
+                </div>
+                <div class="flex items-center gap-1 shrink-0">
+                  <Button
+                    variant="ghost"
+                    size="icon"
+                    class="h-7 w-7"
+                    title="编辑"
+                    @click="startEdit(endpoint)"
+                  >
+                    <Edit class="w-3.5 h-3.5" />
+                  </Button>
+                  <Button
+                    variant="ghost"
+                    size="icon"
+                    class="h-7 w-7"
+                    :title="endpoint.is_active ? '停用' : '启用'"
+                    :disabled="togglingEndpointId === endpoint.id"
+                    @click="handleToggleEndpoint(endpoint)"
+                  >
+                    <Power class="w-3.5 h-3.5" />
+                  </Button>
+                  <Button
+                    variant="ghost"
+                    size="icon"
+                    class="h-7 w-7 text-destructive hover:text-destructive"
+                    title="删除"
+                    :disabled="deletingEndpointId === endpoint.id"
+                    @click="handleDeleteEndpoint(endpoint)"
+                  >
+                    <Trash2 class="w-3.5 h-3.5" />
+                  </Button>
+                </div>
+              </div>
+            </template>
           </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"
+      <!-- 添加新端点 -->
+      <div
+        v-if="availableFormats.length > 0"
+        class="space-y-3 pt-3 border-t"
+      >
+        <Label class="text-muted-foreground">添加新端点</Label>
+        <div class="flex items-end gap-3">
+          <div class="w-32 shrink-0 space-y-1.5">
+            <Label class="text-xs">API 格式</Label>
+            <Select
+              v-model="newEndpoint.api_format"
+              v-model:open="formatSelectOpen"
             >
-              {{ proxyUrlError }}
-            </p>
-            <p
-              v-else
-              class="text-xs text-muted-foreground"
-            >
-              支持 HTTP、HTTPS、SOCKS5 代理
-            </p>
+              <SelectTrigger class="h-9">
+                <SelectValue placeholder="选择格式" />
+              </SelectTrigger>
+              <SelectContent>
+                <SelectItem
+                  v-for="format in availableFormats"
+                  :key="format.value"
+                  :value="format.value"
+                >
+                  {{ format.label }}
+                </SelectItem>
+              </SelectContent>
+            </Select>
           </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_${formId}`"
-                v-model="form.proxy_username"
-                :name="`proxy_user_${formId}`"
-                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}`"
-                v-model="form.proxy_password"
-                :name="`proxy_pass_${formId}`"
-                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 class="flex-1 space-y-1.5">
+            <Label class="text-xs">Base URL</Label>
+            <Input
+              v-model="newEndpoint.base_url"
+              placeholder="https://api.example.com"
+              class="h-9"
+            />
           </div>
+          <div class="w-40 shrink-0 space-y-1.5">
+            <Label class="text-xs">自定义路径</Label>
+            <Input
+              v-model="newEndpoint.custom_path"
+              :placeholder="newEndpointDefaultPath || '可选'"
+              class="h-9"
+            />
+          </div>
+          <Button
+            variant="outline"
+            size="sm"
+            class="h-9 shrink-0"
+            :disabled="!newEndpoint.api_format || !newEndpoint.base_url || addingEndpoint"
+            @click="handleAddEndpoint"
+          >
+            {{ addingEndpoint ? '添加中...' : '添加' }}
+          </Button>
         </div>
       </div>
-    </form>
+
+      <!-- 空状态 -->
+      <div
+        v-if="localEndpoints.length === 0 && availableFormats.length === 0"
+        class="text-center py-8 text-muted-foreground"
+      >
+        <p>所有 API 格式都已配置</p>
+      </div>
+    </div>
 
     <template #footer>
       <Button
-        type="button"
         variant="outline"
-        :disabled="loading"
-        @click="handleCancel"
+        @click="handleClose"
       >
-        取消
-      </Button>
-      <Button
-        :disabled="loading || !form.base_url || (!isEditMode && !form.api_format)"
-        @click="handleSubmit()"
-      >
-        {{ loading ? (isEditMode ? '保存中...' : '创建中...') : (isEditMode ? '保存修改' : '创建') }}
+        关闭
       </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">
-import { ref, computed, onMounted } from 'vue'
+import { ref, computed, onMounted, watch } from 'vue'
 import {
   Dialog,
   Button,
@@ -250,17 +205,15 @@ import {
   SelectValue,
   SelectContent,
   SelectItem,
-  Switch,
 } from '@/components/ui'
-import AlertDialog from '@/components/common/AlertDialog.vue'
-import { Link, SquarePen } from 'lucide-vue-next'
+import { Settings, Edit, Trash2, Check, X, Power } from 'lucide-vue-next'
 import { useToast } from '@/composables/useToast'
-import { useFormDialog } from '@/composables/useFormDialog'
-import { parseNumberInput } from '@/utils/form'
 import { log } from '@/utils/logger'
 import {
   createEndpoint,
   updateEndpoint,
+  deleteEndpoint,
+  API_FORMAT_LABELS,
   type ProviderEndpoint,
   type ProviderWithEndpointsSummary
 } from '@/api/endpoints'
@@ -269,7 +222,7 @@ import { adminApi } from '@/api/admin'
 const props = defineProps<{
   modelValue: boolean
   provider: ProviderWithEndpointsSummary | null
-  endpoint?: ProviderEndpoint | null  // 编辑模式时传入
+  endpoints?: ProviderEndpoint[]
 }>()
 
 const emit = defineEmits<{
@@ -279,258 +232,184 @@ 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)
+// 状态
+const addingEndpoint = ref(false)
+const editingEndpointId = ref<string | null>(null)
+const editingUrl = ref('')
+const editingPath = ref('')
+const savingEndpointId = ref<string | null>(null)
+const deletingEndpointId = ref<string | null>(null)
+const togglingEndpointId = ref<string | null>(null)
+const formatSelectOpen = ref(false)
 
 // 内部状态
 const internalOpen = computed(() => props.modelValue)
 
-// 表单数据
-const form = ref({
+// 新端点表单
+const newEndpoint = ref({
   api_format: '',
   base_url: '',
   custom_path: '',
-  timeout: 300,
-  max_retries: 3,
-  max_concurrent: undefined as number | undefined,
-  rate_limit: undefined as number | undefined,
-  is_active: true,
-  // 代理配置
-  proxy_url: '',
-  proxy_username: '',
-  proxy_password: '',
 })
 
 // API 格式列表
-const apiFormats = ref<Array<{ value: string; label: string; default_path: string; aliases: string[] }>>([])
+const apiFormats = ref<Array<{ value: string; label: string; default_path: string }>>([])
 
-// 加载API格式列表
+// 本地端点列表
+const localEndpoints = ref<ProviderEndpoint[]>([])
+
+// 可用的格式（未添加的）
+const availableFormats = computed(() => {
+  const existingFormats = localEndpoints.value.map(e => e.api_format)
+  return apiFormats.value.filter(f => !existingFormats.includes(f.value))
+})
+
+// 获取指定 API 格式的默认路径
+function getDefaultPath(apiFormat: string): string {
+  const format = apiFormats.value.find(f => f.value === apiFormat)
+  return format?.default_path || ''
+}
+
+// 当前编辑端点的默认路径
+const editingDefaultPath = computed(() => {
+  const endpoint = localEndpoints.value.find(e => e.id === editingEndpointId.value)
+  return endpoint ? getDefaultPath(endpoint.api_format) : ''
+})
+
+// 新端点选择的格式的默认路径
+const newEndpointDefaultPath = computed(() => {
+  return getDefaultPath(newEndpoint.value.api_format)
+})
+
+// 加载 API 格式列表
 const loadApiFormats = async () => {
   try {
     const response = await adminApi.getApiFormats()
     apiFormats.value = response.formats
   } catch (error) {
     log.error('加载API格式失败:', error)
-    if (!isEditMode.value) {
-      showError('加载API格式失败', '错误')
-    }
   }
 }
 
-// 根据选择的 API 格式计算默认路径
-const defaultPath = computed(() => {
-  const format = apiFormats.value.find(f => f.value === form.value.api_format)
-  return format?.default_path || '/'
-})
-
-// 动态 placeholder
-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()
 })
 
-// 重置表单
-function resetForm() {
-  form.value = {
-    api_format: '',
-    base_url: '',
-    custom_path: '',
-    timeout: 300,
-    max_retries: 3,
-    max_concurrent: undefined,
-    rate_limit: undefined,
-    is_active: true,
-    proxy_url: '',
-    proxy_username: '',
-    proxy_password: '',
+// 监听 props 变化
+watch(() => props.modelValue, (open) => {
+  if (open) {
+    localEndpoints.value = [...(props.endpoints || [])]
+    // 重置编辑状态
+    editingEndpointId.value = null
+    editingUrl.value = ''
+    editingPath.value = ''
+  } else {
+    // 关闭对话框时完全清空新端点表单
+    newEndpoint.value = { api_format: '', base_url: '', custom_path: '' }
   }
-  proxyEnabled.value = false
+}, { immediate: true })
+
+watch(() => props.endpoints, (endpoints) => {
+  if (props.modelValue) {
+    localEndpoints.value = [...(endpoints || [])]
+  }
+}, { deep: true })
+
+// 开始编辑
+function startEdit(endpoint: ProviderEndpoint) {
+  editingEndpointId.value = endpoint.id
+  editingUrl.value = endpoint.base_url
+  editingPath.value = endpoint.custom_path || ''
 }
 
-// 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,
-    custom_path: props.endpoint.custom_path || '',
-    timeout: props.endpoint.timeout,
-    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,
-    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
+// 取消编辑
+function cancelEdit() {
+  editingEndpointId.value = null
+  editingUrl.value = ''
+  editingPath.value = ''
 }
 
-// 使用 useFormDialog 统一处理对话框逻辑
-const { isEditMode, handleDialogUpdate, handleCancel } = useFormDialog({
-  isOpen: () => props.modelValue,
-  entity: () => props.endpoint,
-  isLoading: loading,
-  onClose: () => emit('update:modelValue', false),
-  loadData: loadEndpointData,
-  resetForm,
-})
+// 保存端点
+async function saveEndpointUrl(endpoint: ProviderEndpoint) {
+  if (!editingUrl.value) return
 
-// 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 (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
+  savingEndpointId.value = endpoint.id
   try {
-    const proxyConfig = buildProxyConfig()
-
-    if (isEditMode.value && props.endpoint) {
-      // 更新端点
-      await updateEndpoint(props.endpoint.id, {
-        base_url: form.value.base_url,
-        custom_path: form.value.custom_path || undefined,
-        timeout: form.value.timeout,
-        max_retries: form.value.max_retries,
-        max_concurrent: form.value.max_concurrent,
-        rate_limit: form.value.rate_limit,
-        is_active: form.value.is_active,
-        proxy: proxyConfig,
-      })
-
-      success('端点已更新', '保存成功')
-      emit('endpointUpdated')
-    } else if (props.provider) {
-      // 创建端点
-      await createEndpoint(props.provider.id, {
-        provider_id: props.provider.id,
-        api_format: form.value.api_format,
-        base_url: form.value.base_url,
-        custom_path: form.value.custom_path || undefined,
-        timeout: form.value.timeout,
-        max_retries: form.value.max_retries,
-        max_concurrent: form.value.max_concurrent,
-        rate_limit: form.value.rate_limit,
-        is_active: form.value.is_active,
-        proxy: proxyConfig,
-      })
-
-      success('端点创建成功', '成功')
-      emit('endpointCreated')
-      resetForm()
-    }
-
-    emit('update:modelValue', false)
+    await updateEndpoint(endpoint.id, {
+      base_url: editingUrl.value,
+      custom_path: editingPath.value || null,  // 空字符串时传 null 清空
+    })
+    success('端点已更新')
+    emit('endpointUpdated')
+    cancelEdit()
   } catch (error: any) {
-    const action = isEditMode.value ? '更新' : '创建'
-    showError(error.response?.data?.detail || `${action}端点失败`, '错误')
+    showError(error.response?.data?.detail || '更新失败', '错误')
   } finally {
-    loading.value = false
+    savingEndpointId.value = null
   }
 }
 
-// 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
+// 添加端点
+async function handleAddEndpoint() {
+  if (!props.provider || !newEndpoint.value.api_format || !newEndpoint.value.base_url) return
+
+  addingEndpoint.value = true
+  try {
+    await createEndpoint(props.provider.id, {
+      provider_id: props.provider.id,
+      api_format: newEndpoint.value.api_format,
+      base_url: newEndpoint.value.base_url,
+      custom_path: newEndpoint.value.custom_path || undefined,
+      is_active: true,
+    })
+    success(`已添加 ${API_FORMAT_LABELS[newEndpoint.value.api_format] || newEndpoint.value.api_format} 端点`)
+    // 重置表单，保留 URL
+    const url = newEndpoint.value.base_url
+    newEndpoint.value = { api_format: '', base_url: url, custom_path: '' }
+    emit('endpointCreated')
+  } catch (error: any) {
+    showError(error.response?.data?.detail || '添加失败', '错误')
+  } finally {
+    addingEndpoint.value = false
+  }
+}
+
+// 切换端点启用状态
+async function handleToggleEndpoint(endpoint: ProviderEndpoint) {
+  togglingEndpointId.value = endpoint.id
+  try {
+    const newStatus = !endpoint.is_active
+    await updateEndpoint(endpoint.id, { is_active: newStatus })
+    success(newStatus ? '端点已启用' : '端点已停用')
+    emit('endpointUpdated')
+  } catch (error: any) {
+    showError(error.response?.data?.detail || '操作失败', '错误')
+  } finally {
+    togglingEndpointId.value = null
+  }
+}
+
+// 删除端点
+async function handleDeleteEndpoint(endpoint: ProviderEndpoint) {
+  deletingEndpointId.value = endpoint.id
+  try {
+    await deleteEndpoint(endpoint.id)
+    success(`已删除 ${API_FORMAT_LABELS[endpoint.api_format] || endpoint.api_format} 端点`)
+    emit('endpointUpdated')
+  } catch (error: any) {
+    showError(error.response?.data?.detail || '删除失败', '错误')
+  } finally {
+    deletingEndpointId.value = null
+  }
+}
+
+// 关闭对话框
+function handleDialogUpdate(value: boolean) {
+  emit('update:modelValue', value)
+}
+
+function handleClose() {
+  emit('update:modelValue', false)
 }
 </script>

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

@@ -1,165 +1,179 @@
 <template>
   <Dialog
     :model-value="isOpen"
-    title="配置允许的模型"
-    description="选择该 API Key 允许访问的模型，留空则允许访问所有模型"
-    :icon="Settings2"
+    title="获取上游模型"
+    :description="`使用密钥 ${props.apiKey?.name || props.apiKey?.api_key_masked || ''} 从上游获取模型列表。导入的模型需要关联全局模型后才能参与路由。`"
+    :icon="Layers"
     size="2xl"
     @update:model-value="handleDialogUpdate"
   >
     <div class="space-y-4 py-2">
-      <!-- 已选模型展示 -->
-      <div
-        v-if="selectedModels.length > 0"
-        class="space-y-2"
-      >
-        <div class="flex items-center justify-between px-1">
-          <div class="text-xs font-medium text-muted-foreground">
-            已选模型 ({{ selectedModels.length }})
-          </div>
-          <Button
-            type="button"
-            variant="ghost"
-            size="sm"
-            class="h-6 text-xs hover:text-destructive"
-            @click="clearModels"
-          >
-            清空
-          </Button>
-        </div>
-        <div class="flex flex-wrap gap-1.5 p-2 bg-muted/20 rounded-lg border border-border/40 min-h-[40px]">
-          <Badge
-            v-for="modelName in selectedModels"
-            :key="modelName"
-            variant="secondary"
-            class="text-[11px] px-2 py-0.5 bg-background border-border/60 shadow-sm"
-          >
-            {{ getModelLabel(modelName) }}
-            <button
-              class="ml-0.5 hover:text-destructive focus:outline-none"
-              @click.stop="toggleModel(modelName, false)"
-            >
-              &times;
-            </button>
-          </Badge>
+      <!-- 操作区域 -->
+      <div class="flex items-center justify-between">
+        <div class="text-sm text-muted-foreground">
+          <span v-if="!hasQueried">点击获取按钮查询上游可用模型</span>
+          <span v-else-if="upstreamModels.length > 0">
+            共 {{ upstreamModels.length }} 个模型，已选 {{ selectedModels.length }} 个
+          </span>
+          <span v-else>未找到可用模型</span>
         </div>
+        <Button
+          variant="outline"
+          size="sm"
+          :disabled="loading"
+          @click="fetchUpstreamModels"
+        >
+          <RefreshCw
+            class="w-3.5 h-3.5 mr-1.5"
+            :class="{ 'animate-spin': loading }"
+          />
+          {{ hasQueried ? '刷新' : '获取模型' }}
+        </Button>
       </div>
 
-      <!-- 模型列表区域 -->
-      <div class="space-y-2">
+      <!-- 加载状态 -->
+      <div
+        v-if="loading"
+        class="flex flex-col items-center justify-center py-12 space-y-3"
+      >
+        <div class="animate-spin rounded-full h-8 w-8 border-2 border-primary/20 border-t-primary" />
+        <span class="text-xs text-muted-foreground">正在从上游获取模型列表...</span>
+      </div>
+
+      <!-- 错误状态 -->
+      <div
+        v-else-if="errorMessage"
+        class="flex flex-col items-center justify-center py-12 text-destructive border border-dashed border-destructive/30 rounded-lg bg-destructive/5"
+      >
+        <AlertCircle class="w-10 h-10 mb-2 opacity-50" />
+        <span class="text-sm text-center px-4">{{ errorMessage }}</span>
+        <Button
+          variant="outline"
+          size="sm"
+          class="mt-3"
+          @click="fetchUpstreamModels"
+        >
+          重试
+        </Button>
+      </div>
+
+      <!-- 未查询状态 -->
+      <div
+        v-else-if="!hasQueried"
+        class="flex flex-col items-center justify-center py-12 text-muted-foreground border border-dashed rounded-lg bg-muted/10"
+      >
+        <Layers class="w-10 h-10 mb-2 opacity-20" />
+        <span class="text-sm">点击上方按钮获取模型列表</span>
+      </div>
+
+      <!-- 无模型 -->
+      <div
+        v-else-if="upstreamModels.length === 0"
+        class="flex flex-col items-center justify-center py-12 text-muted-foreground border border-dashed rounded-lg bg-muted/10"
+      >
+        <Box class="w-10 h-10 mb-2 opacity-20" />
+        <span class="text-sm">上游 API 未返回可用模型</span>
+      </div>
+
+      <!-- 模型列表 -->
+      <div v-else class="space-y-2">
+        <!-- 全选/取消 -->
         <div class="flex items-center justify-between px-1">
-          <div class="text-xs font-medium text-muted-foreground">
-            可选模型列表
+          <div class="flex items-center gap-2">
+            <Checkbox
+              :checked="isAllSelected"
+              :indeterminate="isPartiallySelected"
+              @update:checked="toggleSelectAll"
+            />
+            <span class="text-xs text-muted-foreground">
+              {{ isAllSelected ? '取消全选' : '全选' }}
+            </span>
           </div>
-          <div
-            v-if="!loadingModels && availableModels.length > 0"
-            class="text-[10px] text-muted-foreground/60"
-          >
-            共 {{ availableModels.length }} 个模型
+          <div class="text-xs text-muted-foreground">
+            {{ newModelsCount }} 个新模型（不在本地）
           </div>
         </div>
 
-        <!-- 加载状态 -->
-        <div
-          v-if="loadingModels"
-          class="flex flex-col items-center justify-center py-12 space-y-3"
-        >
-          <div class="animate-spin rounded-full h-8 w-8 border-2 border-primary/20 border-t-primary" />
-          <span class="text-xs text-muted-foreground">正在加载模型列表...</span>
-        </div>
-
-        <!-- 无模型 -->
-        <div
-          v-else-if="availableModels.length === 0"
-          class="flex flex-col items-center justify-center py-12 text-muted-foreground border border-dashed rounded-lg bg-muted/10"
-        >
-          <Box class="w-10 h-10 mb-2 opacity-20" />
-          <span class="text-sm">暂无可选模型</span>
-        </div>
-
-        <!-- 模型列表 -->
-        <div
-          v-else
-          class="max-h-[320px] overflow-y-auto pr-1 space-y-1.5 custom-scrollbar"
-        >
+        <div class="max-h-[320px] overflow-y-auto pr-1 space-y-1 custom-scrollbar">
           <div
-            v-for="model in availableModels"
-            :key="model.global_model_name"
+            v-for="model in upstreamModels"
+            :key="`${model.id}:${model.api_format || ''}`"
             class="group flex items-center gap-3 px-3 py-2.5 rounded-lg border transition-all duration-200 cursor-pointer select-none"
             :class="[
-              selectedModels.includes(model.global_model_name)
+              selectedModels.includes(model.id)
                 ? 'border-primary/40 bg-primary/5 shadow-sm'
                 : 'border-border/40 bg-background hover:border-primary/20 hover:bg-muted/30'
             ]"
-            @click="toggleModel(model.global_model_name, !selectedModels.includes(model.global_model_name))"
+            @click="toggleModel(model.id)"
           >
-            <!-- Checkbox -->
             <Checkbox
-              :checked="selectedModels.includes(model.global_model_name)"
+              :checked="selectedModels.includes(model.id)"
               class="data-[state=checked]:bg-primary data-[state=checked]:border-primary"
               @click.stop
-              @update:checked="checked => toggleModel(model.global_model_name, checked)"
+              @update:checked="checked => toggleModel(model.id, checked)"
             />
-
-            <!-- Info -->
             <div class="flex-1 min-w-0">
-              <div class="flex items-center justify-between gap-2">
-                <span class="text-sm font-medium truncate text-foreground/90">{{ model.display_name }}</span>
-                <span
-                  v-if="hasPricing(model)"
-                  class="text-[10px] font-mono text-muted-foreground/80 bg-muted/30 px-1.5 py-0.5 rounded border border-border/30 shrink-0"
-                >
-                  {{ formatPricingShort(model) }}
+              <div class="flex items-center gap-2">
+                <span class="text-sm font-medium truncate text-foreground/90">
+                  {{ model.display_name || model.id }}
                 </span>
+                <Badge
+                  v-if="model.api_format"
+                  variant="outline"
+                  class="text-[10px] px-1.5 py-0 shrink-0"
+                >
+                  {{ API_FORMAT_LABELS[model.api_format] || model.api_format }}
+                </Badge>
+                <Badge
+                  v-if="isModelExisting(model.id)"
+                  variant="secondary"
+                  class="text-[10px] px-1.5 py-0 shrink-0"
+                >
+                  已存在
+                </Badge>
               </div>
               <div class="text-[11px] text-muted-foreground/60 font-mono truncate mt-0.5">
-                {{ model.global_model_name }}
+                {{ model.id }}
               </div>
             </div>
-
-            <!-- 测试按钮 -->
-            <Button
-              variant="ghost"
-              size="icon"
-              class="h-7 w-7 shrink-0"
-              title="测试模型连接"
-              :disabled="testingModelName === model.global_model_name"
-              @click.stop="testModelConnection(model)"
+            <div
+              v-if="model.owned_by"
+              class="text-[10px] text-muted-foreground/50 shrink-0"
             >
-              <Loader2
-                v-if="testingModelName === model.global_model_name"
-                class="w-3.5 h-3.5 animate-spin"
-              />
-              <Play
-                v-else
-                class="w-3.5 h-3.5"
-              />
-            </Button>
+              {{ model.owned_by }}
+            </div>
           </div>
         </div>
       </div>
     </div>
 
     <template #footer>
-      <div class="flex items-center justify-end gap-2 w-full pt-2">
-        <Button
-          variant="outline"
-          class="h-9"
-          @click="handleCancel"
-        >
-          取消
-        </Button>
-        <Button
-          :disabled="saving"
-          class="h-9 min-w-[80px]"
-          @click="handleSave"
-        >
-          <Loader2
-            v-if="saving"
-            class="w-3.5 h-3.5 mr-1.5 animate-spin"
-          />
-          {{ saving ? '保存中' : '保存配置' }}
-        </Button>
+      <div class="flex items-center justify-between w-full pt-2">
+        <div class="text-xs text-muted-foreground">
+          <span v-if="selectedModels.length > 0 && newSelectedCount > 0">
+            将导入 {{ newSelectedCount }} 个新模型
+          </span>
+        </div>
+        <div class="flex items-center gap-2">
+          <Button
+            variant="outline"
+            class="h-9"
+            @click="handleCancel"
+          >
+            取消
+          </Button>
+          <Button
+            :disabled="importing || selectedModels.length === 0 || newSelectedCount === 0"
+            class="h-9 min-w-[100px]"
+            @click="handleImport"
+          >
+            <Loader2
+              v-if="importing"
+              class="w-3.5 h-3.5 mr-1.5 animate-spin"
+            />
+            {{ importing ? '导入中' : `导入 ${newSelectedCount} 个模型` }}
+          </Button>
+        </div>
       </div>
     </template>
   </Dialog>
@@ -167,19 +181,20 @@
 
 <script setup lang="ts">
 import { ref, computed, watch } from 'vue'
-import { Box, Loader2, Settings2, Play } from 'lucide-vue-next'
+import { Box, Layers, Loader2, RefreshCw, AlertCircle } from 'lucide-vue-next'
 import { Dialog } from '@/components/ui'
 import Button from '@/components/ui/button.vue'
 import Badge from '@/components/ui/badge.vue'
 import Checkbox from '@/components/ui/checkbox.vue'
 import { useToast } from '@/composables/useToast'
-import { parseApiError, parseTestModelError } from '@/utils/errorParser'
+import { parseUpstreamModelError } from '@/utils/errorParser'
+import { adminApi } from '@/api/admin'
 import {
-  updateEndpointKey,
-  getProviderAvailableSourceModels,
-  testModel,
+  importModelsFromUpstream,
+  getProviderModels,
   type EndpointAPIKey,
-  type ProviderAvailableSourceModel
+  type UpstreamModel,
+  API_FORMAT_LABELS,
 } from '@/api/endpoints'
 
 const props = defineProps<{
@@ -196,130 +211,121 @@ const emit = defineEmits<{
 const { success, error: showError } = useToast()
 
 const isOpen = computed(() => props.open)
-const saving = ref(false)
-const loadingModels = ref(false)
-const availableModels = ref<ProviderAvailableSourceModel[]>([])
+const loading = ref(false)
+const importing = ref(false)
+const hasQueried = ref(false)
+const errorMessage = ref('')
+const upstreamModels = ref<UpstreamModel[]>([])
 const selectedModels = ref<string[]>([])
-const initialModels = ref<string[]>([])
-const testingModelName = ref<string | null>(null)
+const existingModelIds = ref<Set<string>>(new Set())
+
+// 计算属性
+const isAllSelected = computed(() =>
+  upstreamModels.value.length > 0 &&
+  selectedModels.value.length === upstreamModels.value.length
+)
+
+const isPartiallySelected = computed(() =>
+  selectedModels.value.length > 0 &&
+  selectedModels.value.length < upstreamModels.value.length
+)
+
+const newModelsCount = computed(() =>
+  upstreamModels.value.filter(m => !existingModelIds.value.has(m.id)).length
+)
+
+const newSelectedCount = computed(() =>
+  selectedModels.value.filter(id => !existingModelIds.value.has(id)).length
+)
+
+// 检查模型是否已存在
+function isModelExisting(modelId: string): boolean {
+  return existingModelIds.value.has(modelId)
+}
 
 // 监听对话框打开
 watch(() => props.open, (open) => {
   if (open) {
-    loadData()
+    resetState()
+    loadExistingModels()
   }
 })
 
-async function loadData() {
-  // 初始化已选模型
-  if (props.apiKey?.allowed_models) {
-    selectedModels.value = [...props.apiKey.allowed_models]
-    initialModels.value = [...props.apiKey.allowed_models]
-  } else {
-    selectedModels.value = []
-    initialModels.value = []
-  }
-
-  // 加载可选模型
-  if (props.providerId) {
-    await loadAvailableModels()
-  }
-}
-
-async function loadAvailableModels() {
-  if (!props.providerId) return
-  try {
-    loadingModels.value = true
-    const response = await getProviderAvailableSourceModels(props.providerId)
-    availableModels.value = response.models
-  } catch (err: any) {
-    const errorMessage = parseApiError(err, '加载模型列表失败')
-    showError(errorMessage, '错误')
-  } finally {
-    loadingModels.value = false
-  }
-}
-
-const modelLabelMap = computed(() => {
-  const map = new Map<string, string>()
-  availableModels.value.forEach(model => {
-    map.set(model.global_model_name, model.display_name || model.global_model_name)
-  })
-  return map
-})
-
-function getModelLabel(modelName: string): string {
-  return modelLabelMap.value.get(modelName) ?? modelName
-}
-
-function hasPricing(model: ProviderAvailableSourceModel): boolean {
-  const input = model.price.input_price_per_1m ?? 0
-  const output = model.price.output_price_per_1m ?? 0
-  return input > 0 || output > 0
-}
-
-function formatPricingShort(model: ProviderAvailableSourceModel): string {
-  const input = model.price.input_price_per_1m ?? 0
-  const output = model.price.output_price_per_1m ?? 0
-  if (input > 0 || output > 0) {
-    return `$${formatPrice(input)}/$${formatPrice(output)}`
-  }
-  return ''
-}
-
-function formatPrice(value?: number | null): string {
-  if (value === undefined || value === null || value === 0) return '0'
-  if (value >= 1) {
-    return value.toFixed(2)
-  }
-  return value.toFixed(2)
-}
-
-function toggleModel(modelName: string, checked: boolean) {
-  if (checked) {
-    if (!selectedModels.value.includes(modelName)) {
-      selectedModels.value = [...selectedModels.value, modelName]
-    }
-  } else {
-    selectedModels.value = selectedModels.value.filter(name => name !== modelName)
-  }
-}
-
-function clearModels() {
+function resetState() {
+  hasQueried.value = false
+  errorMessage.value = ''
+  upstreamModels.value = []
   selectedModels.value = []
 }
 
-// 测试模型连接
-async function testModelConnection(model: ProviderAvailableSourceModel) {
-  if (!props.providerId || !props.apiKey || testingModelName.value) return
-
-  testingModelName.value = model.global_model_name
+// 加载已存在的模型列表
+async function loadExistingModels() {
+  if (!props.providerId) return
   try {
-    const result = await testModel({
-      provider_id: props.providerId,
-      model_name: model.provider_model_name,
-      api_key_id: props.apiKey.id,
-      message: "hello"
-    })
-
-    if (result.success) {
-      success(`模型 "${model.display_name}" 测试成功`)
-    } else {
-      showError(`模型测试失败: ${parseTestModelError(result)}`)
-    }
-  } catch (err: any) {
-    const errorMsg = err.response?.data?.detail || err.message || '测试请求失败'
-    showError(`模型测试失败: ${errorMsg}`)
-  } finally {
-    testingModelName.value = null
+    const models = await getProviderModels(props.providerId)
+    existingModelIds.value = new Set(
+      models.map((m: { provider_model_name: string }) => m.provider_model_name)
+    )
+  } catch {
+    existingModelIds.value = new Set()
   }
 }
 
-function areArraysEqual(a: string[], b: string[]): boolean {
-  if (a.length !== b.length) return false
-  const sortedA = [...a].sort()
-  const sortedB = [...b].sort()
-  return sortedA.every((value, index) => value === sortedB[index])
+// 获取上游模型
+async function fetchUpstreamModels() {
+  if (!props.providerId || !props.apiKey) return
+
+  loading.value = true
+  errorMessage.value = ''
+
+  try {
+    const response = await adminApi.queryProviderModels(props.providerId, props.apiKey.id)
+
+    if (response.success && response.data?.models) {
+      upstreamModels.value = response.data.models
+      // 默认选中所有新模型
+      selectedModels.value = response.data.models
+        .filter((m: UpstreamModel) => !existingModelIds.value.has(m.id))
+        .map((m: UpstreamModel) => m.id)
+      hasQueried.value = true
+      // 如果有部分失败，显示警告提示
+      if (response.data.error) {
+        // 使用友好的错误解析
+        showError(`部分格式获取失败: ${parseUpstreamModelError(response.data.error)}`, '警告')
+      }
+    } else {
+      // 使用友好的错误解析
+      const rawError = response.data?.error || '获取上游模型失败'
+      errorMessage.value = parseUpstreamModelError(rawError)
+    }
+  } catch (err: any) {
+    // 使用友好的错误解析
+    const rawError = err.response?.data?.detail || err.message || '获取上游模型失败'
+    errorMessage.value = parseUpstreamModelError(rawError)
+  } finally {
+    loading.value = false
+  }
+}
+
+// 切换模型选择
+function toggleModel(modelId: string, checked?: boolean) {
+  const shouldSelect = checked !== undefined ? checked : !selectedModels.value.includes(modelId)
+  if (shouldSelect) {
+    if (!selectedModels.value.includes(modelId)) {
+      selectedModels.value = [...selectedModels.value, modelId]
+    }
+  } else {
+    selectedModels.value = selectedModels.value.filter(id => id !== modelId)
+  }
+}
+
+// 全选/取消全选
+function toggleSelectAll(checked: boolean) {
+  if (checked) {
+    selectedModels.value = upstreamModels.value.map(m => m.id)
+  } else {
+    selectedModels.value = []
+  }
 }
 
 function handleDialogUpdate(value: boolean) {
@@ -332,30 +338,44 @@ function handleCancel() {
   emit('close')
 }
 
-async function handleSave() {
-  if (!props.apiKey) return
+// 导入选中的模型
+async function handleImport() {
+  if (!props.providerId || selectedModels.value.length === 0) return
 
-  // 检查是否有变化
-  const hasChanged = !areArraysEqual(selectedModels.value, initialModels.value)
-  if (!hasChanged) {
-    emit('close')
+  // 过滤出新模型（不在已存在列表中的）
+  const modelsToImport = selectedModels.value.filter(id => !existingModelIds.value.has(id))
+  if (modelsToImport.length === 0) {
+    showError('所选模型都已存在', '提示')
     return
   }
 
-  saving.value = true
+  importing.value = true
   try {
-    await updateEndpointKey(props.apiKey.id, {
-      // 空数组时发送 null，表示允许所有模型
-      allowed_models: selectedModels.value.length > 0 ? [...selectedModels.value] : null
-    })
-    success('允许的模型已更新', '成功')
-    emit('saved')
-    emit('close')
+    const response = await importModelsFromUpstream(props.providerId, modelsToImport)
+
+    const successCount = response.success?.length || 0
+    const errorCount = response.errors?.length || 0
+
+    if (successCount > 0 && errorCount === 0) {
+      success(`成功导入 ${successCount} 个模型`, '导入成功')
+      emit('saved')
+      emit('close')
+    } else if (successCount > 0 && errorCount > 0) {
+      success(`成功导入 ${successCount} 个模型，${errorCount} 个失败`, '部分成功')
+      emit('saved')
+      // 刷新列表以更新已存在状态
+      await loadExistingModels()
+      // 更新选中列表，移除已成功导入的
+      const successIds = new Set(response.success?.map((s: { model_id: string }) => s.model_id) || [])
+      selectedModels.value = selectedModels.value.filter(id => !successIds.has(id))
+    } else {
+      const errorMsg = response.errors?.[0]?.error || '导入失败'
+      showError(errorMsg, '导入失败')
+    }
   } catch (err: any) {
-    const errorMessage = parseApiError(err, '保存失败')
-    showError(errorMessage, '错误')
+    showError(err.response?.data?.detail || '导入失败', '错误')
   } finally {
-    saving.value = false
+    importing.value = false
   }
 }
 </script>

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

@@ -0,0 +1,640 @@
+<template>
+  <Dialog
+    :model-value="isOpen"
+    :title="props.apiKey?.name ? `模型权限 - ${props.apiKey.name}` : '模型权限'"
+    description="选中的模型将被允许访问，不选择则允许全部"
+    :icon="Shield"
+    size="2xl"
+    @update:model-value="handleDialogUpdate"
+  >
+    <template #default>
+      <div class="space-y-4">
+        <!-- 字典模式警告 -->
+        <div
+          v-if="isDictMode"
+          class="rounded-lg border border-amber-500/50 bg-amber-50 dark:bg-amber-950/30 p-3"
+        >
+          <p class="text-sm text-amber-700 dark:text-amber-400">
+            <strong>注意：</strong>此密钥使用按 API 格式区分的模型权限配置。
+            编辑后将转换为统一列表模式，原有的格式区分信息将丢失。
+          </p>
+        </div>
+
+        <!-- 常驻选择面板 -->
+        <div class="border rounded-lg overflow-hidden">
+          <!-- 搜索 + 操作栏 -->
+          <div class="flex items-center gap-2 p-2 border-b bg-muted/30">
+            <div class="relative flex-1">
+              <Search class="absolute left-2 top-1/2 -translate-y-1/2 w-4 h-4 text-muted-foreground" />
+              <Input
+                v-model="searchQuery"
+                placeholder="搜索模型或添加自定义模型..."
+                class="pl-8 h-8 text-sm"
+              />
+            </div>
+            <!-- 已选数量徽章 -->
+            <span
+              v-if="selectedModels.length === 0"
+              class="h-6 px-2 text-xs rounded flex items-center bg-muted text-muted-foreground shrink-0"
+            >
+              全部模型
+            </span>
+            <span
+              v-else
+              class="h-6 px-2 text-xs rounded flex items-center bg-primary/10 text-primary shrink-0"
+            >
+              已选 {{ selectedModels.length }} 个
+            </span>
+            <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()"
+            >
+              <RefreshCw
+                class="w-4 h-4"
+                :class="{ 'animate-spin': fetchingUpstreamModels }"
+              />
+            </button>
+            <Button
+              v-else-if="!fetchingUpstreamModels"
+              variant="outline"
+              size="sm"
+              class="h-8"
+              title="从提供<E68F90><E4BE9B><EFBFBD>获取模型"
+              @click="fetchUpstreamModels()"
+            >
+              <Zap class="w-4 h-4" />
+            </Button>
+            <Loader2
+              v-else
+              class="w-4 h-4 animate-spin text-muted-foreground shrink-0"
+            />
+          </div>
+
+          <!-- 分组列表 -->
+          <div class="max-h-96 overflow-y-auto">
+            <!-- 加载中 -->
+            <div
+              v-if="loadingGlobalModels"
+              class="flex items-center justify-center py-12"
+            >
+              <Loader2 class="w-6 h-6 animate-spin text-primary" />
+            </div>
+
+            <template v-else>
+              <!-- 添加自定义模型（搜索内容不在列表中时显示，固定在顶部） -->
+              <div
+                v-if="searchQuery && canAddAsCustom"
+                class="px-3 py-2 border-b bg-background sticky top-0 z-10"
+              >
+                <div
+                  class="flex items-center justify-between px-3 py-2 rounded-lg border border-dashed hover:border-primary hover:bg-primary/5 cursor-pointer transition-colors"
+                  @click="addCustomModel"
+                >
+                  <div class="flex items-center gap-2">
+                    <Plus class="w-4 h-4 text-muted-foreground" />
+                    <span class="text-sm font-mono">{{ searchQuery }}</span>
+                  </div>
+                  <span class="text-xs text-muted-foreground">添加自定义模型</span>
+                </div>
+              </div>
+
+              <!-- 自定义模型（手动添加的，始终显示全部，搜索命中的排前面） -->
+              <div v-if="customModels.length > 0">
+                <div
+                  class="flex items-center justify-between px-3 py-2 bg-muted sticky top-0 z-10 cursor-pointer hover:bg-muted/80 transition-colors"
+                  @click="toggleGroupCollapse('custom')"
+                >
+                  <div class="flex items-center gap-2">
+                    <ChevronDown
+                      class="w-4 h-4 transition-transform shrink-0"
+                      :class="collapsedGroups.has('custom') ? '-rotate-90' : ''"
+                    />
+                    <span class="text-xs font-medium">自定义模型</span>
+                    <span class="text-xs text-muted-foreground">({{ customModels.length }})</span>
+                  </div>
+                </div>
+                <div
+                  v-show="!collapsedGroups.has('custom')"
+                  class="space-y-1 p-2"
+                >
+                  <div
+                    v-for="model in sortedCustomModels"
+                    :key="model"
+                    class="flex items-center gap-2 px-2 py-1.5 rounded hover:bg-muted cursor-pointer"
+                    @click="toggleModel(model)"
+                  >
+                    <div
+                      class="w-4 h-4 border rounded flex items-center justify-center shrink-0"
+                      :class="selectedModels.includes(model) ? 'bg-primary border-primary' : ''"
+                    >
+                      <Check v-if="selectedModels.includes(model)" class="w-3 h-3 text-primary-foreground" />
+                    </div>
+                    <span class="text-sm font-mono truncate">{{ model }}</span>
+                  </div>
+                </div>
+              </div>
+
+              <!-- 全局模型 -->
+              <div v-if="filteredGlobalModels.length > 0">
+                <div
+                  class="flex items-center justify-between px-3 py-2 bg-muted sticky top-0 z-10 cursor-pointer hover:bg-muted/80 transition-colors"
+                  @click="toggleGroupCollapse('global')"
+                >
+                  <div class="flex items-center gap-2">
+                    <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">({{ filteredGlobalModels.length }})</span>
+                  </div>
+                  <button
+                    type="button"
+                    class="text-xs text-primary hover:underline"
+                    @click.stop="toggleAllGlobalModels"
+                  >
+                    {{ isAllGlobalModelsSelected ? '取消全选' : '全选' }}
+                  </button>
+                </div>
+                <div
+                  v-show="!collapsedGroups.has('global')"
+                  class="space-y-1 p-2"
+                >
+                  <div
+                    v-for="model in filteredGlobalModels"
+                    :key="model.name"
+                    class="flex items-center gap-2 px-2 py-1.5 rounded hover:bg-muted cursor-pointer"
+                    @click="toggleModel(model.name)"
+                  >
+                    <div
+                      class="w-4 h-4 border rounded flex items-center justify-center shrink-0"
+                      :class="selectedModels.includes(model.name) ? 'bg-primary border-primary' : ''"
+                    >
+                      <Check v-if="selectedModels.includes(model.name)" class="w-3 h-3 text-primary-foreground" />
+                    </div>
+                    <div class="flex-1 min-w-0">
+                      <p class="text-sm font-medium truncate">{{ model.display_name }}</p>
+                      <p class="text-xs text-muted-foreground truncate font-mono">{{ model.name }}</p>
+                    </div>
+                  </div>
+                </div>
+              </div>
+
+              <!-- 上游模型组 -->
+              <div v-for="group in filteredUpstreamGroups" :key="group.api_format">
+                <div
+                  class="flex items-center justify-between px-3 py-2 bg-muted sticky top-0 z-10 cursor-pointer hover:bg-muted/80 transition-colors"
+                  @click="toggleGroupCollapse(group.api_format)"
+                >
+                  <div class="flex items-center gap-2">
+                    <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>
+                  </div>
+                  <button
+                    type="button"
+                    class="text-xs text-primary hover:underline"
+                    @click.stop="toggleAllUpstreamGroup(group.api_format)"
+                  >
+                    {{ isUpstreamGroupAllSelected(group.api_format) ? '取消全选' : '全选' }}
+                  </button>
+                </div>
+                <div
+                  v-show="!collapsedGroups.has(group.api_format)"
+                  class="space-y-1 p-2"
+                >
+                  <div
+                    v-for="model in group.models"
+                    :key="model.id"
+                    class="flex items-center gap-2 px-2 py-1.5 rounded hover:bg-muted cursor-pointer"
+                    @click="toggleModel(model.id)"
+                  >
+                    <div
+                      class="w-4 h-4 border rounded flex items-center justify-center shrink-0"
+                      :class="selectedModels.includes(model.id) ? 'bg-primary border-primary' : ''"
+                    >
+                      <Check v-if="selectedModels.includes(model.id)" class="w-3 h-3 text-primary-foreground" />
+                    </div>
+                    <span class="text-sm font-mono truncate">{{ model.id }}</span>
+                  </div>
+                </div>
+              </div>
+
+              <!-- 空状态 -->
+              <div
+                v-if="filteredGlobalModels.length === 0 && filteredUpstreamGroups.length === 0 && customModels.length === 0"
+                class="flex flex-col items-center justify-center py-12 text-muted-foreground"
+              >
+                <Shield class="w-10 h-10 mb-2 opacity-30" />
+                <p class="text-sm">{{ searchQuery ? '无匹配结果' : '暂无可选模型' }}</p>
+                <p v-if="!upstreamModelsLoaded" class="text-xs mt-1">点击闪电按钮从上游获取模型</p>
+              </div>
+            </template>
+          </div>
+        </div>
+      </div>
+    </template>
+
+    <template #footer>
+      <div class="flex items-center justify-between w-full">
+        <p class="text-xs text-muted-foreground">
+          {{ hasChanges ? '有未保存的更改' : '' }}
+        </p>
+        <div class="flex items-center gap-2">
+          <Button :disabled="saving || !hasChanges" @click="handleSave">
+            {{ saving ? '保存中...' : '保存' }}
+          </Button>
+          <Button variant="outline" @click="handleCancel">取消</Button>
+        </div>
+      </div>
+    </template>
+  </Dialog>
+</template>
+
+<script setup lang="ts">
+import { ref, computed, watch, onUnmounted } from 'vue'
+import {
+  Shield,
+  Search,
+  RefreshCw,
+  Loader2,
+  Zap,
+  Plus,
+  Check,
+  ChevronDown
+} from 'lucide-vue-next'
+import { Dialog, Button, Input } from '@/components/ui'
+import { useToast } from '@/composables/useToast'
+import { useConfirm } from '@/composables/useConfirm'
+import { parseApiError, parseUpstreamModelError } from '@/utils/errorParser'
+import {
+  updateProviderKey,
+  API_FORMAT_LABELS,
+  type EndpointAPIKey,
+  type AllowedModels,
+} from '@/api/endpoints'
+import { getGlobalModels, type GlobalModelResponse } from '@/api/global-models'
+import { adminApi } from '@/api/admin'
+import type { UpstreamModel } from '@/api/endpoints/types'
+
+interface AvailableModel {
+  name: string
+  display_name: string
+}
+
+const props = defineProps<{
+  open: boolean
+  apiKey: EndpointAPIKey | null
+  providerId: string
+}>()
+
+const emit = defineEmits<{
+  close: []
+  saved: []
+}>()
+
+const { success, error: showError } = useToast()
+const { confirmWarning } = useConfirm()
+
+const isOpen = computed(() => props.open)
+const saving = ref(false)
+const loadingGlobalModels = ref(false)
+const fetchingUpstreamModels = ref(false)
+const upstreamModelsLoaded = ref(false)
+
+// 用于取消异步操作的标志
+let loadingCancelled = false
+
+// 搜索
+const searchQuery = ref('')
+
+// 可用模型列表（全局模型）
+const allGlobalModels = ref<AvailableModel[]>([])
+// 上游模型列表
+const upstreamModels = ref<UpstreamModel[]>([])
+
+// 已选中的模型
+const selectedModels = ref<string[]>([])
+const initialSelectedModels = ref<string[]>([])
+
+// 所有添加过的自定义模型（包括已取消勾选的，保存前不消失）
+const allCustomModels = ref<string[]>([])
+
+// 是否为字典模式（按 API 格式区分）
+const isDictMode = ref(false)
+
+// 折叠状态
+const collapsedGroups = ref<Set<string>>(new Set())
+
+// 是否有更改
+const hasChanges = computed(() => {
+  if (selectedModels.value.length !== initialSelectedModels.value.length) return true
+  const sorted1 = [...selectedModels.value].sort()
+  const sorted2 = [...initialSelectedModels.value].sort()
+  return sorted1.some((v, i) => v !== sorted2[i])
+})
+
+// 所有已知模型的集合（全局 + 上游）
+const allKnownModels = computed(() => {
+  const set = new Set<string>()
+  allGlobalModels.value.forEach(m => set.add(m.name))
+  upstreamModels.value.forEach(m => set.add(m.id))
+  return set
+})
+
+// 自定义模型列表（显示所有添加过的，不因取消勾选而消失）
+const customModels = computed(() => {
+  return allCustomModels.value
+})
+
+// 排序后的自定义模型（搜索命中的排前面）
+const sortedCustomModels = computed(() => {
+  const search = searchQuery.value.toLowerCase().trim()
+  if (!search) return customModels.value
+
+  const matched: string[] = []
+  const unmatched: string[] = []
+  for (const m of customModels.value) {
+    if (m.toLowerCase().includes(search)) {
+      matched.push(m)
+    } else {
+      unmatched.push(m)
+    }
+  }
+  return [...matched, ...unmatched]
+})
+
+// 判断搜索内容是否可以作为自定义模型添加
+const canAddAsCustom = computed(() => {
+  const search = searchQuery.value.trim()
+  if (!search) return false
+  // 已经选中了就不显示
+  if (selectedModels.value.includes(search)) return false
+  // 已经在自定义模型列表中就不显示
+  if (allCustomModels.value.includes(search)) return false
+  // 精确匹配全局模型就不显示
+  if (allGlobalModels.value.some(m => m.name === search)) return false
+  // 精确匹配上游模型就不显示
+  if (upstreamModels.value.some(m => m.id === search)) return false
+  return true
+})
+
+// 搜索过滤后的全局模型
+const filteredGlobalModels = computed(() => {
+  if (!searchQuery.value.trim()) return allGlobalModels.value
+  const query = searchQuery.value.toLowerCase()
+  return allGlobalModels.value.filter(m =>
+    m.name.toLowerCase().includes(query) ||
+    m.display_name.toLowerCase().includes(query)
+  )
+})
+
+// 按 API 格式分组的上游模型（过滤后）
+const filteredUpstreamGroups = computed(() => {
+  if (!upstreamModelsLoaded.value) return []
+
+  const query = searchQuery.value.toLowerCase().trim()
+  const groups: Record<string, UpstreamModel[]> = {}
+
+  for (const model of upstreamModels.value) {
+    // 搜索过滤
+    if (query && !model.id.toLowerCase().includes(query)) continue
+
+    const format = model.api_format || 'unknown'
+    if (!groups[format]) groups[format] = []
+    groups[format].push(model)
+  }
+
+  const order = Object.keys(API_FORMAT_LABELS)
+  return Object.entries(groups)
+    .map(([api_format, models]) => ({ api_format, models }))
+    .filter(g => g.models.length > 0)
+    .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
+    })
+})
+
+// 全局模型是否全选
+const isAllGlobalModelsSelected = computed(() => {
+  if (filteredGlobalModels.value.length === 0) return false
+  return filteredGlobalModels.value.every(m => selectedModels.value.includes(m.name))
+})
+
+// 检查某个上游组是否全选
+function isUpstreamGroupAllSelected(apiFormat: string): boolean {
+  const group = filteredUpstreamGroups.value.find(g => g.api_format === apiFormat)
+  if (!group || group.models.length === 0) return false
+  return group.models.every(m => selectedModels.value.includes(m.id))
+}
+
+// 切换模型选中状态
+function toggleModel(modelId: string) {
+  const idx = selectedModels.value.indexOf(modelId)
+  if (idx === -1) {
+    selectedModels.value.push(modelId)
+  } else {
+    selectedModels.value.splice(idx, 1)
+  }
+}
+
+// 添加自定义模型
+function addCustomModel() {
+  const model = searchQuery.value.trim()
+  if (model && !selectedModels.value.includes(model)) {
+    selectedModels.value.push(model)
+    // 同时添加到自定义模型列表（如果不在已知模型中）
+    if (!allKnownModels.value.has(model) && !allCustomModels.value.includes(model)) {
+      allCustomModels.value.push(model)
+    }
+    searchQuery.value = ''
+  }
+}
+
+// 全选/取消全选全局模型
+function toggleAllGlobalModels() {
+  const allNames = filteredGlobalModels.value.map(m => m.name)
+  if (isAllGlobalModelsSelected.value) {
+    // 取消全选
+    selectedModels.value = selectedModels.value.filter(id => !allNames.includes(id))
+  } else {
+    // 全选
+    allNames.forEach(name => {
+      if (!selectedModels.value.includes(name)) {
+        selectedModels.value.push(name)
+      }
+    })
+  }
+}
+
+// 全选/取消全选某个上游组
+function toggleAllUpstreamGroup(apiFormat: string) {
+  const group = filteredUpstreamGroups.value.find(g => g.api_format === apiFormat)
+  if (!group) return
+
+  const allIds = group.models.map(m => m.id)
+  if (isUpstreamGroupAllSelected(apiFormat)) {
+    // 取消全选
+    selectedModels.value = selectedModels.value.filter(id => !allIds.includes(id))
+  } else {
+    // 全选
+    allIds.forEach(id => {
+      if (!selectedModels.value.includes(id)) {
+        selectedModels.value.push(id)
+      }
+    })
+  }
+}
+
+// 切换折叠状态
+function toggleGroupCollapse(group: string) {
+  if (collapsedGroups.value.has(group)) {
+    collapsedGroups.value.delete(group)
+  } else {
+    collapsedGroups.value.add(group)
+  }
+  collapsedGroups.value = new Set(collapsedGroups.value)
+}
+
+// 加载全局模型
+async function loadGlobalModels() {
+  loadingGlobalModels.value = true
+  try {
+    const response = await getGlobalModels({ limit: 1000 })
+    if (loadingCancelled) return
+    allGlobalModels.value = response.models.map((m: GlobalModelResponse) => ({
+      name: m.name,
+      display_name: m.display_name
+    }))
+  } catch (err) {
+    if (loadingCancelled) return
+    showError('加载全局模型失败', '错误')
+  } finally {
+    loadingGlobalModels.value = false
+  }
+}
+
+// 从提供商获取模型
+async function fetchUpstreamModels() {
+  if (!props.providerId || !props.apiKey) return
+  try {
+    fetchingUpstreamModels.value = true
+    const response = await adminApi.queryProviderModels(props.providerId, props.apiKey.id)
+    if (loadingCancelled) return
+    if (response.success && response.data?.models) {
+      upstreamModels.value = response.data.models
+      upstreamModelsLoaded.value = true
+      // 获取上游模型后，从自定义模型列表中移除已变成已知的模型
+      const upstreamIds = new Set(response.data.models.map((m: UpstreamModel) => m.id))
+      allCustomModels.value = allCustomModels.value.filter(m => !upstreamIds.has(m))
+    } else {
+      const errorMsg = response.data?.error
+        ? parseUpstreamModelError(response.data.error)
+        : '获取上游模型失败'
+      showError(errorMsg, '获取上游模型失败')
+    }
+  } catch (err: any) {
+    if (loadingCancelled) return
+    const rawError = err.response?.data?.detail || err.message || '获取上游模型失败'
+    showError(parseUpstreamModelError(rawError), '获取上游模型失败')
+  } finally {
+    fetchingUpstreamModels.value = false
+  }
+}
+
+// 解析 allowed_models
+function parseAllowedModels(allowed: AllowedModels): string[] {
+  if (allowed === null || allowed === undefined) {
+    isDictMode.value = false
+    return []
+  }
+  if (Array.isArray(allowed)) {
+    isDictMode.value = false
+    return [...allowed]
+  }
+  // 字典模式：合并所有格式的模型，并设置警告标志
+  isDictMode.value = true
+  const all = new Set<string>()
+  for (const models of Object.values(allowed)) {
+    models.forEach(m => all.add(m))
+  }
+  return Array.from(all)
+}
+
+// 监听对话框打开
+watch(() => props.open, async (open) => {
+  if (open && props.apiKey) {
+    loadingCancelled = false
+
+    const parsed = parseAllowedModels(props.apiKey.allowed_models ?? null)
+    selectedModels.value = [...parsed]
+    initialSelectedModels.value = [...parsed]
+    searchQuery.value = ''
+    upstreamModels.value = []
+    upstreamModelsLoaded.value = false
+    allCustomModels.value = []
+
+    await loadGlobalModels()
+
+    // 加载全局模型后，从已选中的模型中提取自定义模型（不在全局模型中的）
+    const globalModelNames = new Set(allGlobalModels.value.map(m => m.name))
+    allCustomModels.value = selectedModels.value.filter(m => !globalModelNames.has(m))
+  } else {
+    loadingCancelled = true
+  }
+})
+
+// 组件卸载时取消所有异步操作
+onUnmounted(() => {
+  loadingCancelled = true
+})
+
+async function handleDialogUpdate(value: boolean) {
+  if (!value && hasChanges.value) {
+    const confirmed = await confirmWarning('有未保存的更改，确定要关闭吗？', '放弃更改')
+    if (!confirmed) return
+  }
+  if (!value) emit('close')
+}
+
+async function handleCancel() {
+  if (hasChanges.value) {
+    const confirmed = await confirmWarning('有未保存的更改，确定要关闭吗？', '放弃更改')
+    if (!confirmed) return
+  }
+  emit('close')
+}
+
+async function handleSave() {
+  if (!props.apiKey) return
+
+  saving.value = true
+  try {
+    const newAllowed: AllowedModels = selectedModels.value.length > 0
+      ? [...selectedModels.value]
+      : null
+
+    await updateProviderKey(props.apiKey.id, { allowed_models: newAllowed })
+    success('模型权限已更新', '成功')
+    emit('saved')
+    emit('close')
+  } catch (err: any) {
+    showError(parseApiError(err, '保存失败'), '错误')
+  } finally {
+    saving.value = false
+  }
+}
+</script>

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

@@ -2,57 +2,36 @@
   <Dialog
     :model-value="isOpen"
     :title="isEditMode ? '编辑密钥' : '添加密钥'"
-    :description="isEditMode ? '修改 API 密钥配置' : '为端点添加新的 API 密钥'"
+    :description="isEditMode ? '修改 API 密钥配置' : '为提供商添加新的 API 密钥'"
     :icon="isEditMode ? SquarePen : Key"
-    size="2xl"
+    size="xl"
     @update:model-value="handleDialogUpdate"
   >
     <form
-      class="space-y-5"
+      class="space-y-4"
       autocomplete="off"
       @submit.prevent="handleSave"
     >
       <!-- 基本信息 -->
-      <div class="space-y-3">
-        <h3 class="text-sm font-medium border-b pb-2">
-          基本信息
-        </h3>
-        <div class="grid grid-cols-2 gap-4">
-          <div>
-            <Label :for="keyNameInputId">密钥名称 *</Label>
-            <Input
-              :id="keyNameInputId"
-              v-model="form.name"
-              :name="keyNameFieldName"
-              required
-              placeholder="例如：主 Key、备用 Key 1"
-              maxlength="100"
-              autocomplete="off"
-              autocapitalize="none"
-              autocorrect="off"
-              spellcheck="false"
-              data-form-type="other"
-              data-lpignore="true"
-              data-1p-ignore="true"
-            />
-          </div>
-          <div>
-            <Label for="rate_multiplier">成本倍率 *</Label>
-            <Input
-              id="rate_multiplier"
-              v-model.number="form.rate_multiplier"
-              type="number"
-              step="0.01"
-              min="0.01"
-              required
-              placeholder="1.0"
-            />
-            <p class="text-xs text-muted-foreground mt-1">
-              真实成本 = 表面成本 × 倍率
-            </p>
-          </div>
+      <div class="grid grid-cols-2 gap-3">
+        <div>
+          <Label :for="keyNameInputId">密钥名称 *</Label>
+          <Input
+            :id="keyNameInputId"
+            v-model="form.name"
+            :name="keyNameFieldName"
+            required
+            placeholder="例如：主 Key、备用 Key 1"
+            maxlength="100"
+            autocomplete="off"
+            autocapitalize="none"
+            autocorrect="off"
+            spellcheck="false"
+            data-form-type="other"
+            data-lpignore="true"
+            data-1p-ignore="true"
+          />
         </div>
-
         <div>
           <Label :for="apiKeyInputId">API 密钥 {{ editingKey ? '' : '*' }}</Label>
           <Input
@@ -83,148 +62,161 @@
             v-else-if="editingKey"
             class="text-xs text-muted-foreground mt-1"
           >
-            留空表示不修改，输入新值则覆盖
+            留空表示不修改
           </p>
         </div>
+      </div>
 
+      <!-- 备注 -->
+      <div>
+        <Label for="note">备注</Label>
+        <Input
+          id="note"
+          v-model="form.note"
+          placeholder="可选的备注信息"
+        />
+      </div>
+
+      <!-- API 格式选择 -->
+      <div v-if="sortedApiFormats.length > 0">
+        <Label class="mb-1.5 block">支持的 API 格式 *</Label>
+        <div class="grid grid-cols-2 gap-2">
+          <div
+            v-for="format in sortedApiFormats"
+            :key="format"
+            class="flex items-center justify-between rounded-md border px-2 py-1.5 transition-colors cursor-pointer"
+            :class="form.api_formats.includes(format)
+              ? 'bg-primary/5 border-primary/30'
+              : 'bg-muted/30 border-border hover:border-muted-foreground/30'"
+            @click="toggleApiFormat(format)"
+          >
+            <div class="flex items-center gap-1.5 min-w-0">
+              <span
+                class="w-4 h-4 rounded border flex items-center justify-center text-xs shrink-0"
+                :class="form.api_formats.includes(format)
+                  ? 'bg-primary border-primary text-primary-foreground'
+                  : 'border-muted-foreground/30'"
+              >
+                <span v-if="form.api_formats.includes(format)">✓</span>
+              </span>
+              <span
+                class="text-sm whitespace-nowrap"
+                :class="form.api_formats.includes(format) ? 'text-primary' : 'text-muted-foreground'"
+              >{{ API_FORMAT_LABELS[format] || format }}</span>
+            </div>
+            <div
+              class="flex items-center shrink-0 ml-2 text-xs text-muted-foreground gap-1"
+              @click.stop
+            >
+              <span>×</span>
+              <input
+                :value="form.rate_multipliers[format] ?? ''"
+                type="number"
+                step="0.01"
+                min="0.01"
+                placeholder="1"
+                class="w-9 bg-transparent text-right outline-none [appearance:textfield] [&::-webkit-outer-spin-button]:appearance-none [&::-webkit-inner-spin-button]:appearance-none"
+                :class="form.api_formats.includes(format) ? 'text-primary' : 'text-muted-foreground'"
+                title="成本倍率"
+                @input="(e) => updateRateMultiplier(format, (e.target as HTMLInputElement).value)"
+              >
+            </div>
+          </div>
+        </div>
+      </div>
+
+      <!-- 配置项 -->
+      <div class="grid grid-cols-4 gap-3">
         <div>
-          <Label for="note">备注</Label>
+          <Label
+            for="internal_priority"
+            class="text-xs"
+          >优先级</Label>
           <Input
-            id="note"
-            v-model="form.note"
-            placeholder="可选的备注信息"
+            id="internal_priority"
+            v-model.number="form.internal_priority"
+            type="number"
+            min="0"
+            class="h-8"
           />
+          <p class="text-xs text-muted-foreground mt-0.5">
+            越小越优先
+          </p>
+        </div>
+        <div>
+          <Label
+            for="rpm_limit"
+            class="text-xs"
+          >RPM 限制</Label>
+          <Input
+            id="rpm_limit"
+            :model-value="form.rpm_limit ?? ''"
+            type="number"
+            min="1"
+            max="10000"
+            placeholder="自适应"
+            class="h-8"
+            @update:model-value="(v) => form.rpm_limit = parseNullableNumberInput(v, { min: 1, max: 10000 })"
+          />
+          <p class="text-xs text-muted-foreground mt-0.5">
+            留空自适应
+          </p>
+        </div>
+        <div>
+          <Label
+            for="cache_ttl_minutes"
+            class="text-xs"
+          >缓存 TTL</Label>
+          <Input
+            id="cache_ttl_minutes"
+            :model-value="form.cache_ttl_minutes ?? ''"
+            type="number"
+            min="0"
+            max="60"
+            class="h-8"
+            @update:model-value="(v) => form.cache_ttl_minutes = parseNumberInput(v, { min: 0, max: 60 }) ?? 5"
+          />
+          <p class="text-xs text-muted-foreground mt-0.5">
+            分钟，0禁用
+          </p>
+        </div>
+        <div>
+          <Label
+            for="max_probe_interval_minutes"
+            class="text-xs"
+          >熔断探测</Label>
+          <Input
+            id="max_probe_interval_minutes"
+            :model-value="form.max_probe_interval_minutes ?? ''"
+            type="number"
+            min="2"
+            max="32"
+            placeholder="32"
+            class="h-8"
+            @update:model-value="(v) => form.max_probe_interval_minutes = parseNumberInput(v, { min: 2, max: 32 }) ?? 32"
+          />
+          <p class="text-xs text-muted-foreground mt-0.5">
+            分钟，2-32
+          </p>
         </div>
       </div>
 
-      <!-- 调度与限流 -->
-      <div class="space-y-3">
-        <h3 class="text-sm font-medium border-b pb-2">
-          调度与限流
-        </h3>
-        <div class="grid grid-cols-2 gap-4">
-          <div>
-            <Label for="internal_priority">内部优先级</Label>
-            <Input
-              id="internal_priority"
-              v-model.number="form.internal_priority"
-              type="number"
-              min="0"
-            />
-            <p class="text-xs text-muted-foreground mt-1">
-              数字越小越优先
-            </p>
-          </div>
-          <div>
-            <Label for="max_concurrent">最大并发</Label>
-            <Input
-              id="max_concurrent"
-              :model-value="form.max_concurrent ?? ''"
-              type="number"
-              min="1"
-              placeholder="留空启用自适应"
-              @update:model-value="(v) => form.max_concurrent = parseNumberInput(v)"
-            />
-            <p class="text-xs text-muted-foreground mt-1">
-              留空 = 自适应模式
-            </p>
-          </div>
-        </div>
-
-        <div class="grid grid-cols-3 gap-4">
-          <div>
-            <Label for="rate_limit">速率限制(/分钟)</Label>
-            <Input
-              id="rate_limit"
-              :model-value="form.rate_limit ?? ''"
-              type="number"
-              min="1"
-              @update:model-value="(v) => form.rate_limit = parseNumberInput(v)"
-            />
-          </div>
-          <div>
-            <Label for="daily_limit">每日限制</Label>
-            <Input
-              id="daily_limit"
-              :model-value="form.daily_limit ?? ''"
-              type="number"
-              min="1"
-              @update:model-value="(v) => form.daily_limit = parseNumberInput(v)"
-            />
-          </div>
-          <div>
-            <Label for="monthly_limit">每月限制</Label>
-            <Input
-              id="monthly_limit"
-              :model-value="form.monthly_limit ?? ''"
-              type="number"
-              min="1"
-              @update:model-value="(v) => form.monthly_limit = parseNumberInput(v)"
-            />
-          </div>
-        </div>
-      </div>
-
-      <!-- 缓存与熔断 -->
-      <div class="space-y-3">
-        <h3 class="text-sm font-medium border-b pb-2">
-          缓存与熔断
-        </h3>
-        <div class="grid grid-cols-2 gap-4">
-          <div>
-            <Label for="cache_ttl_minutes">缓存 TTL (分钟)</Label>
-            <Input
-              id="cache_ttl_minutes"
-              :model-value="form.cache_ttl_minutes ?? ''"
-              type="number"
-              min="0"
-              max="60"
-              @update:model-value="(v) => form.cache_ttl_minutes = parseNumberInput(v, { min: 0, max: 60 }) ?? 5"
-            />
-            <p class="text-xs text-muted-foreground mt-1">
-              0 = 禁用缓存亲和性
-            </p>
-          </div>
-          <div>
-            <Label for="max_probe_interval_minutes">熔断探测间隔 (分钟)</Label>
-            <Input
-              id="max_probe_interval_minutes"
-              :model-value="form.max_probe_interval_minutes ?? ''"
-              type="number"
-              min="2"
-              max="32"
-              placeholder="32"
-              @update:model-value="(v) => form.max_probe_interval_minutes = parseNumberInput(v, { min: 2, max: 32 }) ?? 32"
-            />
-            <p class="text-xs text-muted-foreground mt-1">
-              范围 2-32 分钟
-            </p>
-          </div>
-        </div>
-      </div>
-
-      <!-- 能力标签配置 -->
-      <div
-        v-if="availableCapabilities.length > 0"
-        class="space-y-3"
-      >
-        <h3 class="text-sm font-medium border-b pb-2">
-          能力标签
-        </h3>
-        <div class="flex flex-wrap gap-2">
-          <label
+      <!-- 能力标签 -->
+      <div v-if="availableCapabilities.length > 0">
+        <Label class="text-xs mb-1.5 block">能力标签</Label>
+        <div class="flex flex-wrap gap-1.5">
+          <button
             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"
+            type="button"
+            class="inline-flex items-center gap-1.5 px-2.5 py-1 rounded-md border text-sm transition-colors"
+            :class="form.capabilities[cap.name]
+              ? 'bg-primary/10 border-primary/50 text-primary'
+              : 'bg-card border-border hover:bg-muted/50 text-muted-foreground'"
+            @click="form.capabilities[cap.name] = !form.capabilities[cap.name]"
           >
-            <input
-              type="checkbox"
-              :checked="form.capabilities[cap.name] || false"
-              class="rounded"
-              @change="form.capabilities[cap.name] = !form.capabilities[cap.name]"
-            >
-            <span>{{ cap.display_name }}</span>
-          </label>
+            {{ cap.display_name }}
+          </button>
         </div>
       </div>
     </form>
@@ -240,25 +232,27 @@
         :disabled="saving"
         @click="handleSave"
       >
-        {{ saving ? '保存中...' : '保存' }}
+        {{ saving ? (isEditMode ? '保存中...' : '添加中...') : (isEditMode ? '保存' : '添加') }}
       </Button>
     </template>
   </Dialog>
 </template>
 
 <script setup lang="ts">
-import { ref, computed, onMounted } from 'vue'
+import { ref, computed, onMounted, watch } from 'vue'
 import { Dialog, Button, Input, Label } from '@/components/ui'
 import { Key, SquarePen } from 'lucide-vue-next'
 import { useToast } from '@/composables/useToast'
 import { useFormDialog } from '@/composables/useFormDialog'
 import { parseApiError } from '@/utils/errorParser'
-import { parseNumberInput } from '@/utils/form'
+import { parseNumberInput, parseNullableNumberInput } from '@/utils/form'
 import { log } from '@/utils/logger'
 import {
-  addEndpointKey,
-  updateEndpointKey,
+  addProviderKey,
+  updateProviderKey,
   getAllCapabilities,
+  API_FORMAT_LABELS,
+  sortApiFormats,
   type EndpointAPIKey,
   type EndpointAPIKeyUpdate,
   type ProviderEndpoint,
@@ -270,6 +264,7 @@ const props = defineProps<{
   endpoint: ProviderEndpoint | null
   editingKey: EndpointAPIKey | null
   providerId: string | null
+  availableApiFormats: string[]  // Provider 支持的所有 API 格式
 }>()
 
 const emit = defineEmits<{
@@ -279,6 +274,9 @@ const emit = defineEmits<{
 
 const { success, error: showError } = useToast()
 
+// 排序后的可用 API 格式列表
+const sortedApiFormats = computed(() => sortApiFormats(props.availableApiFormats))
+
 const isOpen = computed(() => props.open)
 const saving = ref(false)
 const formNonce = ref(createFieldNonce())
@@ -297,12 +295,10 @@ const availableCapabilities = ref<CapabilityDefinition[]>([])
 const form = ref({
   name: '',
   api_key: '',
-  rate_multiplier: 1.0,
-  internal_priority: 50,
-  max_concurrent: undefined as number | undefined,
-  rate_limit: undefined as number | undefined,
-  daily_limit: undefined as number | undefined,
-  monthly_limit: undefined as number | undefined,
+  api_formats: [] as string[],  // 支持的 API 格式列表
+  rate_multipliers: {} as Record<string, number>,  // 按 API 格式的成本倍率
+  internal_priority: 10,
+  rpm_limit: undefined as number | null | undefined,  // RPM 限制（null=自适应，undefined=保持原值）
   cache_ttl_minutes: 5,
   max_probe_interval_minutes: 32,
   note: '',
@@ -323,6 +319,43 @@ onMounted(() => {
   loadCapabilities()
 })
 
+// API 格式切换
+function toggleApiFormat(format: string) {
+  const index = form.value.api_formats.indexOf(format)
+  if (index === -1) {
+    // 添加格式
+    form.value.api_formats.push(format)
+  } else {
+    // 移除格式前检查：至少保留一个格式
+    if (form.value.api_formats.length <= 1) {
+      showError('至少需要选择一个 API 格式', '验证失败')
+      return
+    }
+    // 移除格式，但保留倍率配置（用户可能只是临时取消）
+    form.value.api_formats.splice(index, 1)
+  }
+}
+
+// 更新指定格式的成本倍率
+function updateRateMultiplier(format: string, value: string | number) {
+  // 使用对象替换以确保 Vue 3 响应性
+  const newMultipliers = { ...form.value.rate_multipliers }
+
+  if (value === '' || value === null || value === undefined) {
+    // 清空时删除该格式的配置（使用默认倍率）
+    delete newMultipliers[format]
+  } else {
+    const numValue = typeof value === 'string' ? parseFloat(value) : value
+    // 限制倍率范围：0.01 - 100
+    if (!isNaN(numValue) && numValue >= 0.01 && numValue <= 100) {
+      newMultipliers[format] = numValue
+    }
+  }
+
+  // 替换整个对象以触发响应式更新
+  form.value.rate_multipliers = newMultipliers
+}
+
 // API 密钥输入框样式计算
 function getApiKeyInputClass(): string {
   const classes = []
@@ -349,8 +382,8 @@ const apiKeyError = computed(() => {
   }
 
   // 如果输入了值，检查长度
-  if (apiKey.length < 10) {
-    return 'API 密钥至少需要 10 个字符'
+  if (apiKey.length < 3) {
+    return 'API 密钥至少需要 3 个字符'
   }
 
   return ''
@@ -363,12 +396,10 @@ function resetForm() {
   form.value = {
     name: '',
     api_key: '',
-    rate_multiplier: 1.0,
-    internal_priority: 50,
-    max_concurrent: undefined,
-    rate_limit: undefined,
-    daily_limit: undefined,
-    monthly_limit: undefined,
+    api_formats: [],  // 默认不选中任何格式
+    rate_multipliers: {},
+    internal_priority: 10,
+    rpm_limit: undefined,
     cache_ttl_minutes: 5,
     max_probe_interval_minutes: 32,
     note: '',
@@ -377,6 +408,14 @@ function resetForm() {
   }
 }
 
+// 添加成功后清除部分字段以便继续添加
+function clearForNextAdd() {
+  formNonce.value = createFieldNonce()
+  apiKeyFocused.value = false
+  form.value.name = ''
+  form.value.api_key = ''
+}
+
 // 加载密钥数据（编辑模式）
 function loadKeyData() {
   if (!props.editingKey) return
@@ -385,13 +424,13 @@ function loadKeyData() {
   form.value = {
     name: props.editingKey.name,
     api_key: '',
-    rate_multiplier: props.editingKey.rate_multiplier || 1.0,
-    internal_priority: props.editingKey.internal_priority ?? 50,
+    api_formats: props.editingKey.api_formats?.length > 0
+      ? [...props.editingKey.api_formats]
+      : [],  // 编辑模式下保持原有选择，不默认全选
+    rate_multipliers: { ...(props.editingKey.rate_multipliers || {}) },
+    internal_priority: props.editingKey.internal_priority ?? 10,
     // 保留原始的 null/undefined 状态，null 表示自适应模式
-    max_concurrent: props.editingKey.max_concurrent ?? undefined,
-    rate_limit: props.editingKey.rate_limit ?? undefined,
-    daily_limit: props.editingKey.daily_limit ?? undefined,
-    monthly_limit: props.editingKey.monthly_limit ?? undefined,
+    rpm_limit: props.editingKey.rpm_limit ?? undefined,
     cache_ttl_minutes: props.editingKey.cache_ttl_minutes ?? 5,
     max_probe_interval_minutes: props.editingKey.max_probe_interval_minutes ?? 32,
     note: props.editingKey.note || '',
@@ -415,7 +454,11 @@ function createFieldNonce(): string {
 }
 
 async function handleSave() {
-  if (!props.endpoint) return
+  // 必须有 providerId
+  if (!props.providerId) {
+    showError('无法保存：缺少提供商信息', '错误')
+    return
+  }
 
   // 提交前验证
   if (apiKeyError.value) {
@@ -429,6 +472,12 @@ async function handleSave() {
     return
   }
 
+  // 验证至少选择一个 API 格式
+  if (form.value.api_formats.length === 0) {
+    showError('请至少选择一个 API 格式', '验证失败')
+    return
+  }
+
   // 过滤出有效的能力配置（只包含值为 true 的）
   const activeCapabilities: Record<string, boolean> = {}
   for (const [key, value] of Object.entries(form.value.capabilities)) {
@@ -440,21 +489,27 @@ async function handleSave() {
 
   saving.value = true
   try {
+    // 准备 rate_multipliers 数据：只保留已选中格式的倍率配置
+    const filteredMultipliers: Record<string, number> = {}
+    for (const format of form.value.api_formats) {
+      if (form.value.rate_multipliers[format] !== undefined) {
+        filteredMultipliers[format] = form.value.rate_multipliers[format]
+      }
+    }
+    const rateMultipliersData = Object.keys(filteredMultipliers).length > 0
+      ? filteredMultipliers
+      : null
+
     if (props.editingKey) {
       // 更新模式
-      // 注意：max_concurrent 需要显式发送 null 来切换到自适应模式
-      // undefined 会在 JSON 中被忽略，所以用 null 表示"清空/自适应"
+      // 注意：rpm_limit 使用 null 表示自适应模式
+      // undefined 表示"保持原值不变"（会在 JSON 序列化时被忽略）
       const updateData: EndpointAPIKeyUpdate = {
+        api_formats: form.value.api_formats,
         name: form.value.name,
-        rate_multiplier: form.value.rate_multiplier,
+        rate_multipliers: rateMultipliersData,
         internal_priority: form.value.internal_priority,
-        // 显式使用 null 表示自适应模式，这样后端能区分"未提供"和"设置为 null"
-        // 注意：只有 max_concurrent 需要这种处理，因为它有"自适应模式"的概念
-        // 其他限制字段（rate_limit 等）不支持"清空"操作，undefined 会被 JSON 忽略即不更新
-        max_concurrent: form.value.max_concurrent === undefined ? null : form.value.max_concurrent,
-        rate_limit: form.value.rate_limit,
-        daily_limit: form.value.daily_limit,
-        monthly_limit: form.value.monthly_limit,
+        rpm_limit: form.value.rpm_limit,
         cache_ttl_minutes: form.value.cache_ttl_minutes,
         max_probe_interval_minutes: form.value.max_probe_interval_minutes,
         note: form.value.note,
@@ -466,26 +521,27 @@ async function handleSave() {
         updateData.api_key = form.value.api_key
       }
 
-      await updateEndpointKey(props.editingKey.id, updateData)
+      await updateProviderKey(props.editingKey.id, updateData)
       success('密钥已更新', '成功')
     } else {
-      // 新增
-      await addEndpointKey(props.endpoint.id, {
-        endpoint_id: props.endpoint.id,
+      // 新增模式
+      await addProviderKey(props.providerId, {
+        api_formats: form.value.api_formats,
         api_key: form.value.api_key,
         name: form.value.name,
-        rate_multiplier: form.value.rate_multiplier,
+        rate_multipliers: rateMultipliersData,
         internal_priority: form.value.internal_priority,
-        max_concurrent: form.value.max_concurrent,
-        rate_limit: form.value.rate_limit,
-        daily_limit: form.value.daily_limit,
-        monthly_limit: form.value.monthly_limit,
+        rpm_limit: form.value.rpm_limit,
         cache_ttl_minutes: form.value.cache_ttl_minutes,
         max_probe_interval_minutes: form.value.max_probe_interval_minutes,
         note: form.value.note,
         capabilities: capabilitiesData || undefined
       })
       success('密钥已添加', '成功')
+      // 添加模式：不关闭对话框，只清除名称和密钥以便继续添加
+      emit('saved')
+      clearForNextAdd()
+      return
     }
 
     emit('saved')

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

@@ -377,6 +377,7 @@ const props = defineProps<{
   providerApiFormats: string[]
   models: Model[]
   editingGroup?: AliasGroup | null
+  preselectedModelId?: string | null
 }>()
 
 const emit = defineEmits<{
@@ -500,7 +501,7 @@ function initForm() {
     }
   } else {
     formData.value = {
-      modelId: '',
+      modelId: props.preselectedModelId || '',
       apiFormats: [],
       aliases: []
     }

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

@@ -95,7 +95,7 @@
 
               <!-- 提供商信息 -->
               <div class="flex-1 min-w-0 flex items-center gap-2">
-                <span class="font-medium text-sm truncate">{{ provider.display_name }}</span>
+                <span class="font-medium text-sm truncate">{{ provider.name }}</span>
                 <Badge
                   v-if="!provider.is_active"
                   variant="secondary"
@@ -262,17 +262,17 @@
                       <div class="shrink-0 flex items-center gap-3">
                         <!-- 健康度 -->
                         <div
-                          v-if="key.success_rate !== null"
+                          v-if="key.health_score != null"
                           class="text-xs text-right"
                         >
                           <div
                             class="font-medium tabular-nums"
                             :class="[
-                              key.success_rate >= 0.95 ? 'text-green-600' :
-                              key.success_rate >= 0.8 ? 'text-yellow-600' : 'text-red-500'
+                              key.health_score >= 0.95 ? 'text-green-600' :
+                              key.health_score >= 0.5 ? 'text-yellow-600' : 'text-red-500'
                             ]"
                           >
-                            {{ (key.success_rate * 100).toFixed(0) }}%
+                            {{ ((key.health_score || 0) * 100).toFixed(0) }}%
                           </div>
                           <div class="text-[10px] text-muted-foreground opacity-70">
                             {{ key.request_count }} reqs
@@ -319,19 +319,6 @@
           <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"
@@ -345,6 +332,32 @@
               >
                 缓存亲和
               </button>
+              <button
+                type="button"
+                class="px-2 py-1 text-xs font-medium rounded transition-all"
+                :class="[
+                  schedulingMode === 'load_balance'
+                    ? 'bg-primary text-primary-foreground shadow-sm'
+                    : 'text-muted-foreground hover:text-foreground hover:bg-muted/50'
+                ]"
+                title="同优先级内随机轮换，不考虑缓存"
+                @click="schedulingMode = 'load_balance'"
+              >
+                负载均衡
+              </button>
+              <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>
             </div>
           </div>
         </div>
@@ -382,7 +395,7 @@ import { Dialog } from '@/components/ui'
 import Button from '@/components/ui/button.vue'
 import Badge from '@/components/ui/badge.vue'
 import { useToast } from '@/composables/useToast'
-import { updateProvider, updateEndpointKey } from '@/api/endpoints'
+import { updateProvider, updateProviderKey } from '@/api/endpoints'
 import type { ProviderWithEndpointsSummary } from '@/api/endpoints'
 import { adminApi } from '@/api/admin'
 
@@ -400,6 +413,7 @@ interface KeyWithMeta {
   endpoint_base_url: string
   api_format: string
   capabilities: string[]
+  health_score: number | null
   success_rate: number | null
   avg_response_time_ms: number | null
   request_count: number
@@ -444,7 +458,7 @@ const saving = ref(false)
 const editingKeyPriority = ref<Record<string, string | null>>({})  // format -> keyId
 
 // 调度模式状态
-const schedulingMode = ref<'fixed_order' | 'cache_affinity'>('cache_affinity')
+const schedulingMode = ref<'fixed_order' | 'load_balance' | 'cache_affinity'>('cache_affinity')
 
 // 可用的 API 格式
 const availableFormats = computed(() => {
@@ -477,7 +491,11 @@ async function loadCurrentPriorityMode() {
     activeMainTab.value = currentMode === 'global_key' ? 'key' : 'provider'
 
     const currentSchedulingMode = schedulingResponse.value || 'cache_affinity'
-    schedulingMode.value = currentSchedulingMode === 'fixed_order' ? 'fixed_order' : 'cache_affinity'
+    if (currentSchedulingMode === 'fixed_order' || currentSchedulingMode === 'load_balance' || currentSchedulingMode === 'cache_affinity') {
+      schedulingMode.value = currentSchedulingMode
+    } else {
+      schedulingMode.value = 'cache_affinity'
+    }
   } catch {
     activeMainTab.value = 'provider'
     schedulingMode.value = 'cache_affinity'
@@ -678,7 +696,7 @@ async function save() {
       const keys = keysByFormat.value[format]
       keys.forEach((key) => {
         // 使用用户设置的 priority 值，相同 priority 会做负载均衡
-        keyUpdates.push(updateEndpointKey(key.id, { global_priority: key.priority }))
+        keyUpdates.push(updateProviderKey(key.id, { global_priority: key.priority }))
       })
     }
 

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

@@ -4,47 +4,29 @@
     :title="isEditMode ? '编辑提供商' : '添加提供商'"
     :description="isEditMode ? '更新提供商配置。API 端点和密钥需在详情页面单独管理。' : '创建新的提供商配置。创建后可以为其添加 API 端点和密钥。'"
     :icon="isEditMode ? SquarePen : Server"
-    size="2xl"
+    size="xl"
     @update:model-value="handleDialogUpdate"
   >
     <form
-      class="space-y-6"
+      class="space-y-5"
       @submit.prevent="handleSubmit"
     >
       <!-- 基本信息 -->
-      <div class="space-y-4">
+      <div class="space-y-3">
         <h3 class="text-sm font-medium border-b pb-2">
           基本信息
         </h3>
 
-        <!-- 添加模式显示提供商标识 -->
-        <div
-          v-if="!isEditMode"
-          class="space-y-2"
-        >
-          <Label for="name">提供商标识 *</Label>
-          <Input
-            id="name"
-            v-model="form.name"
-            placeholder="例如: openai-primary"
-            required
-          />
-          <p class="text-xs text-muted-foreground">
-            唯一ID，创建后不可修改
-          </p>
-        </div>
-
         <div class="grid grid-cols-2 gap-4">
-          <div class="space-y-2">
-            <Label for="display_name">显示名称 *</Label>
+          <div class="space-y-1.5">
+            <Label for="name">名称 *</Label>
             <Input
-              id="display_name"
-              v-model="form.display_name"
+              id="name"
+              v-model="form.name"
               placeholder="例如: OpenAI 主账号"
-              required
             />
           </div>
-          <div class="space-y-2">
+          <div class="space-y-1.5">
             <Label for="website">主站链接</Label>
             <Input
               id="website"
@@ -55,24 +37,28 @@
           </div>
         </div>
 
-        <div class="space-y-2">
+        <div class="space-y-1.5">
           <Label for="description">描述</Label>
-          <Textarea
+          <Input
             id="description"
             v-model="form.description"
             placeholder="提供商描述（可选）"
-            rows="2"
           />
         </div>
       </div>
 
-      <!-- 计费与限流 -->
-      <div class="space-y-4">
-        <h3 class="text-sm font-medium border-b pb-2">
-          计费与限流
-        </h3>
+      <!-- 计费与限流 / 请求配置 -->
+      <div class="space-y-3">
         <div class="grid grid-cols-2 gap-4">
-          <div class="space-y-2">
+          <h3 class="text-sm font-medium border-b pb-2">
+            计费与限流
+          </h3>
+          <h3 class="text-sm font-medium border-b pb-2">
+            请求配置
+          </h3>
+        </div>
+        <div class="grid grid-cols-2 gap-4">
+          <div class="space-y-1.5">
             <Label>计费类型</Label>
             <Select
               v-model="form.billing_type"
@@ -82,27 +68,35 @@
                 <SelectValue />
               </SelectTrigger>
               <SelectContent>
-                <SelectItem value="monthly_quota">
-                  月卡额度
-                </SelectItem>
-                <SelectItem value="pay_as_you_go">
-                  按量付费
-                </SelectItem>
-                <SelectItem value="free_tier">
-                  免费套餐
-                </SelectItem>
+                <SelectItem value="monthly_quota">月卡额度</SelectItem>
+                <SelectItem value="pay_as_you_go">按量付费</SelectItem>
+                <SelectItem value="free_tier">免费套餐</SelectItem>
               </SelectContent>
             </Select>
           </div>
-          <div class="space-y-2">
-            <Label>RPM 限制</Label>
-            <Input
-              :model-value="form.rpm_limit ?? ''"
-              type="number"
-              min="0"
-              placeholder="不限制请留空"
-              @update:model-value="(v) => form.rpm_limit = parseNumberInput(v)"
-            />
+          <div class="grid grid-cols-2 gap-4">
+            <div class="space-y-1.5">
+              <Label>超时时间 (秒)</Label>
+              <Input
+                :model-value="form.timeout ?? ''"
+                type="number"
+                min="1"
+                max="600"
+                placeholder="默认 300"
+                @update:model-value="(v) => form.timeout = parseNumberInput(v)"
+              />
+            </div>
+            <div class="space-y-1.5">
+              <Label>最大重试次数</Label>
+              <Input
+                :model-value="form.max_retries ?? ''"
+                type="number"
+                min="0"
+                max="10"
+                placeholder="默认 2"
+                @update:model-value="(v) => form.max_retries = parseNumberInput(v)"
+              />
+            </div>
           </div>
         </div>
 
@@ -111,52 +105,94 @@
           v-if="form.billing_type === 'monthly_quota'"
           class="grid grid-cols-2 gap-4 p-3 border rounded-lg bg-muted/50"
         >
-          <div class="space-y-2">
+          <div class="space-y-1.5">
             <Label class="text-xs">周期额度 (USD)</Label>
             <Input
               :model-value="form.monthly_quota_usd ?? ''"
               type="number"
               step="0.01"
               min="0"
-              class="h-9"
               @update:model-value="(v) => form.monthly_quota_usd = parseNumberInput(v, { allowFloat: true })"
             />
           </div>
-          <div class="space-y-2">
+          <div class="space-y-1.5">
             <Label class="text-xs">重置周期 (天)</Label>
             <Input
               :model-value="form.quota_reset_day ?? ''"
               type="number"
               min="1"
               max="365"
-              class="h-9"
               @update:model-value="(v) => form.quota_reset_day = parseNumberInput(v) ?? 30"
             />
           </div>
-          <div class="space-y-2">
+          <div class="space-y-1.5">
             <Label class="text-xs">
-              周期开始时间
-              <span class="text-red-500">*</span>
+              周期开始时间 <span class="text-red-500">*</span>
             </Label>
             <Input
               v-model="form.quota_last_reset_at"
               type="datetime-local"
-              class="h-9"
             />
-            <p class="text-xs text-muted-foreground">
-              系统会自动统计从该时间点开始的使用量
-            </p>
           </div>
-          <div class="space-y-2">
+          <div class="space-y-1.5">
             <Label class="text-xs">过期时间</Label>
             <Input
               v-model="form.quota_expires_at"
               type="datetime-local"
-              class="h-9"
             />
-            <p class="text-xs text-muted-foreground">
-              留空表示永久有效
-            </p>
+          </div>
+        </div>
+      </div>
+
+      <!-- 代理配置 -->
+      <div class="space-y-3">
+        <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
+              :model-value="form.proxy_enabled"
+              @update:model-value="(v: boolean) => form.proxy_enabled = v"
+            />
+            <span class="text-sm text-muted-foreground">启用代理</span>
+          </div>
+        </div>
+        <div
+          v-if="form.proxy_enabled"
+          class="grid grid-cols-2 gap-4 p-3 border rounded-lg bg-muted/50"
+        >
+          <div class="space-y-1.5">
+            <Label class="text-xs">代理地址 *</Label>
+            <Input
+              v-model="form.proxy_url"
+              placeholder="http://proxy:port 或 socks5://proxy:port"
+            />
+          </div>
+          <div class="grid grid-cols-2 gap-3">
+            <div class="space-y-1.5">
+              <Label class="text-xs">用户名</Label>
+              <Input
+                v-model="form.proxy_username"
+                placeholder="可选"
+                autocomplete="off"
+                data-form-type="other"
+                data-lpignore="true"
+                data-1p-ignore="true"
+              />
+            </div>
+            <div class="space-y-1.5">
+              <Label class="text-xs">密码</Label>
+              <Input
+                v-model="form.proxy_password"
+                type="password"
+                placeholder="可选"
+                autocomplete="new-password"
+                data-form-type="other"
+                data-lpignore="true"
+                data-1p-ignore="true"
+              />
+            </div>
           </div>
         </div>
       </div>
@@ -172,7 +208,7 @@
         取消
       </Button>
       <Button
-        :disabled="loading || !form.display_name || (!isEditMode && !form.name)"
+        :disabled="loading || !form.name"
         @click="handleSubmit"
       >
         {{ loading ? (isEditMode ? '保存中...' : '创建中...') : (isEditMode ? '保存' : '创建') }}
@@ -187,13 +223,13 @@ import {
   Dialog,
   Button,
   Input,
-  Textarea,
   Label,
   Select,
   SelectTrigger,
   SelectValue,
   SelectContent,
   SelectItem,
+  Switch,
 } from '@/components/ui'
 import { Server, SquarePen } from 'lucide-vue-next'
 import { useToast } from '@/composables/useToast'
@@ -223,7 +259,6 @@ const internalOpen = computed(() => props.modelValue)
 // 表单数据
 const form = ref({
   name: '',
-  display_name: '',
   description: '',
   website: '',
   // 计费配置
@@ -232,19 +267,25 @@ const form = ref({
   quota_reset_day: 30,
   quota_last_reset_at: '',  // 周期开始时间
   quota_expires_at: '',
-  rpm_limit: undefined as string | number | undefined,
   provider_priority: 999,
   // 状态配置
   is_active: true,
   rate_limit: undefined as number | undefined,
   concurrent_limit: undefined as number | undefined,
+  // 请求配置
+  timeout: undefined as number | undefined,
+  max_retries: undefined as number | undefined,
+  // 代理配置（扁平化便于表单绑定）
+  proxy_enabled: false,
+  proxy_url: '',
+  proxy_username: '',
+  proxy_password: '',
 })
 
 // 重置表单
 function resetForm() {
   form.value = {
     name: '',
-    display_name: '',
     description: '',
     website: '',
     billing_type: 'pay_as_you_go',
@@ -252,11 +293,18 @@ function resetForm() {
     quota_reset_day: 30,
     quota_last_reset_at: '',
     quota_expires_at: '',
-    rpm_limit: undefined,
     provider_priority: 999,
     is_active: true,
     rate_limit: undefined,
     concurrent_limit: undefined,
+    // 请求配置
+    timeout: undefined,
+    max_retries: undefined,
+    // 代理配置
+    proxy_enabled: false,
+    proxy_url: '',
+    proxy_username: '',
+    proxy_password: '',
   }
 }
 
@@ -264,9 +312,9 @@ function resetForm() {
 function loadProviderData() {
   if (!props.provider) return
 
+  const proxy = props.provider.proxy
   form.value = {
     name: props.provider.name,
-    display_name: props.provider.display_name,
     description: props.provider.description || '',
     website: props.provider.website || '',
     billing_type: (props.provider.billing_type as 'monthly_quota' | 'pay_as_you_go' | 'free_tier') || 'pay_as_you_go',
@@ -276,11 +324,18 @@ function loadProviderData() {
       new Date(props.provider.quota_last_reset_at).toISOString().slice(0, 16) : '',
     quota_expires_at: props.provider.quota_expires_at ?
       new Date(props.provider.quota_expires_at).toISOString().slice(0, 16) : '',
-    rpm_limit: props.provider.rpm_limit ?? undefined,
     provider_priority: props.provider.provider_priority || 999,
     is_active: props.provider.is_active,
     rate_limit: undefined,
     concurrent_limit: undefined,
+    // 请求配置
+    timeout: props.provider.timeout ?? undefined,
+    max_retries: props.provider.max_retries ?? undefined,
+    // 代理配置
+    proxy_enabled: proxy?.enabled ?? false,
+    proxy_url: proxy?.url || '',
+    proxy_username: proxy?.username || '',
+    proxy_password: proxy?.password || '',
   }
 }
 
@@ -302,17 +357,37 @@ const handleSubmit = async () => {
     return
   }
 
+  // 启用代理时必须填写代理地址
+  if (form.value.proxy_enabled && !form.value.proxy_url) {
+    showError('启用代理时必须填写代理地址', '验证失败')
+    return
+  }
+
   loading.value = true
   try {
+    // 构建代理配置
+    const proxy = form.value.proxy_enabled ? {
+      url: form.value.proxy_url,
+      username: form.value.proxy_username || undefined,
+      password: form.value.proxy_password || undefined,
+      enabled: true,
+    } : null
+
     const payload = {
-      ...form.value,
-      rpm_limit:
-        form.value.rpm_limit === undefined || form.value.rpm_limit === ''
-          ? null
-          : Number(form.value.rpm_limit),
-      // 空字符串时不发送
+      name: form.value.name,
+      description: form.value.description || undefined,
+      website: form.value.website || undefined,
+      billing_type: form.value.billing_type,
+      monthly_quota_usd: form.value.monthly_quota_usd,
+      quota_reset_day: form.value.quota_reset_day,
       quota_last_reset_at: form.value.quota_last_reset_at || undefined,
       quota_expires_at: form.value.quota_expires_at || undefined,
+      provider_priority: form.value.provider_priority,
+      is_active: form.value.is_active,
+      // 请求配置
+      timeout: form.value.timeout ?? undefined,
+      max_retries: form.value.max_retries ?? undefined,
+      proxy,
     }
 
     if (isEditMode.value && props.provider) {

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

@@ -2,6 +2,7 @@ export { default as ProviderFormDialog } from './ProviderFormDialog.vue'
 export { default as EndpointFormDialog } from './EndpointFormDialog.vue'
 export { default as KeyFormDialog } from './KeyFormDialog.vue'
 export { default as KeyAllowedModelsDialog } from './KeyAllowedModelsDialog.vue'
+export { default as KeyAllowedModelsEditDialog } from './KeyAllowedModelsEditDialog.vue'
 export { default as PriorityManagementDialog } from './PriorityManagementDialog.vue'
 export { default as ProviderModelFormDialog } from './ProviderModelFormDialog.vue'
 export { default as ProviderDetailDrawer } from './ProviderDetailDrawer.vue'

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

@@ -168,6 +168,7 @@
     :provider-api-formats="providerApiFormats"
     :models="models"
     :editing-group="editingGroup"
+    :preselected-model-id="preselectedModelId"
     @saved="onDialogSaved"
   />
 
@@ -219,6 +220,7 @@ const deleteConfirmOpen = ref(false)
 const editingGroup = ref<AliasGroup | null>(null)
 const deletingGroup = ref<AliasGroup | null>(null)
 const testingMapping = ref<string | null>(null)
+const preselectedModelId = ref<string | null>(null)
 
 // 列表展开状态
 const expandedAliasGroups = ref<Set<string>>(new Set())
@@ -311,12 +313,21 @@ function toggleAliasGroupExpand(groupKey: string) {
 // 打开添加对话框
 function openAddDialog() {
   editingGroup.value = null
+  preselectedModelId.value = null
+  dialogOpen.value = true
+}
+
+// 打开添加对话框并预选模型（供外部调用）
+function openAddDialogForModel(modelId: string) {
+  editingGroup.value = null
+  preselectedModelId.value = modelId
   dialogOpen.value = true
 }
 
 // 编辑分组
 function editGroup(group: AliasGroup) {
   editingGroup.value = group
+  preselectedModelId.value = null
   dialogOpen.value = true
 }
 
@@ -416,8 +427,9 @@ onMounted(() => {
   }
 })
 
-// 暴露给父组件，用于检测是否有弹窗打开
+// 暴露给父组件
 defineExpose({
-  dialogOpen: computed(() => dialogOpen.value || deleteConfirmOpen.value)
+  dialogOpen: computed(() => dialogOpen.value || deleteConfirmOpen.value),
+  openAddDialogForModel
 })
 </script>

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

@@ -34,16 +34,13 @@
       <table class="w-full text-sm table-fixed">
         <thead class="bg-muted/50 text-xs uppercase tracking-wide text-muted-foreground">
           <tr>
-            <th class="text-left px-4 py-3 font-semibold w-[40%]">
+            <th class="text-left px-4 py-3 font-semibold w-[45%]">
               模型
             </th>
-            <th class="text-left px-4 py-3 font-semibold w-[15%]">
-              能力
-            </th>
-            <th class="text-left px-4 py-3 font-semibold w-[25%]">
+            <th class="text-left px-4 py-3 font-semibold w-[30%]">
               价格 ($/M)
             </th>
-            <th class="text-center px-4 py-3 font-semibold w-[20%]">
+            <th class="text-center px-4 py-3 font-semibold w-[25%]">
               操作
             </th>
           </tr>
@@ -80,42 +77,6 @@
                 </div>
               </div>
             </td>
-            <td class="align-top px-4 py-3">
-              <div
-                v-if="hasAnyCapability(model)"
-                class="grid grid-cols-3 gap-1 w-fit"
-              >
-                <Zap
-                  v-if="model.effective_supports_streaming ?? model.supports_streaming"
-                  class="w-4 h-4 text-muted-foreground"
-                  title="流式输出"
-                />
-                <Image
-                  v-if="model.effective_supports_image_generation ?? model.supports_image_generation"
-                  class="w-4 h-4 text-muted-foreground"
-                  title="图像生成"
-                />
-                <Eye
-                  v-if="model.effective_supports_vision ?? model.supports_vision"
-                  class="w-4 h-4 text-muted-foreground"
-                  title="视觉理解"
-                />
-                <Wrench
-                  v-if="model.effective_supports_function_calling ?? model.supports_function_calling"
-                  class="w-4 h-4 text-muted-foreground"
-                  title="工具调用"
-                />
-                <Brain
-                  v-if="model.effective_supports_extended_thinking ?? model.supports_extended_thinking"
-                  class="w-4 h-4 text-muted-foreground"
-                  title="深度思考"
-                />
-              </div>
-              <span
-                v-else
-                class="text-xs text-muted-foreground"
-              >—</span>
-            </td>
             <td class="align-top px-4 py-3 text-xs whitespace-nowrap">
               <div
                 class="grid gap-1"
@@ -148,7 +109,7 @@
                     ${{ formatPrice(model.effective_price_per_request ?? model.price_per_request) }}/次
                   </span>
                 </template>
-                <!-- 无计费配置 -->
+                <!-- 无计<EFBFBD><EFBFBD>配置 -->
                 <template v-if="!hasTokenPricing(model) && !hasRequestPricing(model)">
                   <span class="text-muted-foreground">—</span>
                 </template>
@@ -156,6 +117,15 @@
             </td>
             <td class="align-top px-4 py-3">
               <div class="flex justify-center gap-1.5">
+                <Button
+                  variant="ghost"
+                  size="icon"
+                  class="h-8 w-8"
+                  title="添加映射"
+                  @click="addMapping(model)"
+                >
+                  <Link class="w-3.5 h-3.5" />
+                </Button>
                 <Button
                   variant="ghost"
                   size="icon"
@@ -178,7 +148,7 @@
                 <Button
                   variant="ghost"
                   size="icon"
-                  class="h-8 w-8 text-destructive hover:text-destructive"
+                  class="h-8 w-8 hover:text-destructive"
                   title="删除"
                   @click="deleteModel(model)"
                 >
@@ -209,7 +179,7 @@
 
 <script setup lang="ts">
 import { ref, computed, onMounted } from 'vue'
-import { Box, Edit, Trash2, Layers, Eye, Wrench, Zap, Brain, Power, Copy, Image } from 'lucide-vue-next'
+import { Box, Edit, Trash2, Layers, Power, Copy, Link } from 'lucide-vue-next'
 import Card from '@/components/ui/card.vue'
 import Button from '@/components/ui/button.vue'
 import { useToast } from '@/composables/useToast'
@@ -225,6 +195,7 @@ const emit = defineEmits<{
   'editModel': [model: Model]
   'deleteModel': [model: Model]
   'batchAssign': []
+  'addMapping': [model: Model]
 }>()
 
 const { error: showError, success: showSuccess } = useToast()
@@ -276,17 +247,6 @@ function formatPrice(price: number | null | undefined): string {
   return price.toFixed(4)
 }
 
-// 检查模型是否有任何能力
-function hasAnyCapability(model: Model): boolean {
-  return !!(
-    (model.effective_supports_vision ?? model.supports_vision) ||
-    (model.effective_supports_function_calling ?? model.supports_function_calling) ||
-    (model.effective_supports_streaming ?? model.supports_streaming) ||
-    (model.effective_supports_extended_thinking ?? model.supports_extended_thinking) ||
-    (model.effective_supports_image_generation ?? model.supports_image_generation)
-  )
-}
-
 // 检查是否有按 Token 计费
 function hasTokenPricing(model: Model): boolean {
   const inputPrice = model.effective_input_price
@@ -345,7 +305,7 @@ function getStatusTitle(model: Model): string {
   return '活跃但不可用'
 }
 
-// 编辑模型
+// <EFBFBD><EFBFBD>辑模型
 function editModel(model: Model) {
   emit('editModel', model)
 }
@@ -355,6 +315,11 @@ function deleteModel(model: Model) {
   emit('deleteModel', model)
 }
 
+// 添加映射
+function addMapping(model: Model) {
+  emit('addMapping', model)
+}
+
 // 打开批量关联对话框
 function openBatchAssignDialog() {
   emit('batchAssign')

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

@@ -3,6 +3,7 @@
  */
 import { ref } from 'vue'
 import { adminApi } from '@/api/admin'
+import { parseUpstreamModelError } from '@/utils/errorParser'
 import type { UpstreamModel } from '@/api/endpoints/types'
 
 // 扩展类型，包含可能的额外字段
@@ -63,10 +64,14 @@ export function useUpstreamModelsCache() {
           })
           return { models: response.data.models }
         } else {
-          return { models: [], error: response.data?.error || '获取上游模型失败' }
+          // 使用友好的错误解析
+          const rawError = response.data?.error || '获取上游模型失败'
+          return { models: [], error: parseUpstreamModelError(rawError) }
         }
       } catch (err: any) {
-        return { models: [], error: err.response?.data?.detail || '获取上游模型失败' }
+        // 使用友好的错误解析
+        const rawError = err.response?.data?.detail || err.message || '获取上游模型失败'
+        return { models: [], error: parseUpstreamModelError(rawError) }
       } finally {
         loadingMap.value.set(providerId, false)
         pendingRequests.delete(providerId)

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

@@ -289,14 +289,14 @@
                 />
               </div>
 
-              <!-- 错误信息卡片 -->
+              <!-- 响应客户端错误卡片 -->
               <Card
                 v-if="detail.error_message"
                 class="border-red-200 dark:border-red-800"
               >
                 <div class="p-4">
                   <h4 class="text-sm font-semibold text-red-600 dark:text-red-400 mb-2">
-                    错误信息
+                    响应客户端错误
                   </h4>
                   <div class="bg-red-50 dark:bg-red-900/20 rounded-lg p-3">
                     <p class="text-sm text-red-800 dark:text-red-300">
@@ -431,7 +431,7 @@
 
                     <TabsContent value="response-headers">
                       <JsonContent
-                        :data="detail.response_headers"
+                        :data="actualResponseHeaders"
                         :view-mode="viewMode"
                         :expand-depth="currentExpandDepth"
                         :is-dark="isDark"
@@ -614,6 +614,25 @@ const tabs = [
   { name: 'metadata', label: '元数据' },
 ]
 
+// 判断数据是否有实际内容（非空对象/数组）
+function hasContent(data: unknown): boolean {
+  if (data === null || data === undefined) return false
+  if (typeof data === 'object') {
+    return Object.keys(data as object).length > 0
+  }
+  return true
+}
+
+// 获取实际的响应头（优先 client_response_headers，回退到 response_headers）
+const actualResponseHeaders = computed(() => {
+  if (!detail.value) return null
+  // 优先返回客户端响应头，如果没有则回退到提供商响应头
+  if (hasContent(detail.value.client_response_headers)) {
+    return detail.value.client_response_headers
+  }
+  return detail.value.response_headers
+})
+
 // 根据实际数据决定显示哪些 Tab
 const visibleTabs = computed(() => {
   if (!detail.value) return []
@@ -621,15 +640,15 @@ const visibleTabs = computed(() => {
   return tabs.filter(tab => {
     switch (tab.name) {
       case 'request-headers':
-        return detail.value!.request_headers && Object.keys(detail.value!.request_headers).length > 0
+        return hasContent(detail.value!.request_headers)
       case 'request-body':
-        return detail.value!.request_body !== null && detail.value!.request_body !== undefined
+        return hasContent(detail.value!.request_body)
       case 'response-headers':
-        return detail.value!.response_headers && Object.keys(detail.value!.response_headers).length > 0
+        return hasContent(actualResponseHeaders.value)
       case 'response-body':
-        return detail.value!.response_body !== null && detail.value!.response_body !== undefined
+        return hasContent(detail.value!.response_body)
       case 'metadata':
-        return detail.value!.metadata && Object.keys(detail.value!.metadata).length > 0
+        return hasContent(detail.value!.metadata)
       default:
         return false
     }
@@ -775,7 +794,7 @@ function copyJsonToClipboard(tabName: string) {
       data = detail.value.request_body
       break
     case 'response-headers':
-      data = detail.value.response_headers
+      data = actualResponseHeaders.value
       break
     case 'response-body':
       data = detail.value.response_body

frontend/src/features/users/components/UserFormDialog.vue

@@ -252,7 +252,7 @@
                     @click.stop
                     @change="toggleSelection('allowed_providers', provider.id)"
                   >
-                  <span class="text-sm">{{ provider.display_name || provider.name }}</span>
+                  <span class="text-sm">{{ provider.name }}</span>
                 </div>
                 <div
                   v-if="providers.length === 0"
@@ -273,8 +273,8 @@
                 class="w-full h-10 px-3 border rounded-lg bg-background text-left flex items-center justify-between hover:bg-muted/50 transition-colors"
                 @click="endpointDropdownOpen = !endpointDropdownOpen"
               >
-                <span :class="form.allowed_endpoints.length ? 'text-foreground' : 'text-muted-foreground'">
-                  {{ form.allowed_endpoints.length ? `已选择 ${form.allowed_endpoints.length} 个` : '全部可用' }}
+                <span :class="form.allowed_api_formats.length ? 'text-foreground' : 'text-muted-foreground'">
+                  {{ form.allowed_api_formats.length ? `已选择 ${form.allowed_api_formats.length} 个` : '全部可用' }}
                 </span>
                 <ChevronDown
                   class="h-4 w-4 text-muted-foreground transition-transform"
@@ -294,14 +294,14 @@
                   v-for="format in apiFormats"
                   :key="format.value"
                   class="flex items-center gap-2 px-3 py-2 hover:bg-muted/50 cursor-pointer"
-                  @click="toggleSelection('allowed_endpoints', format.value)"
+                  @click="toggleSelection('allowed_api_formats', format.value)"
                 >
                   <input
                     type="checkbox"
-                    :checked="form.allowed_endpoints.includes(format.value)"
+                    :checked="form.allowed_api_formats.includes(format.value)"
                     class="h-4 w-4 rounded border-gray-300 cursor-pointer"
                     @click.stop
-                    @change="toggleSelection('allowed_endpoints', format.value)"
+                    @change="toggleSelection('allowed_api_formats', format.value)"
                   >
                   <span class="text-sm">{{ format.label }}</span>
                 </div>
@@ -374,7 +374,7 @@ export interface UserFormData {
   role: 'admin' | 'user'
   is_active?: boolean
   allowed_providers?: string[] | null
-  allowed_endpoints?: string[] | null
+  allowed_api_formats?: string[] | null
   allowed_models?: string[] | null
 }
 
@@ -414,7 +414,7 @@ const form = ref({
   unlimited: false,
   is_active: true,
   allowed_providers: [] as string[],
-  allowed_endpoints: [] as string[],
+  allowed_api_formats: [] as string[],
   allowed_models: [] as string[]
 })
 
@@ -435,7 +435,7 @@ function resetForm() {
     unlimited: false,
     is_active: true,
     allowed_providers: [],
-    allowed_endpoints: [],
+    allowed_api_formats: [],
     allowed_models: []
   }
 }
@@ -454,7 +454,7 @@ function loadUserData() {
     unlimited: props.user.quota_usd == null,
     is_active: props.user.is_active ?? true,
     allowed_providers: props.user.allowed_providers || [],
-    allowed_endpoints: props.user.allowed_endpoints || [],
+    allowed_api_formats: props.user.allowed_api_formats || [],
     allowed_models: props.user.allowed_models || []
   }
 }
@@ -495,7 +495,7 @@ async function loadAccessControlOptions() {
 }
 
 // 切换选择
-function toggleSelection(field: 'allowed_providers' | 'allowed_endpoints' | 'allowed_models', value: string) {
+function toggleSelection(field: 'allowed_providers' | 'allowed_api_formats' | 'allowed_models', value: string) {
   const arr = form.value[field]
   const index = arr.indexOf(value)
   if (index === -1) {
@@ -520,7 +520,7 @@ async function handleSubmit() {
       quota_usd: form.value.unlimited ? null : form.value.quota,
       role: form.value.role,
       allowed_providers: form.value.allowed_providers.length > 0 ? form.value.allowed_providers : null,
-      allowed_endpoints: form.value.allowed_endpoints.length > 0 ? form.value.allowed_endpoints : null,
+      allowed_api_formats: form.value.allowed_api_formats.length > 0 ? form.value.allowed_api_formats : null,
       allowed_models: form.value.allowed_models.length > 0 ? form.value.allowed_models : null
     }
 

frontend/src/layouts/MainLayout.vue

@@ -295,6 +295,15 @@
     </template>
 
     <RouterView />
+
+    <!-- 更新提示弹窗 -->
+    <UpdateDialog
+      v-if="updateInfo"
+      v-model="showUpdateDialog"
+      :current-version="updateInfo.current_version"
+      :latest-version="updateInfo.latest_version || ''"
+      :release-url="updateInfo.release_url"
+    />
   </AppShell>
 </template>
 
@@ -304,14 +313,17 @@ import { useRoute, useRouter } from 'vue-router'
 import { useAuthStore } from '@/stores/auth'
 import { useDarkMode } from '@/composables/useDarkMode'
 import { isDemoMode } from '@/config/demo'
+import { adminApi, type CheckUpdateResponse } from '@/api/admin'
 import Button from '@/components/ui/button.vue'
 import AppShell from '@/components/layout/AppShell.vue'
 import SidebarNav from '@/components/layout/SidebarNav.vue'
 import HeaderLogo from '@/components/HeaderLogo.vue'
+import UpdateDialog from '@/components/common/UpdateDialog.vue'
 import {
   Home,
   Users,
   Key,
+  KeyRound,
   BarChart3,
   Cog,
   Settings,
@@ -344,17 +356,67 @@ const showAuthError = ref(false)
 const mobileMenuOpen = ref(false)
 let authCheckInterval: number | null = null
 
+// 更新检查相关
+const showUpdateDialog = ref(false)
+const updateInfo = ref<CheckUpdateResponse | null>(null)
+
 // 路由变化时自动关闭移动端菜单
 watch(() => route.path, () => {
   mobileMenuOpen.value = false
 })
 
+// 检查是否应该显示更新提示
+function shouldShowUpdatePrompt(latestVersion: string): boolean {
+  const ignoreKey = 'aether_update_ignore'
+  const ignoreData = localStorage.getItem(ignoreKey)
+  if (!ignoreData) return true
+
+  try {
+    const { version, until } = JSON.parse(ignoreData)
+    // 如果忽略的是同一版本且未过期，则不显示
+    if (version === latestVersion && Date.now() < until) {
+      return false
+    }
+  } catch {
+    // 解析失败，显示提示
+  }
+  return true
+}
+
+// 检查更新
+async function checkForUpdate() {
+  // 只有管理员才检查更新
+  if (authStore.user?.role !== 'admin') return
+
+  // 同一会话内只检查一次
+  const sessionKey = 'aether_update_checked'
+  if (sessionStorage.getItem(sessionKey)) return
+  sessionStorage.setItem(sessionKey, '1')
+
+  try {
+    const result = await adminApi.checkUpdate()
+    if (result.has_update && result.latest_version) {
+      if (shouldShowUpdatePrompt(result.latest_version)) {
+        updateInfo.value = result
+        showUpdateDialog.value = true
+      }
+    }
+  } catch {
+    // 静默失败，不影响用户体验
+  }
+}
+
 onMounted(() => {
   authCheckInterval = setInterval(() => {
     if (authStore.user && !authStore.token) {
       showAuthError.value = true
     }
   }, 5000)
+
+  // 延迟检查更新，避免影响页面加载
+  setTimeout(() => {
+    checkForUpdate()
+  }, 2000)
 })
 
 onUnmounted(() => {
@@ -398,6 +460,7 @@ const navigation = computed(() => {
       items: [
         { name: '模型目录', href: '/dashboard/models', icon: Box },
         { name: 'API 密钥', href: '/dashboard/api-keys', icon: Key },
+        { name: '访问令牌', href: '/dashboard/management-tokens', icon: KeyRound },
       ]
     },
     {
@@ -423,6 +486,7 @@ const navigation = computed(() => {
         { name: '提供商', href: '/admin/providers', icon: FolderTree },
         { name: '模型管理', href: '/admin/models', icon: Layers },
         { name: '独立密钥', href: '/admin/keys', icon: Key },
+        { name: '访问令牌', href: '/admin/management-tokens', icon: KeyRound },
         { name: '使用记录', href: '/admin/usage', icon: BarChart3 },
       ]
     },

frontend/src/mocks/data.ts

@@ -22,7 +22,7 @@ export const MOCK_ADMIN_USER: User = {
   used_usd: 156.78,
   total_usd: 1234.56,
   allowed_providers: null,
-  allowed_endpoints: null,
+  allowed_api_formats: null,
   allowed_models: null,
   created_at: '2024-01-01T00:00:00Z',
   last_login_at: new Date().toISOString()
@@ -38,7 +38,7 @@ export const MOCK_NORMAL_USER: User = {
   used_usd: 45.32,
   total_usd: 245.32,
   allowed_providers: null,
-  allowed_endpoints: null,
+  allowed_api_formats: null,
   allowed_models: null,
   created_at: '2024-06-01T00:00:00Z',
   last_login_at: new Date().toISOString()
@@ -274,7 +274,7 @@ export const MOCK_ALL_USERS: AdminUser[] = [
     used_usd: 156.78,
     total_usd: 1234.56,
     allowed_providers: null,
-    allowed_endpoints: null,
+    allowed_api_formats: null,
     allowed_models: null,
     created_at: '2024-01-01T00:00:00Z'
   },
@@ -288,7 +288,7 @@ export const MOCK_ALL_USERS: AdminUser[] = [
     used_usd: 45.32,
     total_usd: 245.32,
     allowed_providers: null,
-    allowed_endpoints: null,
+    allowed_api_formats: null,
     allowed_models: null,
     created_at: '2024-06-01T00:00:00Z'
   },
@@ -302,7 +302,7 @@ export const MOCK_ALL_USERS: AdminUser[] = [
     used_usd: 23.45,
     total_usd: 123.45,
     allowed_providers: null,
-    allowed_endpoints: null,
+    allowed_api_formats: null,
     allowed_models: null,
     created_at: '2024-03-15T00:00:00Z'
   },
@@ -316,7 +316,7 @@ export const MOCK_ALL_USERS: AdminUser[] = [
     used_usd: 89.12,
     total_usd: 589.12,
     allowed_providers: null,
-    allowed_endpoints: null,
+    allowed_api_formats: null,
     allowed_models: null,
     created_at: '2024-02-20T00:00:00Z'
   },
@@ -330,7 +330,7 @@ export const MOCK_ALL_USERS: AdminUser[] = [
     used_usd: 30.00,
     total_usd: 30.00,
     allowed_providers: null,
-    allowed_endpoints: null,
+    allowed_api_formats: null,
     allowed_models: null,
     created_at: '2024-04-10T00:00:00Z'
   }
@@ -424,8 +424,7 @@ export const MOCK_ADMIN_API_KEYS: AdminApiKeysResponse = {
 export const MOCK_PROVIDERS: ProviderWithEndpointsSummary[] = [
   {
     id: 'provider-001',
-    name: 'duck_coding_free',
-    display_name: 'DuckCodingFree',
+    name: 'DuckCodingFree',
     description: '',
     website: 'https://duckcoding.com',
     provider_priority: 1,
@@ -451,8 +450,7 @@ export const MOCK_PROVIDERS: ProviderWithEndpointsSummary[] = [
   },
   {
     id: 'provider-002',
-    name: 'open_claude_code',
-    display_name: 'OpenClaudeCode',
+    name: 'OpenClaudeCode',
     description: '',
     website: 'https://www.openclaudecode.cn',
     provider_priority: 2,
@@ -477,8 +475,7 @@ export const MOCK_PROVIDERS: ProviderWithEndpointsSummary[] = [
   },
   {
     id: 'provider-003',
-    name: '88_code',
-    display_name: '88Code',
+    name: '88Code',
     description: '',
     website: 'https://www.88code.org/',
     provider_priority: 3,
@@ -503,8 +500,7 @@ export const MOCK_PROVIDERS: ProviderWithEndpointsSummary[] = [
   },
   {
     id: 'provider-004',
-    name: 'ikun_code',
-    display_name: 'IKunCode',
+    name: 'IKunCode',
     description: '',
     website: 'https://api.ikuncode.cc',
     provider_priority: 4,
@@ -531,8 +527,7 @@ export const MOCK_PROVIDERS: ProviderWithEndpointsSummary[] = [
   },
   {
     id: 'provider-005',
-    name: 'duck_coding',
-    display_name: 'DuckCoding',
+    name: 'DuckCoding',
     description: '',
     website: 'https://duckcoding.com',
     provider_priority: 5,
@@ -561,8 +556,7 @@ export const MOCK_PROVIDERS: ProviderWithEndpointsSummary[] = [
   },
   {
     id: 'provider-006',
-    name: 'privnode',
-    display_name: 'Privnode',
+    name: 'Privnode',
     description: '',
     website: 'https://privnode.com',
     provider_priority: 6,
@@ -584,8 +578,7 @@ export const MOCK_PROVIDERS: ProviderWithEndpointsSummary[] = [
   },
   {
     id: 'provider-007',
-    name: 'undying_api',
-    display_name: 'UndyingAPI',
+    name: 'UndyingAPI',
     description: '',
     website: 'https://vip.undyingapi.com',
     provider_priority: 7,

frontend/src/mocks/handler.ts

@@ -418,16 +418,16 @@ const MOCK_ALIASES = [
 
 // Mock Endpoint Keys
 const MOCK_ENDPOINT_KEYS = [
-  { id: 'ekey-001', endpoint_id: 'ep-001', api_key_masked: 'sk-ant...abc1', name: 'Primary Key', rate_multiplier: 1.0, internal_priority: 1, health_score: 98, consecutive_failures: 0, request_count: 5000, success_count: 4950, error_count: 50, success_rate: 99, avg_response_time_ms: 1200, is_active: true, created_at: '2024-01-01T00:00:00Z', updated_at: new Date().toISOString() },
-  { id: 'ekey-002', endpoint_id: 'ep-001', api_key_masked: 'sk-ant...def2', name: 'Backup Key', rate_multiplier: 1.0, internal_priority: 2, health_score: 95, consecutive_failures: 1, request_count: 2000, success_count: 1950, error_count: 50, success_rate: 97.5, avg_response_time_ms: 1350, is_active: true, created_at: '2024-02-01T00:00:00Z', updated_at: new Date().toISOString() },
-  { id: 'ekey-003', endpoint_id: 'ep-002', api_key_masked: 'sk-oai...ghi3', name: 'OpenAI Main', rate_multiplier: 1.0, internal_priority: 1, health_score: 97, consecutive_failures: 0, request_count: 3500, success_count: 3450, error_count: 50, success_rate: 98.6, avg_response_time_ms: 900, is_active: true, created_at: '2024-01-15T00:00:00Z', updated_at: new Date().toISOString() }
+  { id: 'ekey-001', provider_id: 'provider-001', api_formats: ['CLAUDE'], api_key_masked: 'sk-ant...abc1', name: 'Primary Key', rate_multiplier: 1.0, internal_priority: 1, health_score: 0.98, consecutive_failures: 0, request_count: 5000, success_count: 4950, error_count: 50, success_rate: 0.99, avg_response_time_ms: 1200, cache_ttl_minutes: 5, max_probe_interval_minutes: 32, is_active: true, created_at: '2024-01-01T00:00:00Z', updated_at: new Date().toISOString() },
+  { id: 'ekey-002', provider_id: 'provider-001', api_formats: ['CLAUDE'], api_key_masked: 'sk-ant...def2', name: 'Backup Key', rate_multiplier: 1.0, internal_priority: 2, health_score: 0.95, consecutive_failures: 1, request_count: 2000, success_count: 1950, error_count: 50, success_rate: 0.975, avg_response_time_ms: 1350, cache_ttl_minutes: 5, max_probe_interval_minutes: 32, is_active: true, created_at: '2024-02-01T00:00:00Z', updated_at: new Date().toISOString() },
+  { id: 'ekey-003', provider_id: 'provider-002', api_formats: ['OPENAI'], api_key_masked: 'sk-oai...ghi3', name: 'OpenAI Main', rate_multiplier: 1.0, internal_priority: 1, health_score: 0.97, consecutive_failures: 0, request_count: 3500, success_count: 3450, error_count: 50, success_rate: 0.986, avg_response_time_ms: 900, cache_ttl_minutes: 5, max_probe_interval_minutes: 32, is_active: true, created_at: '2024-01-15T00:00:00Z', updated_at: new Date().toISOString() }
 ]
 
 // Mock Endpoints
 const MOCK_ENDPOINTS = [
-  { id: 'ep-001', provider_id: 'provider-001', provider_name: 'anthropic', api_format: 'claude', base_url: 'https://api.anthropic.com', auth_type: 'bearer', timeout: 120, max_retries: 3, priority: 100, weight: 100, health_score: 98, consecutive_failures: 0, is_active: true, total_keys: 2, active_keys: 2, created_at: '2024-01-01T00:00:00Z', updated_at: new Date().toISOString() },
-  { id: 'ep-002', provider_id: 'provider-002', provider_name: 'openai', api_format: 'openai', base_url: 'https://api.openai.com', auth_type: 'bearer', timeout: 60, max_retries: 3, priority: 90, weight: 100, health_score: 97, consecutive_failures: 0, is_active: true, total_keys: 1, active_keys: 1, created_at: '2024-01-01T00:00:00Z', updated_at: new Date().toISOString() },
-  { id: 'ep-003', provider_id: 'provider-003', provider_name: 'google', api_format: 'gemini', base_url: 'https://generativelanguage.googleapis.com', auth_type: 'api_key', timeout: 60, max_retries: 3, priority: 80, weight: 100, health_score: 96, consecutive_failures: 0, is_active: true, total_keys: 1, active_keys: 1, created_at: '2024-01-15T00:00:00Z', updated_at: new Date().toISOString() }
+  { id: 'ep-001', provider_id: 'provider-001', provider_name: 'anthropic', api_format: 'CLAUDE', base_url: 'https://api.anthropic.com', timeout: 300, max_retries: 2, is_active: true, total_keys: 2, active_keys: 2, created_at: '2024-01-01T00:00:00Z', updated_at: new Date().toISOString() },
+  { id: 'ep-002', provider_id: 'provider-002', provider_name: 'openai', api_format: 'OPENAI', base_url: 'https://api.openai.com', timeout: 60, max_retries: 2, is_active: true, total_keys: 1, active_keys: 1, created_at: '2024-01-01T00:00:00Z', updated_at: new Date().toISOString() },
+  { id: 'ep-003', provider_id: 'provider-003', provider_name: 'google', api_format: 'GEMINI', base_url: 'https://generativelanguage.googleapis.com', timeout: 60, max_retries: 2, is_active: true, total_keys: 1, active_keys: 1, created_at: '2024-01-15T00:00:00Z', updated_at: new Date().toISOString() }
 ]
 
 // Mock 能力定义
@@ -581,7 +581,6 @@ const mockHandlers: Record<string, (config: AxiosRequestConfig) => Promise<Axios
     return createMockResponse(MOCK_PROVIDERS.map(p => ({
       id: p.id,
       name: p.name,
-      display_name: p.display_name,
       is_active: p.is_active
     })))
   },
@@ -690,7 +689,7 @@ const mockHandlers: Record<string, (config: AxiosRequestConfig) => Promise<Axios
       used_usd: 0,
       total_usd: 0,
       allowed_providers: null,
-      allowed_endpoints: null,
+      allowed_api_formats: null,
       allowed_models: null,
       created_at: new Date().toISOString()
     }
@@ -1222,13 +1221,8 @@ function generateMockEndpointsForProvider(providerId: string) {
       base_url: format.includes('CLAUDE') ? 'https://api.anthropic.com' :
                 format.includes('OPENAI') ? 'https://api.openai.com' :
                 'https://generativelanguage.googleapis.com',
-      auth_type: format.includes('GEMINI') ? 'api_key' : 'bearer',
-      timeout: 120,
-      max_retries: 3,
-      priority: 100 - index * 10,
-      weight: 100,
-      health_score: healthDetail?.health_score ?? 1.0,
-      consecutive_failures: healthDetail?.health_score && healthDetail.health_score < 0.7 ? 2 : 0,
+      timeout: 300,
+      max_retries: 2,
       is_active: healthDetail?.is_active ?? true,
       total_keys: Math.ceil(Math.random() * 3) + 1,
       active_keys: Math.ceil(Math.random() * 2) + 1,
@@ -1238,11 +1232,16 @@ function generateMockEndpointsForProvider(providerId: string) {
   })
 }
 
-// 为 endpoint 生成 keys
-function generateMockKeysForEndpoint(endpointId: string, count: number = 2) {
+// 为 provider 生成 keys（Key 归属 Provider，通过 api_formats 关联）
+const PROVIDER_KEYS_CACHE: Record<string, any[]> = {}
+function generateMockKeysForProvider(providerId: string, count: number = 2) {
+  const provider = MOCK_PROVIDERS.find(p => p.id === providerId)
+  const formats = provider?.api_formats || []
+
   return Array.from({ length: count }, (_, i) => ({
-    id: `key-${endpointId}-${i + 1}`,
-    endpoint_id: endpointId,
+    id: `key-${providerId}-${i + 1}`,
+    provider_id: providerId,
+    api_formats: i === 0 ? formats : formats.slice(0, 1),
     api_key_masked: `sk-***...${Math.random().toString(36).substring(2, 6)}`,
     name: i === 0 ? 'Primary Key' : `Backup Key ${i}`,
     rate_multiplier: 1.0,
@@ -1254,6 +1253,8 @@ function generateMockKeysForEndpoint(endpointId: string, count: number = 2) {
     error_count: Math.floor(Math.random() * 100),
     success_rate: 0.95 + Math.random() * 0.04,  // 0.95-0.99
     avg_response_time_ms: 800 + Math.floor(Math.random() * 600),
+    cache_ttl_minutes: 5,
+    max_probe_interval_minutes: 32,
     is_active: true,
     created_at: '2024-01-01T00:00:00Z',
     updated_at: new Date().toISOString()
@@ -1463,29 +1464,63 @@ registerDynamicRoute('PUT', '/api/admin/endpoints/:endpointId', async (config, p
 registerDynamicRoute('DELETE', '/api/admin/endpoints/:endpointId', async (_config, _params) => {
   await delay()
   requireAdmin()
-  return createMockResponse({ message: '删除成功（演示模式）' })
+  return createMockResponse({ message: '删除成功（演示模式）', affected_keys_count: 0 })
 })
 
-// Endpoint Keys 列表
-registerDynamicRoute('GET', '/api/admin/endpoints/:endpointId/keys', async (_config, params) => {
+// Provider Keys 列表
+registerDynamicRoute('GET', '/api/admin/endpoints/providers/:providerId/keys', async (_config, params) => {
   await delay()
   requireAdmin()
-  const keys = generateMockKeysForEndpoint(params.endpointId, 2)
-  return createMockResponse(keys)
+  if (!PROVIDER_KEYS_CACHE[params.providerId]) {
+    PROVIDER_KEYS_CACHE[params.providerId] = generateMockKeysForProvider(params.providerId, 2)
+  }
+  return createMockResponse(PROVIDER_KEYS_CACHE[params.providerId])
 })
 
-// 创建 Key
-registerDynamicRoute('POST', '/api/admin/endpoints/:endpointId/keys', async (config, params) => {
+// 为 Provider 创建 Key
+registerDynamicRoute('POST', '/api/admin/endpoints/providers/:providerId/keys', async (config, params) => {
   await delay()
   requireAdmin()
   const body = JSON.parse(config.data || '{}')
-  return createMockResponse({
+  const apiKeyPlain = body.api_key || 'sk-demo'
+  const masked = apiKeyPlain.length >= 12
+    ? `${apiKeyPlain.slice(0, 8)}***${apiKeyPlain.slice(-4)}`
+    : 'sk-***...demo'
+
+  const newKey = {
     id: `key-demo-${Date.now()}`,
-    endpoint_id: params.endpointId,
-    api_key_masked: 'sk-***...demo',
-    ...body,
-    created_at: new Date().toISOString()
-  })
+    provider_id: params.providerId,
+    api_formats: body.api_formats || [],
+    api_key_masked: masked,
+    api_key_plain: null,
+    name: body.name || 'New Key',
+    note: body.note,
+    rate_multiplier: body.rate_multiplier ?? 1.0,
+    rate_multipliers: body.rate_multipliers ?? null,
+    internal_priority: body.internal_priority ?? 50,
+    global_priority: body.global_priority ?? null,
+    rpm_limit: body.rpm_limit ?? null,
+    allowed_models: body.allowed_models ?? null,
+    capabilities: body.capabilities ?? null,
+    cache_ttl_minutes: body.cache_ttl_minutes ?? 5,
+    max_probe_interval_minutes: body.max_probe_interval_minutes ?? 32,
+    health_score: 1.0,
+    consecutive_failures: 0,
+    request_count: 0,
+    success_count: 0,
+    error_count: 0,
+    success_rate: 0.0,
+    avg_response_time_ms: 0.0,
+    is_active: true,
+    created_at: new Date().toISOString(),
+    updated_at: new Date().toISOString(),
+  }
+
+  if (!PROVIDER_KEYS_CACHE[params.providerId]) {
+    PROVIDER_KEYS_CACHE[params.providerId] = []
+  }
+  PROVIDER_KEYS_CACHE[params.providerId].push(newKey)
+  return createMockResponse(newKey)
 })
 
 // Key 更新
@@ -1503,6 +1538,50 @@ registerDynamicRoute('DELETE', '/api/admin/endpoints/keys/:keyId', async (_confi
   return createMockResponse({ message: '删除成功（演示模式）' })
 })
 
+// Key Reveal
+registerDynamicRoute('GET', '/api/admin/endpoints/keys/:keyId/reveal', async (_config, _params) => {
+  await delay()
+  requireAdmin()
+  return createMockResponse({ api_key: 'sk-demo-reveal' })
+})
+
+// Keys grouped by format
+mockHandlers['GET /api/admin/endpoints/keys/grouped-by-format'] = async () => {
+  await delay()
+  requireAdmin()
+
+  // 确保每个 provider 都有 key 数据
+  for (const provider of MOCK_PROVIDERS) {
+    if (!PROVIDER_KEYS_CACHE[provider.id]) {
+      PROVIDER_KEYS_CACHE[provider.id] = generateMockKeysForProvider(provider.id, 2)
+    }
+  }
+
+  const grouped: Record<string, any[]> = {}
+  for (const provider of MOCK_PROVIDERS) {
+    const endpoints = generateMockEndpointsForProvider(provider.id)
+    const baseUrlByFormat = Object.fromEntries(endpoints.map(e => [e.api_format, e.base_url]))
+    const keys = PROVIDER_KEYS_CACHE[provider.id] || []
+    for (const key of keys) {
+      const formats: string[] = key.api_formats || []
+      for (const fmt of formats) {
+        if (!grouped[fmt]) grouped[fmt] = []
+        grouped[fmt].push({
+          ...key,
+          api_format: fmt,
+          provider_name: provider.name,
+          endpoint_base_url: baseUrlByFormat[fmt],
+          global_priority: key.global_priority ?? null,
+          circuit_breaker_open: false,
+          capabilities: [],
+        })
+      }
+    }
+  }
+
+  return createMockResponse(grouped)
+}
+
 // Provider Models 列表
 registerDynamicRoute('GET', '/api/admin/providers/:providerId/models', async (_config, params) => {
   await delay()

frontend/src/router/index.ts

@@ -34,6 +34,11 @@ const routes: RouteRecordRaw[] = [
         name: 'MyApiKeys',
         component: () => importWithRetry(() => import('@/views/user/MyApiKeys.vue'))
       },
+      {
+        path: 'management-tokens',
+        name: 'ManagementTokens',
+        component: () => importWithRetry(() => import('@/views/user/ManagementTokens.vue'))
+      },
       {
         path: 'announcements',
         name: 'Announcements',
@@ -81,6 +86,11 @@ const routes: RouteRecordRaw[] = [
         name: 'ApiKeys',
         component: () => importWithRetry(() => import('@/views/admin/ApiKeys.vue'))
       },
+      {
+        path: 'management-tokens',
+        name: 'AdminManagementTokens',
+        component: () => importWithRetry(() => import('@/views/user/ManagementTokens.vue'))
+      },
       {
         path: 'providers',
         name: 'ProviderManagement',

frontend/src/utils/errorParser.ts

@@ -20,7 +20,7 @@ interface ValidationError {
 const fieldNameMap: Record<string, string> = {
   'api_key': 'API 密钥',
   'priority': '优先级',
-  'max_concurrent': '最大并发',
+  'rpm_limit': 'RPM 限制',
   'rate_limit': '速率限制',
   'daily_limit': '每日限制',
   'monthly_limit': '每月限制',
@@ -44,7 +44,6 @@ const fieldNameMap: Record<string, string> = {
   'monthly_quota_usd': '月度配额',
   'quota_reset_day': '配额重置日',
   'quota_expires_at': '配额过期时间',
-  'rpm_limit': 'RPM 限制',
   'cache_ttl_minutes': '缓存 TTL',
   'max_probe_interval_minutes': '最大探测间隔',
 }
@@ -54,7 +53,7 @@ const fieldNameMap: Record<string, string> = {
  */
 const errorTypeMap: Record<string, (error: ValidationError) => string> = {
   'string_too_short': (error) => {
-    const minLength = error.ctx?.min_length || 10
+    const minLength = error.ctx?.min_length || 3
     return `长度不能少于 ${minLength} 个字符`
   },
   'string_too_long': (error) => {
@@ -151,11 +150,18 @@ export function parseApiError(err: unknown, defaultMessage: string = '操作失
     return '无法连接到服务器，请检查网络连接'
   }
 
-  const detail = err.response?.data?.detail
+  const data = err.response?.data
+
+  // 1. 处理 {error: {type, message}} 格式（ProxyException 返回格式）
+  if (data?.error?.message) {
+    return data.error.message
+  }
+
+  const detail = data?.detail
 
   // 如果没有 detail 字段
   if (!detail) {
-    return err.response?.data?.message || err.message || defaultMessage
+    return data?.message || err.message || defaultMessage
   }
 
   // 1. 处理 Pydantic 验证错误（数组格式）
@@ -244,3 +250,115 @@ export function parseTestModelError(result: {
 
   return errorMsg
 }
+
+/**
+ * 解析上游模型查询错误信息
+ * 将后端返回的原始错误信息（如 "HTTP 401: {json...}"）转换为友好的错误提示
+ * @param error 错误字符串，格式可能是 "HTTP {status}: {json_body}" 或其他
+ * @returns 友好的错误信息
+ */
+export function parseUpstreamModelError(error: string): string {
+  if (!error) return '获取上游模型失败'
+
+  // 匹配 "HTTP {status}: {body}" 格式
+  const httpMatch = error.match(/^HTTP\s+(\d+):\s*(.*)$/s)
+  if (httpMatch) {
+    const status = parseInt(httpMatch[1], 10)
+    const body = httpMatch[2]
+
+    // 根据状态码生成友好消息
+    let friendlyMsg = ''
+    if (status === 401) {
+      friendlyMsg = '密钥无效或已过期'
+    } else if (status === 403) {
+      friendlyMsg = '密钥权限不足'
+    } else if (status === 404) {
+      friendlyMsg = '模型列表接口不存在'
+    } else if (status === 429) {
+      friendlyMsg = '请求频率过高，请稍后重试'
+    } else if (status >= 500) {
+      friendlyMsg = '上游服务暂时不可用'
+    }
+
+    // 尝试从 JSON body 中提取更详细的错误信息
+    if (body) {
+      try {
+        const parsed = JSON.parse(body)
+        // 常见的错误格式: {error: {message: "..."}} 或 {error: "..."} 或 {message: "..."}
+        let detailMsg = ''
+        if (parsed.error?.message) {
+          detailMsg = parsed.error.message
+        } else if (typeof parsed.error === 'string') {
+          detailMsg = parsed.error
+        } else if (parsed.message) {
+          detailMsg = parsed.message
+        } else if (parsed.detail) {
+          detailMsg = typeof parsed.detail === 'string' ? parsed.detail : JSON.stringify(parsed.detail)
+        }
+
+        // 如果提取到了详细消息，用它来丰富友好消息
+        if (detailMsg) {
+          // 检查是否是 token/认证相关的错误
+          const lowerMsg = detailMsg.toLowerCase()
+          if (lowerMsg.includes('invalid token') || lowerMsg.includes('invalid api key')) {
+            return '密钥无效，请检查密钥是否正确'
+          }
+          if (lowerMsg.includes('expired')) {
+            return '密钥已过期，请更新密钥'
+          }
+          if (lowerMsg.includes('quota') || lowerMsg.includes('exceeded')) {
+            return '配额已用尽或超出限制'
+          }
+          if (lowerMsg.includes('rate limit')) {
+            return '请求频率过高，请稍后重试'
+          }
+          // 没有匹配特定关键词，但有详细信息，使用它作为补充
+          if (friendlyMsg) {
+            const truncated = detailMsg.length > 80 ? detailMsg.substring(0, 80) + '...' : detailMsg
+            return `${friendlyMsg}: ${truncated}`
+          }
+          // 没有友好消息，直接使用详细信息
+          const truncated = detailMsg.length > 100 ? detailMsg.substring(0, 100) + '...' : detailMsg
+          return truncated
+        }
+      } catch {
+        // JSON 解析失败，忽略
+      }
+    }
+
+    // 返回友好消息，附加状态码
+    if (friendlyMsg) {
+      return friendlyMsg
+    }
+    return `请求失败 (HTTP ${status})`
+  }
+
+  // 检查是否是请求错误
+  if (error.startsWith('Request error:')) {
+    const detail = error.replace('Request error:', '').trim()
+    if (detail.toLowerCase().includes('timeout')) {
+      return '请求超时，上游服务响应过慢'
+    }
+    if (detail.toLowerCase().includes('connection')) {
+      return '无法连接到上游服务'
+    }
+    return '网络请求失败'
+  }
+
+  // 检查是否是未知 API 格式
+  if (error.startsWith('Unknown API format:')) {
+    return '不支持的 API 格式'
+  }
+
+  // 如果包含分号，可能是多个错误合并的，取第一个
+  if (error.includes('; ')) {
+    const firstError = error.split('; ')[0]
+    return parseUpstreamModelError(firstError)
+  }
+
+  // 默认返回原始错误（截断过长的部分）
+  if (error.length > 100) {
+    return error.substring(0, 100) + '...'
+  }
+  return error
+}

frontend/src/utils/form.ts

@@ -54,6 +54,57 @@ export function parseNumberInput(
   return result
 }
 
+/**
+ * Parse number input value for nullable fields (like rpm_limit)
+ * Returns `null` when empty (to signal "use adaptive/default mode")
+ * Returns `undefined` when not provided (to signal "keep original value")
+ *
+ * @param value - Input value (string or number)
+ * @param options - Parse options
+ * @returns Parsed number, null (for empty/adaptive), or undefined
+ */
+export function parseNullableNumberInput(
+  value: string | number | null | undefined,
+  options: {
+    allowFloat?: boolean
+    min?: number
+    max?: number
+  } = {}
+): number | null | undefined {
+  const { allowFloat = false, min, max } = options
+
+  // Empty string means "null" (adaptive mode)
+  if (value === '') {
+    return null
+  }
+
+  // null/undefined means "keep original value"
+  if (value === null || value === undefined) {
+    return undefined
+  }
+
+  // Parse the value
+  const num = typeof value === 'string'
+    ? (allowFloat ? parseFloat(value) : parseInt(value, 10))
+    : value
+
+  // Handle NaN - treat as null (adaptive mode)
+  if (isNaN(num)) {
+    return null
+  }
+
+  // Apply min/max constraints
+  let result = num
+  if (min !== undefined && result < min) {
+    result = min
+  }
+  if (max !== undefined && result > max) {
+    result = max
+  }
+
+  return result
+}
+
 /**
  * Create a handler function for number input with specific field
  * Useful for creating inline handlers in templates

frontend/src/views/admin/LdapSettings.vue

@@ -32,7 +32,10 @@
 
         <div class="grid grid-cols-1 md:grid-cols-2 gap-6">
           <div>
-            <Label for="server-url" class="block text-sm font-medium">
+            <Label
+              for="server-url"
+              class="block text-sm font-medium"
+            >
               服务器地址
             </Label>
             <Input
@@ -48,7 +51,10 @@
           </div>
 
           <div>
-            <Label for="bind-dn" class="block text-sm font-medium">
+            <Label
+              for="bind-dn"
+              class="block text-sm font-medium"
+            >
               绑定 DN
             </Label>
             <Input
@@ -64,7 +70,10 @@
           </div>
 
           <div>
-            <Label for="bind-password" class="block text-sm font-medium">
+            <Label
+              for="bind-password"
+              class="block text-sm font-medium"
+            >
               绑定密码
             </Label>
             <div class="relative mt-1">
@@ -80,10 +89,30 @@
                 v-if="hasPassword || ldapConfig.bind_password"
                 type="button"
                 class="absolute right-3 top-1/2 -translate-y-1/2 p-1 rounded-full text-muted-foreground/60 hover:text-muted-foreground hover:bg-muted/50 transition-colors"
-                @click="handleClearPassword"
                 title="清除密码"
+                @click="handleClearPassword"
               >
-                <svg xmlns="http://www.w3.org/2000/svg" width="14" height="14" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><line x1="18" y1="6" x2="6" y2="18"/><line x1="6" y1="6" x2="18" y2="18"/></svg>
+                <svg
+                  xmlns="http://www.w3.org/2000/svg"
+                  width="14"
+                  height="14"
+                  viewBox="0 0 24 24"
+                  fill="none"
+                  stroke="currentColor"
+                  stroke-width="2"
+                  stroke-linecap="round"
+                  stroke-linejoin="round"
+                ><line
+                  x1="18"
+                  y1="6"
+                  x2="6"
+                  y2="18"
+                /><line
+                  x1="6"
+                  y1="6"
+                  x2="18"
+                  y2="18"
+                /></svg>
               </button>
             </div>
             <p class="mt-1 text-xs text-muted-foreground">
@@ -92,7 +121,10 @@
           </div>
 
           <div>
-            <Label for="base-dn" class="block text-sm font-medium">
+            <Label
+              for="base-dn"
+              class="block text-sm font-medium"
+            >
               基础 DN
             </Label>
             <Input
@@ -108,7 +140,10 @@
           </div>
 
           <div>
-            <Label for="user-search-filter" class="block text-sm font-medium">
+            <Label
+              for="user-search-filter"
+              class="block text-sm font-medium"
+            >
               用户搜索过滤器
             </Label>
             <Input
@@ -124,7 +159,10 @@
           </div>
 
           <div>
-            <Label for="username-attr" class="block text-sm font-medium">
+            <Label
+              for="username-attr"
+              class="block text-sm font-medium"
+            >
               用户名属性
             </Label>
             <Input
@@ -140,7 +178,10 @@
           </div>
 
           <div>
-            <Label for="email-attr" class="block text-sm font-medium">
+            <Label
+              for="email-attr"
+              class="block text-sm font-medium"
+            >
               邮箱属性
             </Label>
             <Input
@@ -153,7 +194,10 @@
           </div>
 
           <div>
-            <Label for="display-name-attr" class="block text-sm font-medium">
+            <Label
+              for="display-name-attr"
+              class="block text-sm font-medium"
+            >
               显示名称属性
             </Label>
             <Input
@@ -166,7 +210,10 @@
           </div>
 
           <div>
-            <Label for="connect-timeout" class="block text-sm font-medium">
+            <Label
+              for="connect-timeout"
+              class="block text-sm font-medium"
+            >
               连接超时 (秒)
             </Label>
             <Input

frontend/src/views/admin/ModelManagement.vue

@@ -530,9 +530,6 @@
                     />
                     <div class="flex-1 min-w-0">
                       <p class="font-medium text-sm truncate">
-                        {{ provider.display_name || provider.name }}
-                      </p>
-                      <p class="text-xs text-muted-foreground truncate">
                         {{ provider.name }}
                       </p>
                     </div>
@@ -645,10 +642,7 @@
                     />
                     <div class="flex-1 min-w-0">
                       <p class="font-medium text-sm truncate">
-                        {{ provider.display_name }}
-                      </p>
-                      <p class="text-xs text-muted-foreground truncate">
-                        {{ provider.identifier }}
+                        {{ provider.name }}
                       </p>
                     </div>
                     <Badge
@@ -679,7 +673,7 @@
     <ProviderModelFormDialog
       :open="editProviderDialogOpen"
       :provider-id="editingProvider?.id || ''"
-      :provider-name="editingProvider?.display_name || ''"
+      :provider-name="editingProvider?.name || ''"
       :editing-model="editingProviderModel"
       @update:open="handleEditProviderDialogUpdate"
       @saved="handleEditProviderSaved"
@@ -939,7 +933,7 @@ async function batchAddSelectedProviders() {
       const errorMessages = result.errors
         .map(e => {
           const provider = providerOptions.value.find(p => p.id === e.provider_id)
-          const providerName = provider?.display_name || provider?.name || e.provider_id
+          const providerName = provider?.name || e.provider_id
           return `${providerName}: ${e.error}`
         })
         .join('\n')
@@ -977,7 +971,7 @@ async function batchRemoveSelectedProviders() {
         await deleteModel(providerId, provider.model_id)
         successCount++
       } catch (err: any) {
-        errors.push(`${provider.display_name}: ${parseApiError(err, '删除失败')}`)
+        errors.push(`${provider.name}: ${parseApiError(err, '删除失败')}`)
       }
     }
 
@@ -1088,8 +1082,7 @@ async function loadModelProviders(_globalModelId: string) {
     selectedModelProviders.value = response.providers.map(p => ({
       id: p.provider_id,
       model_id: p.model_id,
-      display_name: p.provider_display_name || p.provider_name,
-      identifier: p.provider_name,
+      name: p.provider_name,
       provider_type: 'API',
       target_model: p.target_model,
       is_active: p.is_active,
@@ -1219,7 +1212,7 @@ async function confirmDeleteProviderImplementation(provider: any) {
   }
 
   const confirmed = await confirmDanger(
-    `确定要删除 ${provider.display_name} 的模型关联吗？\n\n模型: ${provider.target_model}\n\n此操作不可恢复！`,
+    `确定要删除 ${provider.name} 的模型关联吗？\n\n模型: ${provider.target_model}\n\n此操作不可恢复！`,
     '删除关联提供商'
   )
   if (!confirmed) return
@@ -1227,7 +1220,7 @@ async function confirmDeleteProviderImplementation(provider: any) {
   try {
     const { deleteModel } = await import('@/api/endpoints')
     await deleteModel(provider.id, provider.model_id)
-    success(`已删除 ${provider.display_name} 的模型实现`)
+    success(`已删除 ${provider.name} 的模型实现`)
     // 重新加载 Provider 列表
     if (selectedModel.value) {
       await loadModelProviders(selectedModel.value.id)

frontend/src/views/admin/ProviderManagement.vue

@@ -134,10 +134,7 @@
               @click="handleRowClick($event, provider.id)"
             >
               <TableCell class="py-3.5">
-                <div class="flex flex-col gap-0.5">
-                  <span class="text-sm font-medium text-foreground">{{ provider.display_name }}</span>
-                  <span class="text-xs text-muted-foreground/70 font-mono">{{ provider.name }}</span>
-                </div>
+                <span class="text-sm font-medium text-foreground">{{ provider.name }}</span>
               </TableCell>
               <TableCell class="py-3.5">
                 <Badge
@@ -219,17 +216,10 @@
                     >${{ (provider.monthly_used_usd ?? 0).toFixed(2) }}</span> / <span class="font-medium">${{ (provider.monthly_quota_usd ?? 0).toFixed(2) }}</span>
                   </div>
                   <div
-                    v-if="rpmUsage(provider)"
-                    class="flex items-center gap-1"
-                  >
-                    <span class="text-muted-foreground/70">RPM:</span>
-                    <span class="font-medium text-foreground/80">{{ rpmUsage(provider) }}</span>
-                  </div>
-                  <div
-                    v-if="provider.billing_type !== 'monthly_quota' && !rpmUsage(provider)"
+                    v-else
                     class="text-muted-foreground/50"
                   >
-                    无限制
+                    按量付费
                   </div>
                 </div>
               </TableCell>
@@ -304,7 +294,7 @@
           <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-medium text-foreground truncate">{{ provider.display_name }}</span>
+                <span class="font-medium text-foreground truncate">{{ provider.name }}</span>
                 <Badge
                   :variant="provider.is_active ? 'success' : 'secondary'"
                   class="text-xs shrink-0"
@@ -312,7 +302,6 @@
                   {{ provider.is_active ? '活跃' : '停用' }}
                 </Badge>
               </div>
-              <span class="text-xs text-muted-foreground/70 font-mono">{{ provider.name }}</span>
             </div>
             <div
               class="flex items-center gap-0.5 shrink-0"
@@ -383,20 +372,17 @@
             </span>
           </div>
 
-          <!-- 第四行：配额/限流 -->
+          <!-- 第四行：配额 -->
           <div
-            v-if="provider.billing_type === 'monthly_quota' || rpmUsage(provider)"
+            v-if="provider.billing_type === 'monthly_quota'"
             class="flex items-center gap-3 text-xs text-muted-foreground"
           >
-            <span v-if="provider.billing_type === 'monthly_quota'">
+            <span>
               配额: <span
                 class="font-semibold"
                 :class="getQuotaUsedColorClass(provider)"
               >${{ (provider.monthly_used_usd ?? 0).toFixed(2) }}</span> / ${{ (provider.monthly_quota_usd ?? 0).toFixed(2) }}
             </span>
-            <span v-if="rpmUsage(provider)">
-              RPM: {{ rpmUsage(provider) }}
-            </span>
           </div>
         </div>
       </div>
@@ -509,7 +495,7 @@ const filteredProviders = computed(() => {
   if (searchQuery.value.trim()) {
     const keywords = searchQuery.value.toLowerCase().split(/\s+/).filter(k => k.length > 0)
     result = result.filter(p => {
-      const searchableText = `${p.display_name} ${p.name}`.toLowerCase()
+      const searchableText = `${p.name}`.toLowerCase()
       return keywords.every(keyword => searchableText.includes(keyword))
     })
   }
@@ -525,7 +511,7 @@ const filteredProviders = computed(() => {
       return a.provider_priority - b.provider_priority
     }
     // 3. 按名称排序
-    return a.display_name.localeCompare(b.display_name)
+    return a.name.localeCompare(b.name)
   })
 })
 
@@ -586,7 +572,10 @@ function sortEndpoints(endpoints: any[]) {
 
 // 判断端点是否可用（有 key）
 function isEndpointAvailable(endpoint: any, _provider: ProviderWithEndpointsSummary): boolean {
-  // 检查该端点是否有活跃的密钥
+  // 检查端点是否启用，以及是否有活跃的密钥
+  if (endpoint.is_active === false) {
+    return false
+  }
   return (endpoint.active_keys ?? 0) > 0
 }
 
@@ -639,21 +628,6 @@ function getQuotaUsedColorClass(provider: ProviderWithEndpointsSummary): string
   return 'text-foreground'
 }
 
-function rpmUsage(provider: ProviderWithEndpointsSummary): string | null {
-  const rpmLimit = provider.rpm_limit
-  const rpmUsed = provider.rpm_used ?? 0
-
-  if (rpmLimit === null || rpmLimit === undefined) {
-    return rpmUsed > 0 ? `${rpmUsed}` : null
-  }
-
-  if (rpmLimit === 0) {
-    return '已完全禁止'
-  }
-
-  return `${rpmUsed} / ${rpmLimit}`
-}
-
 // 使用复用的行点击逻辑
 const { handleMouseDown, shouldTriggerRowClick } = useRowClick()
 
@@ -706,7 +680,7 @@ function handleProviderAdded() {
 async function handleDeleteProvider(provider: ProviderWithEndpointsSummary) {
   const confirmed = await confirmDanger(
     '删除提供商',
-    `确定要删除提供商 "${provider.display_name}" 吗？\n\n这将同时删除其所有端点、密钥和配置。此操作不可恢复！`
+    `确定要删除提供商 "${provider.name}" 吗？\n\n这将同时删除其所有端点、密钥和配置。此操作不可恢复！`
   )
 
   if (!confirmed) return

frontend/src/views/admin/SystemSettings.vue

@@ -511,7 +511,7 @@
               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
+              API Keys: {{ importPreview.providers?.reduce((sum: number, p: any) => sum + (p.api_keys?.length || 0), 0) }} 个
             </li>
           </ul>
         </div>
@@ -1144,7 +1144,7 @@ function handleConfigFileSelect(event: Event) {
       const data = JSON.parse(content) as ConfigExportData
 
       // 验证版本
-      if (data.version !== '1.0') {
+      if (data.version !== '2.0') {
         error(`不支持的配置版本: ${data.version}`)
         return
       }

frontend/src/views/admin/Users.vue

@@ -907,7 +907,7 @@ function editUser(user: any) {
     role: user.role,
     is_active: user.is_active,
     allowed_providers: user.allowed_providers || [],
-    allowed_endpoints: user.allowed_endpoints || [],
+    allowed_api_formats: user.allowed_api_formats || [],
     allowed_models: user.allowed_models || []
   }
   showUserFormDialog.value = true
@@ -929,7 +929,7 @@ async function handleUserFormSubmit(data: UserFormData & { password?: string })
         quota_usd: data.quota_usd,
         role: data.role,
         allowed_providers: data.allowed_providers,
-        allowed_endpoints: data.allowed_endpoints,
+        allowed_api_formats: data.allowed_api_formats,
         allowed_models: data.allowed_models
       }
       if (data.password) {
@@ -946,7 +946,7 @@ async function handleUserFormSubmit(data: UserFormData & { password?: string })
         quota_usd: data.quota_usd,
         role: data.role,
         allowed_providers: data.allowed_providers,
-        allowed_endpoints: data.allowed_endpoints,
+        allowed_api_formats: data.allowed_api_formats,
         allowed_models: data.allowed_models
       })
       // 如果创建时指定为禁用，则更新状态

frontend/src/views/shared/Dashboard.vue

@@ -16,7 +16,8 @@
 
         <!-- 主要统计卡片 -->
         <div class="grid grid-cols-2 gap-3 sm:gap-4 xl:grid-cols-4">
-          <template v-if="loading && stats.length === 0">
+          <!-- 加载中骨架屏 -->
+          <template v-if="loading">
             <Card
               v-for="i in 4"
               :key="'skeleton-' + i"
@@ -27,62 +28,98 @@
               <Skeleton class="h-4 w-16" />
             </Card>
           </template>
-          <Card
-            v-for="(stat, index) in stats"
-            v-else
-            :key="stat.name"
-            class="relative overflow-hidden p-3 sm:p-5"
-            :class="statCardBorders[index % statCardBorders.length]"
-          >
-            <div
-              class="pointer-events-none absolute -right-4 -top-6 h-28 w-28 rounded-full blur-3xl opacity-40"
-              :class="statCardGlows[index % statCardGlows.length]"
-            />
-            <!-- 图标固定在右上角 -->
-            <div
-              class="absolute top-3 right-3 sm:top-5 sm:right-5 rounded-xl sm:rounded-2xl border border-border bg-card/50 p-2 sm:p-3 shadow-inner backdrop-blur-sm"
-              :class="getStatIconColor(index)"
+          <!-- 有数据时显示统计卡片 -->
+          <template v-else-if="stats.length > 0">
+            <Card
+              v-for="(stat, index) in stats"
+              :key="stat.name"
+              class="relative overflow-hidden p-3 sm:p-5"
+              :class="statCardBorders[index % statCardBorders.length]"
             >
-              <component
-                :is="stat.icon"
-                class="h-4 w-4 sm:h-5 sm:w-5"
-              />
-            </div>
-            <!-- 内容区域 -->
-            <div>
-              <p class="text-[9px] sm:text-[11px] font-semibold uppercase tracking-[0.2em] sm:tracking-[0.4em] text-muted-foreground pr-10 sm:pr-14">
-                {{ stat.name }}
-              </p>
-              <p class="mt-2 sm:mt-4 text-xl sm:text-3xl font-semibold text-foreground">
-                {{ stat.value }}
-              </p>
-              <p
-                v-if="stat.subValue"
-                class="mt-0.5 sm:mt-1 text-[10px] sm:text-sm text-muted-foreground"
-              >
-                {{ stat.subValue }}
-              </p>
               <div
-                v-if="stat.change || stat.extraBadge"
-                class="mt-1.5 sm:mt-2 flex items-center gap-1 sm:gap-1.5 flex-wrap"
+                class="pointer-events-none absolute -right-4 -top-6 h-28 w-28 rounded-full blur-3xl opacity-40"
+                :class="statCardGlows[index % statCardGlows.length]"
+              />
+              <!-- 图标固定在右上角 -->
+              <div
+                class="absolute top-3 right-3 sm:top-5 sm:right-5 rounded-xl sm:rounded-2xl border border-border bg-card/50 p-2 sm:p-3 shadow-inner backdrop-blur-sm"
+                :class="getStatIconColor(index)"
               >
-                <Badge
-                  v-if="stat.change"
-                  variant="secondary"
-                  class="text-[9px] sm:text-xs"
-                >
-                  {{ stat.change }}
-                </Badge>
-                <Badge
-                  v-if="stat.extraBadge"
-                  variant="secondary"
-                  class="text-[9px] sm:text-xs"
-                >
-                  {{ stat.extraBadge }}
-                </Badge>
+                <component
+                  :is="stat.icon"
+                  class="h-4 w-4 sm:h-5 sm:w-5"
+                />
               </div>
-            </div>
-          </Card>
+              <!-- 内容区域 -->
+              <div>
+                <p class="text-[9px] sm:text-[11px] font-semibold uppercase tracking-[0.2em] sm:tracking-[0.4em] text-muted-foreground pr-10 sm:pr-14">
+                  {{ stat.name }}
+                </p>
+                <p class="mt-2 sm:mt-4 text-xl sm:text-3xl font-semibold text-foreground">
+                  {{ stat.value }}
+                </p>
+                <p
+                  v-if="stat.subValue"
+                  class="mt-0.5 sm:mt-1 text-[10px] sm:text-sm text-muted-foreground"
+                >
+                  {{ stat.subValue }}
+                </p>
+                <div
+                  v-if="stat.change || stat.extraBadge"
+                  class="mt-1.5 sm:mt-2 flex items-center gap-1 sm:gap-1.5 flex-wrap"
+                >
+                  <Badge
+                    v-if="stat.change"
+                    variant="secondary"
+                    class="text-[9px] sm:text-xs"
+                  >
+                    {{ stat.change }}
+                  </Badge>
+                  <Badge
+                    v-if="stat.extraBadge"
+                    variant="secondary"
+                    class="text-[9px] sm:text-xs"
+                  >
+                    {{ stat.extraBadge }}
+                  </Badge>
+                </div>
+              </div>
+            </Card>
+          </template>
+          <!-- 无数据时显示占位卡片 -->
+          <template v-else>
+            <Card
+              v-for="(placeholder, index) in emptyStatPlaceholders"
+              :key="'empty-' + index"
+              class="relative overflow-hidden p-3 sm:p-5"
+              :class="statCardBorders[index % statCardBorders.length]"
+            >
+              <div
+                class="pointer-events-none absolute -right-4 -top-6 h-28 w-28 rounded-full blur-3xl opacity-20"
+                :class="statCardGlows[index % statCardGlows.length]"
+              />
+              <div
+                class="absolute top-3 right-3 sm:top-5 sm:right-5 rounded-xl sm:rounded-2xl border border-border bg-card/50 p-2 sm:p-3 shadow-inner backdrop-blur-sm"
+                :class="getStatIconColor(index)"
+              >
+                <component
+                  :is="placeholder.icon"
+                  class="h-4 w-4 sm:h-5 sm:w-5"
+                />
+              </div>
+              <div>
+                <p class="text-[9px] sm:text-[11px] font-semibold uppercase tracking-[0.2em] sm:tracking-[0.4em] text-muted-foreground pr-10 sm:pr-14">
+                  {{ placeholder.name }}
+                </p>
+                <p class="mt-2 sm:mt-4 text-xl sm:text-3xl font-semibold text-muted-foreground/50">
+                  --
+                </p>
+                <p class="mt-0.5 sm:mt-1 text-[10px] sm:text-sm text-muted-foreground/50">
+                  暂无数据
+                </p>
+              </div>
+            </Card>
+          </template>
         </div>
 
         <!-- 管理员：系统健康摘要 -->
@@ -872,6 +909,24 @@ const iconMap: Record<string, any> = {
   Users, Activity, TrendingUp, DollarSign, Key, Hash, Database
 }
 
+// 空状态占位卡片
+const emptyStatPlaceholders = computed(() => {
+  if (isAdmin.value) {
+    return [
+      { name: '今日请求', icon: Activity },
+      { name: '今日 Tokens', icon: Hash },
+      { name: '活跃用户', icon: Users },
+      { name: '今日费用', icon: DollarSign }
+    ]
+  }
+  return [
+    { name: '今日请求', icon: Activity },
+    { name: '今日 Tokens', icon: Hash },
+    { name: 'API Keys', icon: Key },
+    { name: '今日费用', icon: DollarSign }
+  ]
+})
+
 const totalStats = computed(() => {
   if (dailyStats.value.length === 0) {
     return { requests: 0, tokens: 0, cost: 0, avgResponseTime: 0 }

frontend/src/views/user/ManagementTokens.vue

@@ -0,0 +1,859 @@
+<template>
+  <div class="space-y-6 pb-8">
+    <!-- 访问令牌表格 -->
+    <Card
+      variant="default"
+      class="overflow-hidden"
+    >
+      <!-- 标题和操作栏 -->
+      <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>
+            <h3 class="text-sm sm:text-base font-semibold">
+              访问令牌
+            </h3>
+            <p class="text-xs text-muted-foreground mt-0.5">
+              <template v-if="quota">
+                已创建 {{ quota.used }}/{{ quota.max }} 个令牌
+                <span
+                  v-if="quota.used >= quota.max"
+                  class="text-destructive font-medium"
+                >（已达上限）</span>
+              </template>
+              <template v-else>
+                用于程序化访问管理 API 的令牌
+              </template>
+            </p>
+          </div>
+
+          <!-- 操作按钮 -->
+          <div class="flex items-center gap-2">
+            <!-- 新增按钮 -->
+            <Button
+              variant="ghost"
+              size="icon"
+              class="h-8 w-8"
+              title="创建新令牌"
+              :disabled="quota ? quota.used >= quota.max : false"
+              @click="showCreateDialog = true"
+            >
+              <Plus class="w-3.5 h-3.5" />
+            </Button>
+
+            <!-- 刷新按钮 -->
+            <RefreshButton
+              :loading="loading"
+              @click="loadTokens"
+            />
+          </div>
+        </div>
+      </div>
+
+      <!-- 加载状态 -->
+      <div
+        v-if="loading"
+        class="flex items-center justify-center py-12"
+      >
+        <LoadingState message="加载中..." />
+      </div>
+
+      <!-- 空状态 -->
+      <div
+        v-else-if="tokens.length === 0"
+        class="flex items-center justify-center py-12"
+      >
+        <EmptyState
+          title="暂无访问令牌"
+          description="创建你的第一个访问令牌开始使用管理 API"
+          :icon="KeyRound"
+        >
+          <template #actions>
+            <Button
+              size="lg"
+              class="shadow-lg shadow-primary/20"
+              @click="showCreateDialog = true"
+            >
+              <Plus class="mr-2 h-4 w-4" />
+              创建访问令牌
+            </Button>
+          </template>
+        </EmptyState>
+      </div>
+
+      <!-- 桌面端表格 -->
+      <div
+        v-else
+        class="hidden md:block overflow-x-auto"
+      >
+        <Table>
+          <TableHeader>
+            <TableRow class="border-b border-border/60 hover:bg-transparent">
+              <TableHead class="min-w-[180px] h-12 font-semibold">
+                名称
+              </TableHead>
+              <TableHead class="min-w-[160px] h-12 font-semibold">
+                令牌
+              </TableHead>
+              <TableHead class="min-w-[80px] h-12 font-semibold text-center">
+                使用次数
+              </TableHead>
+              <TableHead class="min-w-[70px] h-12 font-semibold text-center">
+                状态
+              </TableHead>
+              <TableHead class="min-w-[100px] h-12 font-semibold">
+                时间
+              </TableHead>
+              <TableHead class="min-w-[100px] h-12 font-semibold text-center">
+                操作
+              </TableHead>
+            </TableRow>
+          </TableHeader>
+          <TableBody>
+            <TableRow
+              v-for="token in paginatedTokens"
+              :key="token.id"
+              class="border-b border-border/40 hover:bg-muted/30 transition-colors"
+            >
+              <!-- 名称 -->
+              <TableCell class="py-4">
+                <div class="flex-1 min-w-0">
+                  <div
+                    class="text-sm font-semibold truncate"
+                    :title="token.name"
+                  >
+                    {{ token.name }}
+                  </div>
+                  <div
+                    v-if="token.description"
+                    class="text-xs text-muted-foreground mt-0.5 truncate"
+                    :title="token.description"
+                  >
+                    {{ token.description }}
+                  </div>
+                </div>
+              </TableCell>
+
+              <!-- Token 显示 -->
+              <TableCell class="py-4">
+                <div class="flex items-center gap-1.5">
+                  <code class="text-xs font-mono text-muted-foreground bg-muted/30 px-2 py-1 rounded">
+                    {{ token.token_display }}
+                  </code>
+                  <Button
+                    variant="ghost"
+                    size="icon"
+                    class="h-6 w-6"
+                    title="重新生成令牌"
+                    @click="confirmRegenerate(token)"
+                  >
+                    <RefreshCw class="h-3.5 w-3.5" />
+                  </Button>
+                </div>
+              </TableCell>
+
+              <!-- 使用次数 -->
+              <TableCell class="py-4 text-center">
+                <span class="text-sm font-medium">
+                  {{ formatNumber(token.usage_count || 0) }}
+                </span>
+              </TableCell>
+
+              <!-- 状态 -->
+              <TableCell class="py-4 text-center">
+                <Badge
+                  :variant="getStatusVariant(token)"
+                  class="font-medium px-3 py-1"
+                >
+                  {{ getStatusText(token) }}
+                </Badge>
+              </TableCell>
+
+              <!-- 时间 -->
+              <TableCell class="py-4 text-sm text-muted-foreground">
+                <div class="text-xs">
+                  创建于 {{ formatDate(token.created_at) }}
+                </div>
+                <div class="text-xs mt-1">
+                  {{ token.last_used_at ? `最后使用 ${formatRelativeTime(token.last_used_at)}` : '从未使用' }}
+                </div>
+              </TableCell>
+
+              <!-- 操作按钮 -->
+              <TableCell class="py-4">
+                <div class="flex justify-center gap-1">
+                  <Button
+                    variant="ghost"
+                    size="icon"
+                    class="h-8 w-8"
+                    title="编辑"
+                    @click="openEditDialog(token)"
+                  >
+                    <Pencil class="h-4 w-4" />
+                  </Button>
+                  <Button
+                    variant="ghost"
+                    size="icon"
+                    class="h-8 w-8"
+                    :title="token.is_active ? '禁用' : '启用'"
+                    @click="toggleToken(token)"
+                  >
+                    <Power class="h-4 w-4" />
+                  </Button>
+                  <Button
+                    variant="ghost"
+                    size="icon"
+                    class="h-8 w-8"
+                    title="删除"
+                    @click="confirmDelete(token)"
+                  >
+                    <Trash2 class="h-4 w-4" />
+                  </Button>
+                </div>
+              </TableCell>
+            </TableRow>
+          </TableBody>
+        </Table>
+      </div>
+
+      <!-- 移动端卡片列表 -->
+      <div
+        v-if="!loading && tokens.length > 0"
+        class="md:hidden space-y-3 p-4"
+      >
+        <Card
+          v-for="token in paginatedTokens"
+          :key="token.id"
+          variant="default"
+          class="group hover:shadow-md hover:border-primary/30 transition-all duration-200"
+        >
+          <div class="p-4">
+            <!-- 第一行：名称、状态、操作 -->
+            <div class="flex items-center justify-between mb-2">
+              <div class="flex items-center gap-2 min-w-0 flex-1">
+                <h3 class="text-sm font-semibold text-foreground truncate">
+                  {{ token.name }}
+                </h3>
+                <Badge
+                  :variant="getStatusVariant(token)"
+                  class="text-xs px-1.5 py-0"
+                >
+                  {{ getStatusText(token) }}
+                </Badge>
+              </div>
+              <div class="flex items-center gap-0.5 flex-shrink-0">
+                <Button
+                  variant="ghost"
+                  size="icon"
+                  class="h-7 w-7"
+                  title="编辑"
+                  @click="openEditDialog(token)"
+                >
+                  <Pencil class="h-3.5 w-3.5" />
+                </Button>
+                <Button
+                  variant="ghost"
+                  size="icon"
+                  class="h-7 w-7"
+                  :title="token.is_active ? '禁用' : '启用'"
+                  @click="toggleToken(token)"
+                >
+                  <Power class="h-3.5 w-3.5" />
+                </Button>
+                <Button
+                  variant="ghost"
+                  size="icon"
+                  class="h-7 w-7"
+                  title="删除"
+                  @click="confirmDelete(token)"
+                >
+                  <Trash2 class="h-3.5 w-3.5" />
+                </Button>
+              </div>
+            </div>
+
+            <!-- Token 显示 -->
+            <div class="flex items-center gap-2 text-xs mb-2">
+              <code class="font-mono text-muted-foreground">{{ token.token_display }}</code>
+              <Button
+                variant="ghost"
+                size="icon"
+                class="h-5 w-5"
+                title="重新生成"
+                @click="confirmRegenerate(token)"
+              >
+                <RefreshCw class="h-3 w-3" />
+              </Button>
+            </div>
+
+            <!-- 统计信息 -->
+            <div class="flex items-center gap-3 text-xs text-muted-foreground">
+              <span>{{ formatNumber(token.usage_count || 0) }} 次使用</span>
+              <span>·</span>
+              <span>{{ token.last_used_at ? formatRelativeTime(token.last_used_at) : '从未使用' }}</span>
+            </div>
+          </div>
+        </Card>
+      </div>
+
+      <!-- 分页 -->
+      <Pagination
+        v-if="totalTokens > 0"
+        :current="currentPage"
+        :total="totalTokens"
+        :page-size="pageSize"
+        @update:current="currentPage = $event"
+        @update:page-size="handlePageSizeChange"
+      />
+    </Card>
+
+    <!-- 创建/编辑 Token 对话框 -->
+    <Dialog
+      v-model="showCreateDialog"
+      size="lg"
+    >
+      <template #header>
+        <div class="border-b border-border px-6 py-4">
+          <div class="flex items-center gap-3">
+            <div class="flex h-9 w-9 items-center justify-center rounded-lg bg-primary/10 flex-shrink-0">
+              <KeyRound class="h-5 w-5 text-primary" />
+            </div>
+            <div class="flex-1 min-w-0">
+              <h3 class="text-lg font-semibold text-foreground leading-tight">
+                {{ editingToken ? '编辑访问令牌' : '创建访问令牌' }}
+              </h3>
+              <p class="text-xs text-muted-foreground">
+                {{ editingToken ? '修改令牌配置' : '创建一个新的令牌用于访问管理 API' }}
+              </p>
+            </div>
+          </div>
+        </div>
+      </template>
+
+      <div class="space-y-4">
+        <!-- 名称 -->
+        <div class="space-y-2">
+          <Label
+            for="token-name"
+            class="text-sm font-semibold"
+          >名称 *</Label>
+          <Input
+            id="token-name"
+            v-model="formData.name"
+            placeholder="例如：CI/CD 自动化"
+            class="h-11 border-border/60"
+            autocomplete="off"
+            required
+          />
+        </div>
+
+        <!-- 描述 -->
+        <div class="space-y-2">
+          <Label
+            for="token-description"
+            class="text-sm font-semibold"
+          >描述</Label>
+          <Input
+            id="token-description"
+            v-model="formData.description"
+            placeholder="用途说明（可选）"
+            class="h-11 border-border/60"
+            autocomplete="off"
+          />
+        </div>
+
+
+        <!-- IP 白名单 -->
+        <div class="space-y-2">
+          <Label
+            for="token-ips"
+            class="text-sm font-semibold"
+          >IP 白名单</Label>
+          <Input
+            id="token-ips"
+            v-model="formData.allowedIpsText"
+            placeholder="例如：192.168.1.0/24, 10.0.0.1（逗号分隔，留空不限制）"
+            class="h-11 border-border/60"
+            autocomplete="off"
+          />
+          <p class="text-xs text-muted-foreground">
+            限制只能从指定 IP 地址使用此令牌，支持 CIDR 格式
+          </p>
+        </div>
+
+        <!-- 过期时间 -->
+        <div class="space-y-2">
+          <Label
+            for="token-expires"
+            class="text-sm font-semibold"
+          >过期时间</Label>
+          <Input
+            id="token-expires"
+            v-model="formData.expiresAt"
+            type="datetime-local"
+            class="h-11 border-border/60"
+          />
+          <p class="text-xs text-muted-foreground">
+            留空表示永不过期
+          </p>
+        </div>
+      </div>
+
+      <template #footer>
+        <Button
+          variant="outline"
+          class="h-11 px-6"
+          @click="closeDialog"
+        >
+          取消
+        </Button>
+        <Button
+          class="h-11 px-6 shadow-lg shadow-primary/20"
+          :disabled="saving || !isFormValid"
+          @click="saveToken"
+        >
+          <Loader2
+            v-if="saving"
+            class="animate-spin h-4 w-4 mr-2"
+          />
+          {{ saving ? '保存中...' : (editingToken ? '保存' : '创建') }}
+        </Button>
+      </template>
+    </Dialog>
+
+    <!-- 新 Token 创建成功对话框 -->
+    <Dialog
+      v-model="showTokenDialog"
+      size="lg"
+      persistent
+    >
+      <template #header>
+        <div class="border-b border-border px-6 py-4">
+          <div class="flex items-center gap-3">
+            <div class="flex h-9 w-9 items-center justify-center rounded-lg bg-emerald-100 dark:bg-emerald-900/30 flex-shrink-0">
+              <CheckCircle class="h-5 w-5 text-emerald-600 dark:text-emerald-400" />
+            </div>
+            <div class="flex-1 min-w-0">
+              <h3 class="text-lg font-semibold text-foreground leading-tight">
+                {{ isRegenerating ? '令牌已重新生成' : '创建成功' }}
+              </h3>
+              <p class="text-xs text-muted-foreground">
+                请妥善保管，此令牌只会显示一次
+              </p>
+            </div>
+          </div>
+        </div>
+      </template>
+
+      <div class="space-y-4">
+        <div class="space-y-2">
+          <Label class="text-sm font-medium">访问令牌</Label>
+          <div class="flex items-center gap-2">
+            <Input
+              type="text"
+              :value="newTokenValue"
+              readonly
+              class="flex-1 font-mono text-sm bg-muted/50 h-11"
+              @click="($event.target as HTMLInputElement)?.select()"
+            />
+            <Button
+              class="h-11"
+              @click="copyToken(newTokenValue)"
+            >
+              复制
+            </Button>
+          </div>
+        </div>
+        <div class="p-3 rounded-lg bg-amber-50 dark:bg-amber-950/30 border border-amber-200 dark:border-amber-800">
+          <div class="flex gap-2">
+            <AlertTriangle class="h-4 w-4 text-amber-600 dark:text-amber-400 flex-shrink-0 mt-0.5" />
+            <p class="text-sm text-amber-800 dark:text-amber-200">
+              此令牌只会显示一次，关闭后将无法再次查看，请妥善保管。
+            </p>
+          </div>
+        </div>
+      </div>
+
+      <template #footer>
+        <Button
+          class="h-10 px-5"
+          @click="showTokenDialog = false"
+        >
+          我已安全保存
+        </Button>
+      </template>
+    </Dialog>
+
+    <!-- 删除确认对话框 -->
+    <AlertDialog
+      v-model="showDeleteDialog"
+      type="danger"
+      title="确认删除"
+      :description="`确定要删除令牌「${tokenToDelete?.name}」吗？此操作不可恢复。`"
+      confirm-text="删除"
+      :loading="deleting"
+      @confirm="deleteToken"
+      @cancel="showDeleteDialog = false"
+    />
+
+    <!-- 重新生成确认对话框 -->
+    <AlertDialog
+      v-model="showRegenerateDialog"
+      type="warning"
+      title="确认重新生成"
+      :description="`重新生成后，原令牌将立即失效。确定要重新生成「${tokenToRegenerate?.name}」吗？`"
+      confirm-text="重新生成"
+      :loading="regenerating"
+      @confirm="regenerateToken"
+      @cancel="showRegenerateDialog = false"
+    />
+  </div>
+</template>
+
+<script setup lang="ts">
+import { ref, computed, onMounted, reactive, watch } from 'vue'
+import {
+  managementTokenApi,
+  type ManagementToken
+} from '@/api/management-tokens'
+import Card from '@/components/ui/card.vue'
+import Button from '@/components/ui/button.vue'
+import Input from '@/components/ui/input.vue'
+import Label from '@/components/ui/label.vue'
+import Badge from '@/components/ui/badge.vue'
+import { Dialog, Pagination } from '@/components/ui'
+import { LoadingState, AlertDialog, EmptyState } from '@/components/common'
+import {
+  Table,
+  TableBody,
+  TableCell,
+  TableHead,
+  TableHeader,
+  TableRow
+} from '@/components/ui'
+import RefreshButton from '@/components/ui/refresh-button.vue'
+import {
+  Plus,
+  KeyRound,
+  Trash2,
+  Loader2,
+  CheckCircle,
+  Power,
+  Pencil,
+  RefreshCw,
+  AlertTriangle
+} from 'lucide-vue-next'
+import { useToast } from '@/composables/useToast'
+import { log } from '@/utils/logger'
+
+const { success, error: showError } = useToast()
+
+// 数据
+const tokens = ref<ManagementToken[]>([])
+const totalTokens = ref(0)
+const loading = ref(false)
+const saving = ref(false)
+const deleting = ref(false)
+const regenerating = ref(false)
+
+// 配额信息
+const quota = ref<{ used: number; max: number } | null>(null)
+
+// 分页
+const currentPage = ref(1)
+const pageSize = ref(10)
+
+const paginatedTokens = computed(() => tokens.value)
+
+// 监听分页变化
+watch([currentPage, pageSize], () => {
+  loadTokens()
+})
+
+function handlePageSizeChange(newSize: number) {
+  pageSize.value = newSize
+  currentPage.value = 1
+}
+
+// 对话框状态
+const showCreateDialog = ref(false)
+const showTokenDialog = ref(false)
+const showDeleteDialog = ref(false)
+const showRegenerateDialog = ref(false)
+
+// 表单数据
+const editingToken = ref<ManagementToken | null>(null)
+const formData = reactive({
+  name: '',
+  description: '',
+  allowedIpsText: '',
+  expiresAt: ''
+})
+
+const newTokenValue = ref('')
+const isRegenerating = ref(false)
+const tokenToDelete = ref<ManagementToken | null>(null)
+const tokenToRegenerate = ref<ManagementToken | null>(null)
+
+// 表单验证
+const isFormValid = computed(() => {
+  return formData.name.trim().length > 0
+})
+
+function getStatusVariant(token: ManagementToken): 'success' | 'secondary' | 'destructive' {
+  if (token.expires_at && isExpired(token.expires_at)) {
+    return 'destructive'
+  }
+  return token.is_active ? 'success' : 'secondary'
+}
+
+function getStatusText(token: ManagementToken): string {
+  if (token.expires_at && isExpired(token.expires_at)) {
+    return '已过期'
+  }
+  return token.is_active ? '活跃' : '禁用'
+}
+
+function isExpired(dateString: string): boolean {
+  return new Date(dateString) < new Date()
+}
+
+// 加载数据
+onMounted(() => {
+  loadTokens()
+})
+
+async function loadTokens() {
+  loading.value = true
+  try {
+    const skip = (currentPage.value - 1) * pageSize.value
+    const response = await managementTokenApi.listTokens({ skip, limit: pageSize.value })
+
+    tokens.value = response.items
+    totalTokens.value = response.total
+
+    if (response.quota) {
+      quota.value = response.quota
+    }
+
+    // 如果当前页超出范围，重置到第一页
+    if (tokens.value.length === 0 && currentPage.value > 1) {
+      currentPage.value = 1
+    }
+  } catch (err: any) {
+    log.error('加载 Management Tokens 失败:', err)
+    if (!err.response) {
+      showError('无法连接到服务器')
+    } else {
+      showError(`加载失败：${err.response?.data?.detail || err.message}`)
+    }
+  } finally {
+    loading.value = false
+  }
+}
+
+// 打开编辑对话框
+function openEditDialog(token: ManagementToken) {
+  editingToken.value = token
+  formData.name = token.name
+  formData.description = token.description || ''
+  formData.allowedIpsText = (token.allowed_ips && token.allowed_ips.length > 0)
+    ? token.allowed_ips.join(', ')
+    : ''
+  formData.expiresAt = token.expires_at
+    ? toLocalDatetimeString(new Date(token.expires_at))
+    : ''
+  showCreateDialog.value = true
+}
+
+// 关闭对话框
+function closeDialog() {
+  showCreateDialog.value = false
+  editingToken.value = null
+  formData.name = ''
+  formData.description = ''
+  formData.allowedIpsText = ''
+  formData.expiresAt = ''
+}
+
+// 保存 Token
+async function saveToken() {
+  if (!isFormValid.value) return
+
+  saving.value = true
+  try {
+    const allowedIps = formData.allowedIpsText
+      .split(',')
+      .map(ip => ip.trim())
+      .filter(ip => ip)
+
+    // 将本地时间转换为 UTC ISO 字符串
+    const expiresAtUtc = formData.expiresAt
+      ? new Date(formData.expiresAt).toISOString()
+      : null
+
+    if (editingToken.value) {
+      // 更新
+      await managementTokenApi.updateToken(editingToken.value.id, {
+        name: formData.name,
+        description: formData.description.trim() || null,
+        allowed_ips: allowedIps.length > 0 ? allowedIps : null,
+        expires_at: expiresAtUtc
+      })
+      success('令牌更新成功')
+    } else {
+      // 创建
+      const result = await managementTokenApi.createToken({
+        name: formData.name,
+        description: formData.description || undefined,
+        allowed_ips: allowedIps.length > 0 ? allowedIps : undefined,
+        expires_at: expiresAtUtc
+      })
+      newTokenValue.value = result.token
+      isRegenerating.value = false
+      showTokenDialog.value = true
+      success('令牌创建成功')
+    }
+
+    closeDialog()
+    await loadTokens()
+  } catch (err: any) {
+    log.error('保存 Token 失败:', err)
+    const message = err.response?.data?.error?.message
+      || err.response?.data?.detail
+      || '保存失败'
+    showError(message)
+  } finally {
+    saving.value = false
+  }
+}
+
+// 切换状态
+async function toggleToken(token: ManagementToken) {
+  try {
+    const result = await managementTokenApi.toggleToken(token.id)
+
+    const index = tokens.value.findIndex(t => t.id === token.id)
+    if (index !== -1) {
+      tokens.value[index] = result.data
+    }
+    success(result.data.is_active ? '令牌已启用' : '令牌已禁用')
+  } catch (err: any) {
+    log.error('切换状态失败:', err)
+    showError('操作失败')
+  }
+}
+
+// 删除
+function confirmDelete(token: ManagementToken) {
+  tokenToDelete.value = token
+  showDeleteDialog.value = true
+}
+
+async function deleteToken() {
+  if (!tokenToDelete.value) return
+
+  deleting.value = true
+  try {
+    await managementTokenApi.deleteToken(tokenToDelete.value.id)
+
+    showDeleteDialog.value = false
+    success('令牌已删除')
+    await loadTokens()
+  } catch (err: any) {
+    log.error('删除 Token 失败:', err)
+    showError('删除失败')
+  } finally {
+    deleting.value = false
+    tokenToDelete.value = null
+  }
+}
+
+// 重新生成
+function confirmRegenerate(token: ManagementToken) {
+  tokenToRegenerate.value = token
+  showRegenerateDialog.value = true
+}
+
+async function regenerateToken() {
+  if (!tokenToRegenerate.value) return
+
+  regenerating.value = true
+  try {
+    const result = await managementTokenApi.regenerateToken(tokenToRegenerate.value.id)
+    newTokenValue.value = result.token
+    isRegenerating.value = true
+    showRegenerateDialog.value = false
+    showTokenDialog.value = true
+    await loadTokens()
+    success('令牌已重新生成')
+  } catch (err: any) {
+    log.error('重新生成失败:', err)
+    showError('重新生成失败')
+  } finally {
+    regenerating.value = false
+    tokenToRegenerate.value = null
+  }
+}
+
+// 复制 Token
+async function copyToken(text: string) {
+  try {
+    if (navigator.clipboard && window.isSecureContext) {
+      await navigator.clipboard.writeText(text)
+      success('已复制到剪贴板')
+    } else {
+      const textArea = document.createElement('textarea')
+      textArea.value = text
+      textArea.style.position = 'fixed'
+      textArea.style.left = '-999999px'
+      document.body.appendChild(textArea)
+      textArea.select()
+      document.execCommand('copy')
+      document.body.removeChild(textArea)
+      success('已复制到剪贴板')
+    }
+  } catch (err) {
+    log.error('复制失败:', err)
+    showError('复制失败')
+  }
+}
+
+// 格式化
+function formatNumber(num: number): string {
+  return num.toLocaleString('zh-CN')
+}
+
+function toLocalDatetimeString(date: Date): string {
+  const year = date.getFullYear()
+  const month = String(date.getMonth() + 1).padStart(2, '0')
+  const day = String(date.getDate()).padStart(2, '0')
+  const hours = String(date.getHours()).padStart(2, '0')
+  const minutes = String(date.getMinutes()).padStart(2, '0')
+  return `${year}-${month}-${day}T${hours}:${minutes}`
+}
+
+function formatDate(dateString: string): string {
+  const date = new Date(dateString)
+  return date.toLocaleDateString('zh-CN', {
+    year: 'numeric',
+    month: '2-digit',
+    day: '2-digit'
+  })
+}
+
+function formatRelativeTime(dateString: string): string {
+  const date = new Date(dateString)
+  const now = new Date()
+  const diffMs = now.getTime() - date.getTime()
+  const diffMins = Math.floor(diffMs / (1000 * 60))
+  const diffHours = Math.floor(diffMs / (1000 * 60 * 60))
+  const diffDays = Math.floor(diffMs / (1000 * 60 * 60 * 24))
+
+  if (diffMins < 1) return '刚刚'
+  if (diffMins < 60) return `${diffMins}分钟前`
+  if (diffHours < 24) return `${diffHours}小时前`
+  if (diffDays < 7) return `${diffDays}天前`
+
+  return formatDate(dateString)
+}
+</script>

frontend/tailwind.config.js

@@ -78,6 +78,20 @@ export default {
         md: "calc(var(--radius) - 2px)",
         sm: "calc(var(--radius) - 4px)",
       },
+      keyframes: {
+        "collapsible-down": {
+          from: { height: "0" },
+          to: { height: "var(--radix-collapsible-content-height)" },
+        },
+        "collapsible-up": {
+          from: { height: "var(--radix-collapsible-content-height)" },
+          to: { height: "0" },
+        },
+      },
+      animation: {
+        "collapsible-down": "collapsible-down 0.2s ease-out",
+        "collapsible-up": "collapsible-up 0.2s ease-out",
+      },
     },
   },
   plugins: [require("tailwindcss-animate")],

migrate.sh

@@ -1,12 +0,0 @@
-#!/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/_version.py

@@ -1,34 +0,0 @@
-# file generated by setuptools-scm
-# don't change, don't track in version control
-
-__all__ = [
-    "__version__",
-    "__version_tuple__",
-    "version",
-    "version_tuple",
-    "__commit_id__",
-    "commit_id",
-]
-
-TYPE_CHECKING = False
-if TYPE_CHECKING:
-    from typing import Tuple
-    from typing import Union
-
-    VERSION_TUPLE = Tuple[Union[int, str], ...]
-    COMMIT_ID = Union[str, None]
-else:
-    VERSION_TUPLE = object
-    COMMIT_ID = object
-
-version: str
-__version__: str
-__version_tuple__: VERSION_TUPLE
-version_tuple: VERSION_TUPLE
-commit_id: COMMIT_ID
-__commit_id__: COMMIT_ID
-
-__version__ = version = '0.2.3.dev0+g0f78d5cbf.d20260105'
-__version_tuple__ = version_tuple = (0, 2, 3, 'dev0', 'g0f78d5cbf.d20260105')
-
-__commit_id__ = commit_id = None

src/api/admin/__init__.py

@@ -6,6 +6,7 @@ from .adaptive import router as adaptive_router
 from .api_keys import router as api_keys_router
 from .endpoints import router as endpoints_router
 from .ldap import router as ldap_router
+from .management_tokens import router as management_tokens_router
 from .models import router as models_router
 from .monitoring import router as monitoring_router
 from .provider_query import router as provider_query_router
@@ -30,5 +31,6 @@ router.include_router(models_router)
 router.include_router(security_router)
 router.include_router(provider_query_router)
 router.include_router(ldap_router)
+router.include_router(management_tokens_router)
 
 __all__ = ["router"]

src/api/admin/adaptive.py

@@ -1,12 +1,12 @@
 """
-自适应并发管理 API 端点
+自适应 RPM 管理 API 端点
 
 设计原则：
-- 自适应模式由 max_concurrent 字段决定：
-  - max_concurrent = NULL：启用自适应模式，系统自动学习并调整并发限制
-  - max_concurrent = 数字：固定限制模式，使用用户指定的并发限制
-- learned_max_concurrent：自适应模式下学习到的并发限制值
-- adaptive_mode 是计算字段，基于 max_concurrent 是否为 NULL
+- 自适应模式由 rpm_limit 字段决定：
+  - rpm_limit = NULL：启用自适应模式，系统自动学习并调整 RPM 限制
+  - rpm_limit = 数字：固定限制模式，使用用户指定的 RPM 限制
+- learned_rpm_limit：自适应模式下学习到的 RPM 限制值
+- adaptive_mode 是计算字段，基于 rpm_limit 是否为 NULL
 """
 
 from dataclasses import dataclass
@@ -18,12 +18,13 @@ from sqlalchemy.orm import Session
 
 from src.api.base.admin_adapter import AdminApiAdapter
 from src.api.base.pipeline import ApiRequestPipeline
+from src.config.constants import RPMDefaults
 from src.core.exceptions import InvalidRequestException, translate_pydantic_error
 from src.database import get_db
 from src.models.database import ProviderAPIKey
-from src.services.rate_limit.adaptive_concurrency import get_adaptive_manager
+from src.services.rate_limit.adaptive_rpm import get_adaptive_rpm_manager
 
-router = APIRouter(prefix="/api/admin/adaptive", tags=["Adaptive Concurrency"])
+router = APIRouter(prefix="/api/admin/adaptive", tags=["Adaptive RPM"])
 pipeline = ApiRequestPipeline()
 
 
@@ -35,19 +36,19 @@ class EnableAdaptiveRequest(BaseModel):
 
     enabled: bool = Field(..., description="是否启用自适应模式（true=自适应，false=固定限制）")
     fixed_limit: Optional[int] = Field(
-        None, ge=1, le=100, description="固定并发限制（仅当 enabled=false 时生效）"
+        None, ge=1, le=100, description="固定 RPM 限制（仅当 enabled=false 时生效，1-100）"
     )
 
 
 class AdaptiveStatsResponse(BaseModel):
     """自适应统计响应"""
 
-    adaptive_mode: bool = Field(..., description="是否为自适应模式（max_concurrent=NULL）")
-    max_concurrent: Optional[int] = Field(None, description="用户配置的固定限制（NULL=自适应）")
+    adaptive_mode: bool = Field(..., description="是否为自适应模式（rpm_limit=NULL）")
+    rpm_limit: Optional[int] = Field(None, description="用户配置的固定限制（NULL=自适应）")
     effective_limit: Optional[int] = Field(
         None, description="当前有效限制（自适应使用学习值，固定使用配置值）"
     )
-    learned_limit: Optional[int] = Field(None, description="学习到的并发限制")
+    learned_limit: Optional[int] = Field(None, description="学习到的 RPM 限制")
     concurrent_429_count: int
     rpm_429_count: int
     last_429_at: Optional[str]
@@ -61,11 +62,12 @@ class KeyListItem(BaseModel):
 
     id: str
     name: Optional[str]
-    endpoint_id: str
-    is_adaptive: bool = Field(..., description="是否为自适应模式（max_concurrent=NULL）")
-    max_concurrent: Optional[int] = Field(None, description="固定并发限制（NULL=自适应）")
+    provider_id: str
+    api_formats: List[str] = Field(default_factory=list)
+    is_adaptive: bool = Field(..., description="是否为自适应模式（rpm_limit=NULL）")
+    rpm_limit: Optional[int] = Field(None, description="固定 RPM 限制（NULL=自适应）")
     effective_limit: Optional[int] = Field(None, description="当前有效限制")
-    learned_max_concurrent: Optional[int] = Field(None, description="学习到的并发限制")
+    learned_rpm_limit: Optional[int] = Field(None, description="学习到的 RPM 限制")
     concurrent_429_count: int
     rpm_429_count: int
 
@@ -80,22 +82,22 @@ class KeyListItem(BaseModel):
 )
 async def list_adaptive_keys(
     request: Request,
-    endpoint_id: Optional[str] = Query(None, description="按 Endpoint 过滤"),
+    provider_id: Optional[str] = Query(None, description="按 Provider 过滤"),
     db: Session = Depends(get_db),
 ):
     """
     获取所有启用自适应模式的Key列表
 
     可选参数：
-    - endpoint_id: 按 Endpoint 过滤
+    - provider_id: 按 Provider 过滤
     """
-    adapter = ListAdaptiveKeysAdapter(endpoint_id=endpoint_id)
+    adapter = ListAdaptiveKeysAdapter(provider_id=provider_id)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
 @router.patch(
     "/keys/{key_id}/mode",
-    summary="Toggle key's concurrency control mode",
+    summary="Toggle key's RPM control mode",
 )
 async def toggle_adaptive_mode(
     key_id: str,
@@ -103,10 +105,10 @@ async def toggle_adaptive_mode(
     db: Session = Depends(get_db),
 ):
     """
-    Toggle the concurrency control mode for a specific key
+    Toggle the RPM control mode for a specific key
 
     Parameters:
-    - enabled: true=adaptive mode (max_concurrent=NULL), false=fixed limit mode
+    - enabled: true=adaptive mode (rpm_limit=NULL), false=fixed limit mode
     - fixed_limit: fixed limit value (required when enabled=false)
     """
     adapter = ToggleAdaptiveModeAdapter(key_id=key_id)
@@ -124,7 +126,7 @@ async def get_adaptive_stats(
     db: Session = Depends(get_db),
 ):
     """
-    获取指定Key的自适应并发统计信息
+    获取指定Key的自适应 RPM 统计信息
 
     包括：
     - 当前配置
@@ -149,12 +151,12 @@ async def reset_adaptive_learning(
     Reset the adaptive learning state for a specific key
 
     Clears:
-    - Learned concurrency limit (learned_max_concurrent)
+    - Learned RPM limit (learned_rpm_limit)
     - 429 error counts
     - Adjustment history
 
     Does not change:
-    - max_concurrent config (determines adaptive mode)
+    - rpm_limit config (determines adaptive mode)
     """
     adapter = ResetAdaptiveLearningAdapter(key_id=key_id)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
@@ -162,40 +164,40 @@ async def reset_adaptive_learning(
 
 @router.patch(
     "/keys/{key_id}/limit",
-    summary="Set key to fixed concurrency limit mode",
+    summary="Set key to fixed RPM limit mode",
 )
-async def set_concurrent_limit(
+async def set_rpm_limit(
     key_id: str,
     request: Request,
-    limit: int = Query(..., ge=1, le=100, description="Concurrency limit value"),
+    limit: int = Query(..., ge=1, le=100, description="RPM limit value (1-100)"),
     db: Session = Depends(get_db),
 ):
     """
-    Set key to fixed concurrency limit mode
+    Set key to fixed RPM limit mode
 
     Note:
     - After setting this value, key switches to fixed limit mode and won't auto-adjust
     - To restore adaptive mode, use PATCH /keys/{key_id}/mode
     """
-    adapter = SetConcurrentLimitAdapter(key_id=key_id, limit=limit)
+    adapter = SetRPMLimitAdapter(key_id=key_id, limit=limit)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
 @router.get(
     "/summary",
-    summary="获取自适应并发的全局统计",
+    summary="获取自适应 RPM 的全局统计",
 )
 async def get_adaptive_summary(
     request: Request,
     db: Session = Depends(get_db),
 ):
     """
-    获取自适应并发的全局统计摘要
+    获取自适应 RPM 的全局统计摘要
 
     包括：
     - 启用自适应模式的Key数量
     - 总429错误数
-    - 并发限制调整次数
+    - RPM 限制调整次数
     """
     adapter = AdaptiveSummaryAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
@@ -206,26 +208,29 @@ async def get_adaptive_summary(
 
 @dataclass
 class ListAdaptiveKeysAdapter(AdminApiAdapter):
-    endpoint_id: Optional[str] = None
+    provider_id: Optional[str] = None
 
     async def handle(self, context):  # type: ignore[override]
-        # 自适应模式：max_concurrent = NULL
-        query = context.db.query(ProviderAPIKey).filter(ProviderAPIKey.max_concurrent.is_(None))
-        if self.endpoint_id:
-            query = query.filter(ProviderAPIKey.endpoint_id == self.endpoint_id)
+        # 自适应模式：rpm_limit = NULL
+        query = context.db.query(ProviderAPIKey).filter(ProviderAPIKey.rpm_limit.is_(None))
+        if self.provider_id:
+            query = query.filter(ProviderAPIKey.provider_id == self.provider_id)
 
         keys = query.all()
         return [
             KeyListItem(
                 id=key.id,
                 name=key.name,
-                endpoint_id=key.endpoint_id,
-                is_adaptive=key.max_concurrent is None,
-                max_concurrent=key.max_concurrent,
+                provider_id=key.provider_id,
+                api_formats=key.api_formats or [],
+                is_adaptive=key.rpm_limit is None,
+                rpm_limit=key.rpm_limit,
                 effective_limit=(
-                    key.learned_max_concurrent if key.max_concurrent is None else key.max_concurrent
+                    (key.learned_rpm_limit if key.learned_rpm_limit is not None else RPMDefaults.INITIAL_LIMIT)
+                    if key.rpm_limit is None
+                    else key.rpm_limit
                 ),
-                learned_max_concurrent=key.learned_max_concurrent,
+                learned_rpm_limit=key.learned_rpm_limit,
                 concurrent_429_count=key.concurrent_429_count or 0,
                 rpm_429_count=key.rpm_429_count or 0,
             )
@@ -252,28 +257,32 @@ class ToggleAdaptiveModeAdapter(AdminApiAdapter):
             raise InvalidRequestException("请求数据验证失败")
 
         if body.enabled:
-            # 启用自适应模式：将 max_concurrent 设为 NULL
-            key.max_concurrent = None
-            message = "已切换为自适应模式，系统将自动学习并调整并发限制"
+            # 启用自适应模式：将 rpm_limit 设为 NULL
+            key.rpm_limit = None
+            message = "已切换为自适应模式，系统将自动学习并调整 RPM 限制"
         else:
             # 禁用自适应模式：设置固定限制
             if body.fixed_limit is None:
                 raise HTTPException(
                     status_code=400, detail="禁用自适应模式时必须提供 fixed_limit 参数"
                 )
-            key.max_concurrent = body.fixed_limit
-            message = f"已切换为固定限制模式，并发限制设为 {body.fixed_limit}"
+            key.rpm_limit = body.fixed_limit
+            message = f"已切换为固定限制模式，RPM 限制设为 {body.fixed_limit}"
 
         context.db.commit()
         context.db.refresh(key)
 
-        is_adaptive = key.max_concurrent is None
+        is_adaptive = key.rpm_limit is None
         return {
             "message": message,
             "key_id": key.id,
             "is_adaptive": is_adaptive,
-            "max_concurrent": key.max_concurrent,
-            "effective_limit": key.learned_max_concurrent if is_adaptive else key.max_concurrent,
+            "rpm_limit": key.rpm_limit,
+            "effective_limit": (
+                (key.learned_rpm_limit if key.learned_rpm_limit is not None else RPMDefaults.INITIAL_LIMIT)
+                if is_adaptive
+                else key.rpm_limit
+            ),
         }
 
 
@@ -286,13 +295,13 @@ class GetAdaptiveStatsAdapter(AdminApiAdapter):
         if not key:
             raise HTTPException(status_code=404, detail="Key not found")
 
-        adaptive_manager = get_adaptive_manager()
+        adaptive_manager = get_adaptive_rpm_manager()
         stats = adaptive_manager.get_adjustment_stats(key)
 
         # 转换字段名以匹配响应模型
         return AdaptiveStatsResponse(
             adaptive_mode=stats["adaptive_mode"],
-            max_concurrent=stats["max_concurrent"],
+            rpm_limit=stats["rpm_limit"],
             effective_limit=stats["effective_limit"],
             learned_limit=stats["learned_limit"],
             concurrent_429_count=stats["concurrent_429_count"],
@@ -313,13 +322,13 @@ class ResetAdaptiveLearningAdapter(AdminApiAdapter):
         if not key:
             raise HTTPException(status_code=404, detail="Key not found")
 
-        adaptive_manager = get_adaptive_manager()
+        adaptive_manager = get_adaptive_rpm_manager()
         adaptive_manager.reset_learning(context.db, key)
         return {"message": "学习状态已重置", "key_id": key.id}
 
 
 @dataclass
-class SetConcurrentLimitAdapter(AdminApiAdapter):
+class SetRPMLimitAdapter(AdminApiAdapter):
     key_id: str
     limit: int
 
@@ -328,25 +337,25 @@ class SetConcurrentLimitAdapter(AdminApiAdapter):
         if not key:
             raise HTTPException(status_code=404, detail="Key not found")
 
-        was_adaptive = key.max_concurrent is None
-        key.max_concurrent = self.limit
+        was_adaptive = key.rpm_limit is None
+        key.rpm_limit = self.limit
         context.db.commit()
         context.db.refresh(key)
 
         return {
-            "message": f"已设置为固定限制模式，并发限制为 {self.limit}",
+            "message": f"已设置为固定限制模式，RPM 限制为 {self.limit}",
             "key_id": key.id,
             "is_adaptive": False,
-            "max_concurrent": key.max_concurrent,
+            "rpm_limit": key.rpm_limit,
             "previous_mode": "adaptive" if was_adaptive else "fixed",
         }
 
 
 class AdaptiveSummaryAdapter(AdminApiAdapter):
     async def handle(self, context):  # type: ignore[override]
-        # 自适应模式：max_concurrent = NULL
+        # 自适应模式：rpm_limit = NULL
         adaptive_keys = (
-            context.db.query(ProviderAPIKey).filter(ProviderAPIKey.max_concurrent.is_(None)).all()
+            context.db.query(ProviderAPIKey).filter(ProviderAPIKey.rpm_limit.is_(None)).all()
         )
 
         total_keys = len(adaptive_keys)

src/api/admin/api_keys/routes.py

@@ -73,7 +73,26 @@ async def list_standalone_api_keys(
     is_active: Optional[bool] = None,
     db: Session = Depends(get_db),
 ):
-    """列出所有独立余额API Keys"""
+    """
+    列出所有独立余额 API Keys
+
+    获取系统中所有独立余额 API Key 的列表。独立余额 Key 不关联用户配额，
+    有独立的余额限制，主要用于给非注册用户使用。
+
+    **查询参数**:
+    - `skip`: 跳过的记录数（分页偏移量），默认 0
+    - `limit`: 返回的记录数（分页限制），默认 100，最大 500
+    - `is_active`: 可选，根据启用状态筛选（true/false）
+
+    **返回字段**:
+    - `api_keys`: API Key 列表，包含 id, name, key_display, is_active, current_balance_usd,
+      balance_used_usd, total_requests, total_cost_usd, rate_limit, allowed_providers,
+      allowed_api_formats, allowed_models, last_used_at, expires_at, created_at, updated_at,
+      auto_delete_on_expiry 等字段
+    - `total`: 符合条件的总记录数
+    - `limit`: 当前分页限制
+    - `skip`: 当前分页偏移量
+    """
     adapter = AdminListStandaloneKeysAdapter(skip=skip, limit=limit, is_active=is_active)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -84,7 +103,35 @@ async def create_standalone_api_key(
     key_data: CreateApiKeyRequest,
     db: Session = Depends(get_db),
 ):
-    """创建独立余额API Key（必须设置余额限制）"""
+    """
+    创建独立余额 API Key
+
+    创建一个新的独立余额 API Key。独立余额 Key 必须设置初始余额限制。
+
+    **请求体字段**:
+    - `name`: API Key 的名称
+    - `initial_balance_usd`: 必需，初始余额（美元），必须大于 0
+    - `allowed_providers`: 可选，允许使用的提供商列表
+    - `allowed_api_formats`: 可选，允许使用的 API 格式列表
+    - `allowed_models`: 可选，允许使用的模型列表
+    - `rate_limit`: 可选，速率限制配置（请求数/秒）
+    - `expire_days`: 可选，过期天数（兼容旧版）
+    - `expires_at`: 可选，过期时间（ISO 格式或 YYYY-MM-DD 格式，优先级高于 expire_days）
+    - `auto_delete_on_expiry`: 可选，过期后是否自动删除
+
+    **返回字段**:
+    - `id`: API Key ID
+    - `key`: 完整的 API Key（仅在创建时返回一次）
+    - `name`: API Key 名称
+    - `key_display`: 脱敏显示的 Key
+    - `is_standalone`: 是否为独立余额 Key（始终为 true）
+    - `current_balance_usd`: 当前余额
+    - `balance_used_usd`: 已使用余额
+    - `rate_limit`: 速率限制配置
+    - `expires_at`: 过期时间
+    - `created_at`: 创建时间
+    - `message`: 提示信息
+    """
     adapter = AdminCreateStandaloneKeyAdapter(key_data=key_data)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -93,20 +140,72 @@ async def create_standalone_api_key(
 async def update_api_key(
     key_id: str, request: Request, key_data: CreateApiKeyRequest, db: Session = Depends(get_db)
 ):
-    """更新独立余额Key（可修改名称、过期时间、余额限制等）"""
+    """
+    更新独立余额 API Key
+
+    更新指定 ID 的独立余额 API Key 的配置信息。
+
+    **路径参数**:
+    - `key_id`: API Key ID
+
+    **请求体字段**:
+    - `name`: 可选，API Key 的名称
+    - `rate_limit`: 可选，速率限制配置（null 表示无限制）
+    - `allowed_providers`: 可选，允许使用的提供商列表
+    - `allowed_api_formats`: 可选，允许使用的 API 格式列表
+    - `allowed_models`: 可选，允许使用的模型列表
+    - `expire_days`: 可选，过期天数（兼容旧版）
+    - `expires_at`: 可选，过期时间（ISO 格式或 YYYY-MM-DD 格式，优先级高于 expire_days，null 或空字符串表示永不过期）
+    - `auto_delete_on_expiry`: 可选，过期后是否自动删除
+
+    **返回字段**:
+    - `id`: API Key ID
+    - `name`: API Key 名称
+    - `key_display`: 脱敏显示的 Key
+    - `is_active`: 是否启用
+    - `current_balance_usd`: 当前余额
+    - `balance_used_usd`: 已使用余额
+    - `rate_limit`: 速率限制配置
+    - `expires_at`: 过期时间
+    - `updated_at`: 更新时间
+    - `message`: 提示信息
+    """
     adapter = AdminUpdateApiKeyAdapter(key_id=key_id, key_data=key_data)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
 @router.patch("/{key_id}")
 async def toggle_api_key(key_id: str, request: Request, db: Session = Depends(get_db)):
-    """Toggle API key active status (PATCH with is_active in body)"""
+    """
+    切换 API Key 启用状态
+
+    切换指定 API Key 的启用/禁用状态。
+
+    **路径参数**:
+    - `key_id`: API Key ID
+
+    **返回字段**:
+    - `id`: API Key ID
+    - `is_active`: 新的启用状态
+    - `message`: 提示信息
+    """
     adapter = AdminToggleApiKeyAdapter(key_id=key_id)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
 @router.delete("/{key_id}")
 async def delete_api_key(key_id: str, request: Request, db: Session = Depends(get_db)):
+    """
+    删除 API Key
+
+    删除指定的 API Key。此操作不可逆。
+
+    **路径参数**:
+    - `key_id`: API Key ID
+
+    **返回字段**:
+    - `message`: 提示信息
+    """
     adapter = AdminDeleteApiKeyAdapter(key_id=key_id)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -117,7 +216,24 @@ async def add_balance_to_key(
     request: Request,
     db: Session = Depends(get_db),
 ):
-    """Adjust balance for standalone API key (positive to add, negative to deduct)"""
+    """
+    调整独立余额 API Key 的余额
+
+    为指定的独立余额 API Key 增加或扣除余额。
+
+    **路径参数**:
+    - `key_id`: API Key ID
+
+    **请求体字段**:
+    - `amount_usd`: 调整金额（美元），正数为充值，负数为扣除
+
+    **返回字段**:
+    - `id`: API Key ID
+    - `name`: API Key 名称
+    - `current_balance_usd`: 调整后的当前余额
+    - `balance_used_usd`: 已使用余额
+    - `message`: 提示信息
+    """
     # 从请求体获取调整金额
     body = await request.json()
     amount_usd = body.get("amount_usd")
@@ -162,7 +278,24 @@ async def get_api_key_detail(
     include_key: bool = Query(False, description="Include full decrypted key in response"),
     db: Session = Depends(get_db),
 ):
-    """Get API key detail, optionally include full key"""
+    """
+    获取 API Key 详情
+
+    获取指定 API Key 的详细信息。可选择是否返回完整的解密密钥。
+
+    **路径参数**:
+    - `key_id`: API Key ID
+
+    **查询参数**:
+    - `include_key`: 是否包含完整的解密密钥，默认 false
+
+    **返回字段**:
+    - 当 include_key=false 时，返回基本信息：id, user_id, name, key_display, is_active,
+      is_standalone, current_balance_usd, balance_used_usd, total_requests, total_cost_usd,
+      rate_limit, allowed_providers, allowed_api_formats, allowed_models, last_used_at,
+      expires_at, created_at, updated_at
+    - 当 include_key=true 时，返回完整密钥：key
+    """
     if include_key:
         adapter = AdminGetFullKeyAdapter(key_id=key_id)
     else:

src/api/admin/endpoints/__init__.py

@@ -7,7 +7,7 @@ from .health import router as health_router
 from .keys import router as keys_router
 from .routes import router as routes_router
 
-router = APIRouter(prefix="/api/admin/endpoints", tags=["Endpoint Management"])
+router = APIRouter(prefix="/api/admin/endpoints", tags=["Admin - Endpoints"])
 
 # Endpoint CRUD
 router.include_router(routes_router)

src/api/admin/endpoints/concurrency.py

@@ -1,9 +1,8 @@
 """
-Endpoint 并发控制管理 API
+Key RPM 限制管理 API
 """
 
 from dataclasses import dataclass
-from typing import Optional
 
 from fastapi import APIRouter, Depends, Request
 from sqlalchemy.orm import Session
@@ -12,47 +11,56 @@ from src.api.base.admin_adapter import AdminApiAdapter
 from src.api.base.pipeline import ApiRequestPipeline
 from src.core.exceptions import NotFoundException
 from src.database import get_db
-from src.models.database import ProviderAPIKey, ProviderEndpoint
-from src.models.endpoint_models import (
-    ConcurrencyStatusResponse,
-    ResetConcurrencyRequest,
-)
+from src.models.database import ProviderAPIKey
+from src.models.endpoint_models import KeyRpmStatusResponse
 from src.services.rate_limit.concurrency_manager import get_concurrency_manager
 
-router = APIRouter(tags=["Concurrency Control"])
+router = APIRouter(tags=["RPM Control"])
 pipeline = ApiRequestPipeline()
 
 
-@router.get("/concurrency/endpoint/{endpoint_id}", response_model=ConcurrencyStatusResponse)
-async def get_endpoint_concurrency(
-    endpoint_id: str,
-    request: Request,
-    db: Session = Depends(get_db),
-) -> ConcurrencyStatusResponse:
-    """获取 Endpoint 当前并发状态"""
-    adapter = AdminEndpointConcurrencyAdapter(endpoint_id=endpoint_id)
-    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
-
-
-@router.get("/concurrency/key/{key_id}", response_model=ConcurrencyStatusResponse)
-async def get_key_concurrency(
+@router.get("/rpm/key/{key_id}", response_model=KeyRpmStatusResponse)
+async def get_key_rpm(
     key_id: str,
     request: Request,
     db: Session = Depends(get_db),
-) -> ConcurrencyStatusResponse:
-    """获取 Key 当前并发状态"""
-    adapter = AdminKeyConcurrencyAdapter(key_id=key_id)
+) -> KeyRpmStatusResponse:
+    """
+    获取 Key 当前 RPM 状态
+
+    查询指定 API Key 的实时 RPM 使用情况，包括当前 RPM 计数和最大 RPM 限制。
+
+    **路径参数**:
+    - `key_id`: API Key ID
+
+    **返回字段**:
+    - `key_id`: API Key ID
+    - `current_rpm`: 当前 RPM 计数
+    - `rpm_limit`: RPM 限制
+    """
+    adapter = AdminKeyRpmAdapter(key_id=key_id)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
-@router.delete("/concurrency")
-async def reset_concurrency(
-    request: ResetConcurrencyRequest,
+@router.delete("/rpm/key/{key_id}")
+async def reset_key_rpm(
+    key_id: str,
     http_request: Request,
     db: Session = Depends(get_db),
 ) -> dict:
-    """Reset concurrency counters (admin function, use with caution)"""
-    adapter = AdminResetConcurrencyAdapter(endpoint_id=request.endpoint_id, key_id=request.key_id)
+    """
+    重置 Key RPM 计数器
+
+    重置指定 API Key 的 RPM 计数器，用于解决计数不准确的问题。
+    管理员功能，请谨慎使用。
+
+    **路径参数**:
+    - `key_id`: API Key ID
+
+    **返回字段**:
+    - `message`: 操作结果消息
+    """
+    adapter = AdminResetKeyRpmAdapter(key_id=key_id)
     return await pipeline.run(adapter=adapter, http_request=http_request, db=db, mode=adapter.mode)
 
 
@@ -60,31 +68,7 @@ async def reset_concurrency(
 
 
 @dataclass
-class AdminEndpointConcurrencyAdapter(AdminApiAdapter):
-    endpoint_id: str
-
-    async def handle(self, context):  # type: ignore[override]
-        db = context.db
-        endpoint = (
-            db.query(ProviderEndpoint).filter(ProviderEndpoint.id == self.endpoint_id).first()
-        )
-        if not endpoint:
-            raise NotFoundException(f"Endpoint {self.endpoint_id} 不存在")
-
-        concurrency_manager = await get_concurrency_manager()
-        endpoint_count, _ = await concurrency_manager.get_current_concurrency(
-            endpoint_id=self.endpoint_id
-        )
-
-        return ConcurrencyStatusResponse(
-            endpoint_id=self.endpoint_id,
-            endpoint_current_concurrency=endpoint_count,
-            endpoint_max_concurrent=endpoint.max_concurrent,
-        )
-
-
-@dataclass
-class AdminKeyConcurrencyAdapter(AdminApiAdapter):
+class AdminKeyRpmAdapter(AdminApiAdapter):
     key_id: str
 
     async def handle(self, context):  # type: ignore[override]
@@ -94,23 +78,20 @@ class AdminKeyConcurrencyAdapter(AdminApiAdapter):
             raise NotFoundException(f"Key {self.key_id} 不存在")
 
         concurrency_manager = await get_concurrency_manager()
-        _, key_count = await concurrency_manager.get_current_concurrency(key_id=self.key_id)
+        key_count = await concurrency_manager.get_key_rpm_count(key_id=self.key_id)
 
-        return ConcurrencyStatusResponse(
+        return KeyRpmStatusResponse(
             key_id=self.key_id,
-            key_current_concurrency=key_count,
-            key_max_concurrent=key.max_concurrent,
+            current_rpm=key_count,
+            rpm_limit=key.rpm_limit,
         )
 
 
 @dataclass
-class AdminResetConcurrencyAdapter(AdminApiAdapter):
-    endpoint_id: Optional[str]
-    key_id: Optional[str]
+class AdminResetKeyRpmAdapter(AdminApiAdapter):
+    key_id: str
 
     async def handle(self, context):  # type: ignore[override]
         concurrency_manager = await get_concurrency_manager()
-        await concurrency_manager.reset_concurrency(
-            endpoint_id=self.endpoint_id, key_id=self.key_id
-        )
-        return {"message": "并发计数已重置"}
+        await concurrency_manager.reset_key_rpm(key_id=self.key_id)
+        return {"message": "RPM 计数已重置"}

src/api/admin/endpoints/health.py

@@ -5,7 +5,7 @@ Endpoint 健康监控 API
 from collections import defaultdict
 from dataclasses import dataclass
 from datetime import datetime, timedelta, timezone
-from typing import Dict, List
+from typing import Dict, List, Optional
 
 from fastapi import APIRouter, Depends, Query, Request
 from sqlalchemy import func
@@ -36,7 +36,20 @@ async def get_health_summary(
     request: Request,
     db: Session = Depends(get_db),
 ) -> HealthSummaryResponse:
-    """获取健康状态摘要"""
+    """
+    获取健康状态摘要
+
+    获取系统整体健康状态摘要，包括所有 Provider、Endpoint 和 Key 的健康状态统计。
+
+    **返回字段**:
+    - `total_providers`: Provider 总数
+    - `active_providers`: 活跃 Provider 数量
+    - `total_endpoints`: Endpoint 总数
+    - `active_endpoints`: 活跃 Endpoint 数量
+    - `total_keys`: Key 总数
+    - `active_keys`: 活跃 Key 数量
+    - `circuit_breaker_open_keys`: 熔断的 Key 数量
+    """
     adapter = AdminHealthSummaryAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -50,9 +63,21 @@ async def get_endpoint_health_status(
     """
     获取端点健康状态（简化视图，与用户端点统一）
 
+    获取按 API 格式聚合的端点健康状态时间线，基于 Usage 表统计，
+    返回 50 个时间段的聚合状态，适用于快速查看整体健康趋势。
+
     与 /health/api-formats 的区别：
     - /health/status: 返回聚合的时间线状态（50个时间段），基于 Usage 表
     - /health/api-formats: 返回详细的事件列表，基于 RequestCandidate 表
+
+    **查询参数**:
+    - `lookback_hours`: 回溯的小时数（1-72），默认 6
+
+    **返回字段**:
+    - `api_format`: API 格式名称
+    - `timeline`: 时间线数据（50个时间段）
+    - `time_range_start`: 时间范围起始
+    - `time_range_end`: 时间范围结束
     """
     adapter = AdminEndpointHealthStatusAdapter(lookback_hours=lookback_hours)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
@@ -65,7 +90,33 @@ async def get_api_format_health_monitor(
     per_format_limit: int = Query(60, ge=10, le=200, description="每个 API 格式的事件数量"),
     db: Session = Depends(get_db),
 ) -> ApiFormatHealthMonitorResponse:
-    """获取按 API 格式聚合的健康监控时间线（详细事件列表）"""
+    """
+    获取按 API 格式聚合的健康监控时间线（详细事件列表）
+
+    获取每个 API 格式的详细健康监控数据，包括请求事件列表、成功率统计、
+    时间线数据等，基于 RequestCandidate 表查询，适用于详细分析。
+
+    **查询参数**:
+    - `lookback_hours`: 回溯的小时数（1-72），默认 6
+    - `per_format_limit`: 每个 API 格式返回的事件数量（10-200），默认 60
+
+    **返回字段**:
+    - `generated_at`: 数据生成时间
+    - `formats`: API 格式健康监控数据列表
+      - `api_format`: API 格式名称
+      - `total_attempts`: 总请求数
+      - `success_count`: 成功请求数
+      - `failed_count`: 失败请求数
+      - `skipped_count`: 跳过请求数
+      - `success_rate`: 成功率
+      - `provider_count`: Provider 数量
+      - `key_count`: Key 数量
+      - `last_event_at`: 最后事件时间
+      - `events`: 事件列表
+      - `timeline`: 时间线数据
+      - `time_range_start`: 时间范围起始
+      - `time_range_end`: 时间范围结束
+    """
     adapter = AdminApiFormatHealthMonitorAdapter(
         lookback_hours=lookback_hours,
         per_format_limit=per_format_limit,
@@ -77,10 +128,32 @@ async def get_api_format_health_monitor(
 async def get_key_health(
     key_id: str,
     request: Request,
+    api_format: Optional[str] = Query(None, description="API 格式（可选，如 CLAUDE、OPENAI）"),
     db: Session = Depends(get_db),
 ) -> HealthStatusResponse:
-    """获取 Key 健康状态"""
-    adapter = AdminKeyHealthAdapter(key_id=key_id)
+    """
+    获取 Key 健康状态
+
+    获取指定 API Key 的健康状态详情，包括健康分数、连续失败次数、
+    熔断器状态等信息。支持按 API 格式查询。
+
+    **路径参数**:
+    - `key_id`: API Key ID
+
+    **查询参数**:
+    - `api_format`: 可选，指定 API 格式（如 CLAUDE、OPENAI）。
+      - 指定时返回该格式的健康度详情
+      - 不指定时返回所有格式的健康度摘要
+
+    **返回字段**:
+    - `key_id`: API Key ID
+    - `key_health_score`: 健康分数（0.0-1.0）
+    - `key_is_active`: 是否活跃
+    - `key_statistics`: 统计信息
+    - `health_by_format`: 按格式的健康度数据（无 api_format 参数时）
+    - `circuit_breaker_open`: 熔断器是否打开（有 api_format 参数时）
+    """
+    adapter = AdminKeyHealthAdapter(key_id=key_id, api_format=api_format)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
@@ -88,18 +161,31 @@ async def get_key_health(
 async def recover_key_health(
     key_id: str,
     request: Request,
+    api_format: Optional[str] = Query(None, description="API 格式（可选，不指定则恢复所有格式）"),
     db: Session = Depends(get_db),
 ) -> dict:
     """
-    Recover key health status
+    恢复 Key 健康状态
 
-    Resets health_score to 1.0, closes circuit breaker,
-    cancels auto-disable, and resets all failure counts.
+    手动恢复指定 Key 的健康状态，将健康分数重置为 1.0，关闭熔断器，
+    取消自动禁用，并重置所有失败计数。支持按 API 格式恢复。
 
-    Parameters:
-    - key_id: Key ID (path parameter)
+    **路径参数**:
+    - `key_id`: API Key ID
+
+    **查询参数**:
+    - `api_format`: 可选，指定 API 格式（如 CLAUDE、OPENAI）
+      - 指定时仅恢复该格式的健康度
+      - 不指定时恢复所有格式
+
+    **返回字段**:
+    - `message`: 操作结果消息
+    - `details`: 详细信息
+      - `health_score`: 健康分数
+      - `circuit_breaker_open`: 熔断器状态
+      - `is_active`: 是否活跃
     """
-    adapter = AdminRecoverKeyHealthAdapter(key_id=key_id)
+    adapter = AdminRecoverKeyHealthAdapter(key_id=key_id, api_format=api_format)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
@@ -109,12 +195,21 @@ async def recover_all_keys_health(
     db: Session = Depends(get_db),
 ) -> dict:
     """
-    Batch recover all circuit-broken keys
+    批量恢复所有熔断 Key 的健康状态
 
-    Finds all keys with circuit_breaker_open=True and:
-    1. Resets health_score to 1.0
-    2. Closes circuit breaker
-    3. Resets failure counts
+    查找所有处于熔断状态的 Key（circuit_breaker_open=True），
+    并批量执行以下操作：
+    1. 将健康分数重置为 1.0
+    2. 关闭熔断器
+    3. 重置失败计数
+
+    **返回字段**:
+    - `message`: 操作结果消息
+    - `recovered_count`: 恢复的 Key 数量
+    - `recovered_keys`: 恢复的 Key 列表
+      - `key_id`: Key ID
+      - `key_name`: Key 名称
+      - `endpoint_id`: Endpoint ID
     """
     adapter = AdminRecoverAllKeysHealthAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
@@ -190,34 +285,9 @@ class AdminApiFormatHealthMonitorAdapter(AdminApiAdapter):
             )
             all_formats[api_format] = provider_count
 
-        # 1.1 获取所有活跃的 API 格式及其 API Key 数量
-        active_keys = (
-            db.query(
-                ProviderEndpoint.api_format,
-                func.count(ProviderAPIKey.id).label("key_count"),
-            )
-            .join(ProviderAPIKey, ProviderEndpoint.id == ProviderAPIKey.endpoint_id)
-            .join(Provider, ProviderEndpoint.provider_id == Provider.id)
-            .filter(
-                ProviderEndpoint.is_active.is_(True),
-                Provider.is_active.is_(True),
-                ProviderAPIKey.is_active.is_(True),
-            )
-            .group_by(ProviderEndpoint.api_format)
-            .all()
-        )
-
-        # 构建所有格式的 key_count 映射
-        key_counts: Dict[str, int] = {}
-        for api_format_enum, key_count in active_keys:
-            api_format = (
-                api_format_enum.value if hasattr(api_format_enum, "value") else str(api_format_enum)
-            )
-            key_counts[api_format] = key_count
-
-        # 1.2 建立每个 API 格式对应的 Endpoint ID 列表，供 Usage 时间线生成使用
+        # 1.1 建立每个 API 格式对应的 Endpoint ID 列表（用于时间线生成），并收集活跃的 provider+format 组合
         endpoint_rows = (
-            db.query(ProviderEndpoint.api_format, ProviderEndpoint.id)
+            db.query(ProviderEndpoint.api_format, ProviderEndpoint.id, ProviderEndpoint.provider_id)
             .join(Provider, ProviderEndpoint.provider_id == Provider.id)
             .filter(
                 ProviderEndpoint.is_active.is_(True),
@@ -226,11 +296,32 @@ class AdminApiFormatHealthMonitorAdapter(AdminApiAdapter):
             .all()
         )
         endpoint_map: Dict[str, List[str]] = defaultdict(list)
-        for api_format_enum, endpoint_id in endpoint_rows:
+        active_provider_formats: set[tuple[str, str]] = set()
+        for api_format_enum, endpoint_id, provider_id in endpoint_rows:
             api_format = (
                 api_format_enum.value if hasattr(api_format_enum, "value") else str(api_format_enum)
             )
             endpoint_map[api_format].append(endpoint_id)
+            active_provider_formats.add((str(provider_id), api_format))
+
+        # 1.2 统计每个 API 格式可用的活跃 Key 数量（Key 属于 Provider，通过 api_formats 关联格式）
+        key_counts: Dict[str, int] = {}
+        if active_provider_formats:
+            active_provider_keys = (
+                db.query(ProviderAPIKey.provider_id, ProviderAPIKey.api_formats)
+                .join(Provider, ProviderAPIKey.provider_id == Provider.id)
+                .filter(
+                    Provider.is_active.is_(True),
+                    ProviderAPIKey.is_active.is_(True),
+                )
+                .all()
+            )
+            for provider_id, api_formats in active_provider_keys:
+                pid = str(provider_id)
+                for fmt in (api_formats or []):
+                    if (pid, fmt) not in active_provider_formats:
+                        continue
+                    key_counts[fmt] = key_counts.get(fmt, 0) + 1
 
         # 2. 统计窗口内每个 API 格式的请求状态分布（真实统计）
         # 只统计最终状态：success, failed, skipped
@@ -371,28 +462,45 @@ class AdminApiFormatHealthMonitorAdapter(AdminApiAdapter):
 @dataclass
 class AdminKeyHealthAdapter(AdminApiAdapter):
     key_id: str
+    api_format: Optional[str] = None
 
     async def handle(self, context):  # type: ignore[override]
-        health_data = health_monitor.get_key_health(context.db, self.key_id)
+        health_data = health_monitor.get_key_health(context.db, self.key_id, self.api_format)
         if not health_data:
             raise NotFoundException(f"Key {self.key_id} 不存在")
 
-        return HealthStatusResponse(
-            key_id=health_data["key_id"],
-            key_health_score=health_data["health_score"],
-            key_consecutive_failures=health_data["consecutive_failures"],
-            key_last_failure_at=health_data["last_failure_at"],
-            key_is_active=health_data["is_active"],
-            key_statistics=health_data["statistics"],
-            circuit_breaker_open=health_data["circuit_breaker_open"],
-            circuit_breaker_open_at=health_data["circuit_breaker_open_at"],
-            next_probe_at=health_data["next_probe_at"],
-        )
+        # 构建响应
+        response_data = {
+            "key_id": health_data["key_id"],
+            "key_is_active": health_data["is_active"],
+            "key_statistics": health_data.get("statistics"),
+            "key_health_score": health_data.get("health_score", 1.0),
+        }
+
+        if self.api_format:
+            # 单格式查询
+            response_data["api_format"] = self.api_format
+            response_data["key_consecutive_failures"] = health_data.get("consecutive_failures")
+            response_data["key_last_failure_at"] = health_data.get("last_failure_at")
+            circuit = health_data.get("circuit_breaker", {})
+            response_data["circuit_breaker_open"] = circuit.get("open", False)
+            response_data["circuit_breaker_open_at"] = circuit.get("open_at")
+            response_data["next_probe_at"] = circuit.get("next_probe_at")
+            response_data["half_open_until"] = circuit.get("half_open_until")
+            response_data["half_open_successes"] = circuit.get("half_open_successes", 0)
+            response_data["half_open_failures"] = circuit.get("half_open_failures", 0)
+        else:
+            # 全格式查询
+            response_data["any_circuit_open"] = health_data.get("any_circuit_open", False)
+            response_data["health_by_format"] = health_data.get("health_by_format")
+
+        return HealthStatusResponse(**response_data)
 
 
 @dataclass
 class AdminRecoverKeyHealthAdapter(AdminApiAdapter):
     key_id: str
+    api_format: Optional[str] = None
 
     async def handle(self, context):  # type: ignore[override]
         db = context.db
@@ -400,28 +508,38 @@ class AdminRecoverKeyHealthAdapter(AdminApiAdapter):
         if not key:
             raise NotFoundException(f"Key {self.key_id} 不存在")
 
-        key.health_score = 1.0
-        key.consecutive_failures = 0
-        key.last_failure_at = None
-        key.circuit_breaker_open = False
-        key.circuit_breaker_open_at = None
-        key.next_probe_at = None
+        # 使用 health_monitor.reset_health 重置健康度
+        success = health_monitor.reset_health(db, key_id=self.key_id, api_format=self.api_format)
+        if not success:
+            raise Exception("重置健康度失败")
+
+        # 如果 Key 被禁用，重新启用
         if not key.is_active:
-            key.is_active = True
+            key.is_active = True  # type: ignore[assignment]
 
         db.commit()
 
-        admin_name = context.user.username if context.user else "admin"
-        logger.info(f"管理员恢复Key健康状态: {self.key_id} (health_score: 1.0, circuit_breaker: closed)")
-
-        return {
-            "message": "Key已完全恢复",
-            "details": {
-                "health_score": 1.0,
-                "circuit_breaker_open": False,
-                "is_active": True,
-            },
-        }
+        if self.api_format:
+            logger.info(f"管理员恢复Key健康状态: {self.key_id}/{self.api_format}")
+            return {
+                "message": f"Key 的 {self.api_format} 格式已恢复",
+                "details": {
+                    "api_format": self.api_format,
+                    "health_score": 1.0,
+                    "circuit_breaker_open": False,
+                    "is_active": True,
+                },
+            }
+        else:
+            logger.info(f"管理员恢复Key健康状态: {self.key_id} (所有格式)")
+            return {
+                "message": "Key 所有格式已恢复",
+                "details": {
+                    "health_score": 1.0,
+                    "circuit_breaker_open": False,
+                    "is_active": True,
+                },
+            }
 
 
 class AdminRecoverAllKeysHealthAdapter(AdminApiAdapter):
@@ -430,10 +548,17 @@ class AdminRecoverAllKeysHealthAdapter(AdminApiAdapter):
     async def handle(self, context):  # type: ignore[override]
         db = context.db
 
-        # 查找所有熔断的 Key
-        circuit_open_keys = (
-            db.query(ProviderAPIKey).filter(ProviderAPIKey.circuit_breaker_open == True).all()
-        )
+        # 查找所有有熔断格式的 Key（检查 circuit_breaker_by_format JSON 字段）
+        all_keys = db.query(ProviderAPIKey).all()
+
+        # 筛选出有任何格式熔断的 Key
+        circuit_open_keys = []
+        for key in all_keys:
+            circuit_by_format = key.circuit_breaker_by_format or {}
+            for fmt, circuit_data in circuit_by_format.items():
+                if circuit_data.get("open"):
+                    circuit_open_keys.append(key)
+                    break
 
         if not circuit_open_keys:
             return {
@@ -444,17 +569,15 @@ class AdminRecoverAllKeysHealthAdapter(AdminApiAdapter):
 
         recovered_keys = []
         for key in circuit_open_keys:
-            key.health_score = 1.0
-            key.consecutive_failures = 0
-            key.last_failure_at = None
-            key.circuit_breaker_open = False
-            key.circuit_breaker_open_at = None
-            key.next_probe_at = None
+            # 重置所有格式的健康度
+            key.health_by_format = {}  # type: ignore[assignment]
+            key.circuit_breaker_by_format = {}  # type: ignore[assignment]
             recovered_keys.append(
                 {
                     "key_id": key.id,
                     "key_name": key.name,
-                    "endpoint_id": key.endpoint_id,
+                    "provider_id": key.provider_id,
+                    "api_formats": key.api_formats,
                 }
             )
 
@@ -466,7 +589,6 @@ class AdminRecoverAllKeysHealthAdapter(AdminApiAdapter):
         HealthMonitor._open_circuit_keys = 0
         health_open_circuits.set(0)
 
-        admin_name = context.user.username if context.user else "admin"
         logger.info(f"管理员批量恢复 {len(recovered_keys)} 个 Key 的健康状态")
 
         return {

src/api/admin/endpoints/keys.py

@@ -1,5 +1,5 @@
 """
-Endpoint API Keys 管理
+Provider API Keys 管理
 """
 
 import uuid
@@ -12,52 +12,23 @@ from sqlalchemy.orm import Session
 
 from src.api.base.admin_adapter import AdminApiAdapter
 from src.api.base.pipeline import ApiRequestPipeline
+from src.config.constants import RPMDefaults
 from src.core.crypto import crypto_service
 from src.core.exceptions import InvalidRequestException, NotFoundException
 from src.core.key_capabilities import get_capability
 from src.core.logger import logger
 from src.database import get_db
 from src.models.database import Provider, ProviderAPIKey, ProviderEndpoint
+from src.services.cache.provider_cache import ProviderCacheService
 from src.models.endpoint_models import (
-    BatchUpdateKeyPriorityRequest,
     EndpointAPIKeyCreate,
     EndpointAPIKeyResponse,
     EndpointAPIKeyUpdate,
 )
 
-router = APIRouter(tags=["Endpoint Keys"])
+router = APIRouter(tags=["Provider Keys"])
 pipeline = ApiRequestPipeline()
 
-
-@router.get("/{endpoint_id}/keys", response_model=List[EndpointAPIKeyResponse])
-async def list_endpoint_keys(
-    endpoint_id: str,
-    request: Request,
-    skip: int = Query(0, ge=0, description="跳过的记录数"),
-    limit: int = Query(100, ge=1, le=1000, description="返回的最大记录数"),
-    db: Session = Depends(get_db),
-) -> List[EndpointAPIKeyResponse]:
-    """获取 Endpoint 的所有 Keys"""
-    adapter = AdminListEndpointKeysAdapter(
-        endpoint_id=endpoint_id,
-        skip=skip,
-        limit=limit,
-    )
-    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
-
-
-@router.post("/{endpoint_id}/keys", response_model=EndpointAPIKeyResponse)
-async def add_endpoint_key(
-    endpoint_id: str,
-    key_data: EndpointAPIKeyCreate,
-    request: Request,
-    db: Session = Depends(get_db),
-) -> EndpointAPIKeyResponse:
-    """为 Endpoint 添加 Key"""
-    adapter = AdminCreateEndpointKeyAdapter(endpoint_id=endpoint_id, key_data=key_data)
-    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
-
-
 @router.put("/keys/{key_id}", response_model=EndpointAPIKeyResponse)
 async def update_endpoint_key(
     key_id: str,
@@ -65,7 +36,29 @@ async def update_endpoint_key(
     request: Request,
     db: Session = Depends(get_db),
 ) -> EndpointAPIKeyResponse:
-    """更新 Endpoint Key"""
+    """
+    更新 Provider Key
+
+    更新指定 Key 的配置，支持修改并发限制、速率倍数、优先级、
+    配额限制、能力限制等。支持部分更新。
+
+    **路径参数**:
+    - `key_id`: Key ID
+
+    **请求体字段**（均为可选）:
+    - `api_key`: 新的 API Key 原文
+    - `name`: Key 名称
+    - `note`: 备注
+    - `rate_multiplier`: 速率倍数
+    - `internal_priority`: 内部优先级
+    - `rpm_limit`: RPM 限制（设置为 null 可切换到自适应模式）
+    - `allowed_models`: 允许的模型列表
+    - `capabilities`: 能力配置
+    - `is_active`: 是否活跃
+
+    **返回字段**:
+    - 包含更新后的完整 Key 信息
+    """
     adapter = AdminUpdateEndpointKeyAdapter(key_id=key_id, key_data=key_data)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -75,7 +68,31 @@ async def get_keys_grouped_by_format(
     request: Request,
     db: Session = Depends(get_db),
 ) -> dict:
-    """获取按 API 格式分组的所有 Keys（用于全局优先级管理）"""
+    """
+    获取按 API 格式分组的所有 Keys
+
+    获取所有活跃的 Key，按 API 格式分组返回，用于全局优先级管理。
+    每个 Key 包含基本信息、健康度指标、能力标签等。
+
+    **返回字段**:
+    - 返回一个字典，键为 API 格式，值为该格式下的 Key 列表
+    - 每个 Key 包含：
+      - `id`: Key ID
+      - `name`: Key 名称
+      - `api_key_masked`: 脱敏后的 API Key
+      - `internal_priority`: 内部优先级
+      - `global_priority`: 全局优先级
+      - `rate_multiplier`: 速率倍数
+      - `is_active`: 是否活跃
+      - `circuit_breaker_open`: 熔断器状态
+      - `provider_name`: Provider 名称
+      - `endpoint_base_url`: Endpoint 基础 URL
+      - `api_format`: API 格式
+      - `capabilities`: 能力简称列表
+      - `success_rate`: 成功率
+      - `avg_response_time_ms`: 平均响应时间
+      - `request_count`: 请求总数
+    """
     adapter = AdminGetKeysGroupedByFormatAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -86,7 +103,18 @@ async def reveal_endpoint_key(
     request: Request,
     db: Session = Depends(get_db),
 ) -> dict:
-    """获取完整的 API Key（用于查看和复制）"""
+    """
+    获取完整的 API Key
+
+    解密并返回指定 Key 的完整原文，用于查看和复制。
+    此操作会被记录到审计日志。
+
+    **路径参数**:
+    - `key_id`: Key ID
+
+    **返回字段**:
+    - `api_key`: 完整的 API Key 原文
+    """
     adapter = AdminRevealEndpointKeyAdapter(key_id=key_id)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -97,151 +125,81 @@ async def delete_endpoint_key(
     request: Request,
     db: Session = Depends(get_db),
 ) -> dict:
-    """删除 Endpoint Key"""
+    """
+    删除 Provider Key
+
+    删除指定的 API Key。此操作不可逆，请谨慎使用。
+
+    **路径参数**:
+    - `key_id`: Key ID
+
+    **返回字段**:
+    - `message`: 操作结果消息
+    """
     adapter = AdminDeleteEndpointKeyAdapter(key_id=key_id)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
-@router.put("/{endpoint_id}/keys/batch-priority")
-async def batch_update_key_priority(
-    endpoint_id: str,
+# ========== Provider Keys API ==========
+
+
+@router.get("/providers/{provider_id}/keys", response_model=List[EndpointAPIKeyResponse])
+async def list_provider_keys(
+    provider_id: str,
     request: Request,
-    priority_data: BatchUpdateKeyPriorityRequest,
+    skip: int = Query(0, ge=0, description="跳过的记录数"),
+    limit: int = Query(100, ge=1, le=1000, description="返回的最大记录数"),
     db: Session = Depends(get_db),
-) -> dict:
-    """批量更新 Endpoint 下 Keys 的优先级（用于拖动排序）"""
-    adapter = AdminBatchUpdateKeyPriorityAdapter(endpoint_id=endpoint_id, priority_data=priority_data)
+) -> List[EndpointAPIKeyResponse]:
+    """
+    获取 Provider 的所有 Keys
+
+    获取指定 Provider 下的所有 API Key 列表，支持多 API 格式。
+    结果按优先级和创建时间排序。
+
+    **路径参数**:
+    - `provider_id`: Provider ID
+
+    **查询参数**:
+    - `skip`: 跳过的记录数，用于分页（默认 0）
+    - `limit`: 返回的最大记录数（1-1000，默认 100）
+    """
+    adapter = AdminListProviderKeysAdapter(
+        provider_id=provider_id,
+        skip=skip,
+        limit=limit,
+    )
+    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
+
+
+@router.post("/providers/{provider_id}/keys", response_model=EndpointAPIKeyResponse)
+async def add_provider_key(
+    provider_id: str,
+    key_data: EndpointAPIKeyCreate,
+    request: Request,
+    db: Session = Depends(get_db),
+) -> EndpointAPIKeyResponse:
+    """
+    为 Provider 添加 Key
+
+    为指定 Provider 添加新的 API Key，支持配置多个 API 格式。
+
+    **路径参数**:
+    - `provider_id`: Provider ID
+
+    **请求体字段**:
+    - `api_formats`: 支持的 API 格式列表（必填）
+    - `api_key`: API Key 原文（将被加密存储）
+    - `name`: Key 名称
+    - 其他配置字段同 Key
+    """
+    adapter = AdminCreateProviderKeyAdapter(provider_id=provider_id, key_data=key_data)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
 # -------- Adapters --------
 
 
-@dataclass
-class AdminListEndpointKeysAdapter(AdminApiAdapter):
-    endpoint_id: str
-    skip: int
-    limit: int
-
-    async def handle(self, context):  # type: ignore[override]
-        db = context.db
-        endpoint = (
-            db.query(ProviderEndpoint).filter(ProviderEndpoint.id == self.endpoint_id).first()
-        )
-        if not endpoint:
-            raise NotFoundException(f"Endpoint {self.endpoint_id} 不存在")
-
-        keys = (
-            db.query(ProviderAPIKey)
-            .filter(ProviderAPIKey.endpoint_id == self.endpoint_id)
-            .order_by(ProviderAPIKey.internal_priority.asc(), ProviderAPIKey.created_at.asc())
-            .offset(self.skip)
-            .limit(self.limit)
-            .all()
-        )
-
-        result: List[EndpointAPIKeyResponse] = []
-        for key in keys:
-            try:
-                decrypted_key = crypto_service.decrypt(key.api_key)
-                masked_key = f"{decrypted_key[:8]}***{decrypted_key[-4:]}"
-            except Exception:
-                masked_key = "***ERROR***"
-
-            success_rate = key.success_count / key.request_count if key.request_count > 0 else 0.0
-            avg_response_time_ms = (
-                key.total_response_time_ms / key.success_count if key.success_count > 0 else 0.0
-            )
-
-            is_adaptive = key.max_concurrent is None
-            key_dict = key.__dict__.copy()
-            key_dict.pop("_sa_instance_state", None)
-            key_dict.update(
-                {
-                    "api_key_masked": masked_key,
-                    "api_key_plain": None,
-                    "success_rate": success_rate,
-                    "avg_response_time_ms": round(avg_response_time_ms, 2),
-                    "is_adaptive": is_adaptive,
-                    "effective_limit": (
-                        key.learned_max_concurrent if is_adaptive else key.max_concurrent
-                    ),
-                }
-            )
-            result.append(EndpointAPIKeyResponse(**key_dict))
-
-        return result
-
-
-@dataclass
-class AdminCreateEndpointKeyAdapter(AdminApiAdapter):
-    endpoint_id: str
-    key_data: EndpointAPIKeyCreate
-
-    async def handle(self, context):  # type: ignore[override]
-        db = context.db
-        endpoint = (
-            db.query(ProviderEndpoint).filter(ProviderEndpoint.id == self.endpoint_id).first()
-        )
-        if not endpoint:
-            raise NotFoundException(f"Endpoint {self.endpoint_id} 不存在")
-
-        if self.key_data.endpoint_id != self.endpoint_id:
-            raise InvalidRequestException("endpoint_id 不匹配")
-
-        encrypted_key = crypto_service.encrypt(self.key_data.api_key)
-        now = datetime.now(timezone.utc)
-        # max_concurrent=NULL 表示自适应模式，数字表示固定限制
-        new_key = ProviderAPIKey(
-            id=str(uuid.uuid4()),
-            endpoint_id=self.endpoint_id,
-            api_key=encrypted_key,
-            name=self.key_data.name,
-            note=self.key_data.note,
-            rate_multiplier=self.key_data.rate_multiplier,
-            internal_priority=self.key_data.internal_priority,
-            max_concurrent=self.key_data.max_concurrent,  # NULL=自适应模式
-            rate_limit=self.key_data.rate_limit,
-            daily_limit=self.key_data.daily_limit,
-            monthly_limit=self.key_data.monthly_limit,
-            allowed_models=self.key_data.allowed_models if self.key_data.allowed_models else None,
-            capabilities=self.key_data.capabilities if self.key_data.capabilities else None,
-            request_count=0,
-            success_count=0,
-            error_count=0,
-            total_response_time_ms=0,
-            is_active=True,
-            last_used_at=None,
-            created_at=now,
-            updated_at=now,
-        )
-
-        db.add(new_key)
-        db.commit()
-        db.refresh(new_key)
-
-        logger.info(f"[OK] 添加 Key: Endpoint={self.endpoint_id}, Key=***{self.key_data.api_key[-4:]}, ID={new_key.id}")
-
-        masked_key = f"{self.key_data.api_key[:8]}***{self.key_data.api_key[-4:]}"
-        is_adaptive = new_key.max_concurrent is None
-        response_dict = new_key.__dict__.copy()
-        response_dict.pop("_sa_instance_state", None)
-        response_dict.update(
-            {
-                "api_key_masked": masked_key,
-                "api_key_plain": self.key_data.api_key,
-                "success_rate": 0.0,
-                "avg_response_time_ms": 0.0,
-                "is_adaptive": is_adaptive,
-                "effective_limit": (
-                    new_key.learned_max_concurrent if is_adaptive else new_key.max_concurrent
-                ),
-            }
-        )
-
-        return EndpointAPIKeyResponse(**response_dict)
-
-
 @dataclass
 class AdminUpdateEndpointKeyAdapter(AdminApiAdapter):
     key_id: str
@@ -257,14 +215,21 @@ class AdminUpdateEndpointKeyAdapter(AdminApiAdapter):
         if "api_key" in update_data:
             update_data["api_key"] = crypto_service.encrypt(update_data["api_key"])
 
-        # 特殊处理 max_concurrent：需要区分"未提供"和"显式设置为 null"
-        # 当 max_concurrent 被显式设置时（在 model_fields_set 中），即使值为 None 也应该更新
-        if "max_concurrent" in self.key_data.model_fields_set:
-            update_data["max_concurrent"] = self.key_data.max_concurrent
-            # 切换到自适应模式时，清空学习到的并发限制，让系统重新学习
-            if self.key_data.max_concurrent is None:
-                update_data["learned_max_concurrent"] = None
-                logger.info("Key %s 切换为自适应并发模式", self.key_id)
+        # 特殊处理 rpm_limit：需要区分"未提供"和"显式设置为 null"
+        if "rpm_limit" in self.key_data.model_fields_set:
+            update_data["rpm_limit"] = self.key_data.rpm_limit
+            if self.key_data.rpm_limit is None:
+                update_data["learned_rpm_limit"] = None
+                logger.info("Key %s 切换为自适应 RPM 模式", self.key_id)
+
+        # 统一处理 allowed_models：空列表/空字典 -> None（表示不限制）
+        if "allowed_models" in update_data:
+            am = update_data["allowed_models"]
+            if am is not None and (
+                (isinstance(am, list) and len(am) == 0)
+                or (isinstance(am, dict) and len(am) == 0)
+            ):
+                update_data["allowed_models"] = None
 
         for field, value in update_data.items():
             setattr(key, field, value)
@@ -273,35 +238,13 @@ class AdminUpdateEndpointKeyAdapter(AdminApiAdapter):
         db.commit()
         db.refresh(key)
 
+        # 任何字段更新都清除缓存，确保缓存一致性
+        # 包括 is_active、allowed_models、capabilities 等影响权限和行为的字段
+        await ProviderCacheService.invalidate_provider_api_key_cache(self.key_id)
+
         logger.info("[OK] 更新 Key: ID=%s, Updates=%s", self.key_id, list(update_data.keys()))
 
-        try:
-            decrypted_key = crypto_service.decrypt(key.api_key)
-            masked_key = f"{decrypted_key[:8]}***{decrypted_key[-4:]}"
-        except Exception:
-            masked_key = "***ERROR***"
-
-        success_rate = key.success_count / key.request_count if key.request_count > 0 else 0.0
-        avg_response_time_ms = (
-            key.total_response_time_ms / key.success_count if key.success_count > 0 else 0.0
-        )
-
-        is_adaptive = key.max_concurrent is None
-        response_dict = key.__dict__.copy()
-        response_dict.pop("_sa_instance_state", None)
-        response_dict.update(
-            {
-                "api_key_masked": masked_key,
-                "api_key_plain": None,
-                "success_rate": success_rate,
-                "avg_response_time_ms": round(avg_response_time_ms, 2),
-                "is_adaptive": is_adaptive,
-                "effective_limit": (
-                    key.learned_max_concurrent if is_adaptive else key.max_concurrent
-                ),
-            }
-        )
-        return EndpointAPIKeyResponse(**response_dict)
+        return _build_key_response(key)
 
 
 @dataclass
@@ -338,7 +281,7 @@ class AdminDeleteEndpointKeyAdapter(AdminApiAdapter):
         if not key:
             raise NotFoundException(f"Key {self.key_id} 不存在")
 
-        endpoint_id = key.endpoint_id
+        provider_id = key.provider_id
         try:
             db.delete(key)
             db.commit()
@@ -347,7 +290,7 @@ class AdminDeleteEndpointKeyAdapter(AdminApiAdapter):
             logger.error(f"删除 Key 失败: ID={self.key_id}, Error={exc}")
             raise
 
-        logger.warning(f"[DELETE] 删除 Key: ID={self.key_id}, Endpoint={endpoint_id}")
+        logger.warning(f"[DELETE] 删除 Key: ID={self.key_id}, Provider={provider_id}")
         return {"message": f"Key {self.key_id} 已删除"}
 
 
@@ -355,31 +298,51 @@ class AdminGetKeysGroupedByFormatAdapter(AdminApiAdapter):
     async def handle(self, context):  # type: ignore[override]
         db = context.db
 
+        # Key 属于 Provider：按 key.api_formats 分组展示
         keys = (
-            db.query(ProviderAPIKey, ProviderEndpoint, Provider)
-            .join(ProviderEndpoint, ProviderAPIKey.endpoint_id == ProviderEndpoint.id)
-            .join(Provider, ProviderEndpoint.provider_id == Provider.id)
+            db.query(ProviderAPIKey, Provider)
+            .join(Provider, ProviderAPIKey.provider_id == Provider.id)
             .filter(
                 ProviderAPIKey.is_active.is_(True),
-                ProviderEndpoint.is_active.is_(True),
                 Provider.is_active.is_(True),
             )
             .order_by(
-                ProviderAPIKey.global_priority.asc().nullslast(), ProviderAPIKey.internal_priority.asc()
+                ProviderAPIKey.global_priority.asc().nullslast(),
+                ProviderAPIKey.internal_priority.asc(),
             )
             .all()
         )
 
+        provider_ids = {str(provider.id) for _key, provider in keys}
+        endpoints = (
+            db.query(
+                ProviderEndpoint.provider_id,
+                ProviderEndpoint.api_format,
+                ProviderEndpoint.base_url,
+            )
+            .filter(
+                ProviderEndpoint.provider_id.in_(provider_ids),
+                ProviderEndpoint.is_active.is_(True),
+            )
+            .all()
+        )
+        endpoint_base_url_map: Dict[tuple[str, str], str] = {}
+        for provider_id, api_format, base_url in endpoints:
+            fmt = api_format.value if hasattr(api_format, "value") else str(api_format)
+            endpoint_base_url_map[(str(provider_id), fmt)] = base_url
+
         grouped: Dict[str, List[dict]] = {}
-        for key, endpoint, provider in keys:
-            api_format = endpoint.api_format
-            if api_format not in grouped:
-                grouped[api_format] = []
+        for key, provider in keys:
+            api_formats = key.api_formats or []
+
+            if not api_formats:
+                continue  # 跳过没有 API 格式的 Key
 
             try:
                 decrypted_key = crypto_service.decrypt(key.api_key)
                 masked_key = f"{decrypted_key[:8]}***{decrypted_key[-4:]}"
-            except Exception:
+            except Exception as e:
+                logger.error(f"解密 Key 失败: key_id={key.id}, error={e}")
                 masked_key = "***ERROR***"
 
             # 计算健康度指标
@@ -398,72 +361,209 @@ class AdminGetKeysGroupedByFormatAdapter(AdminApiAdapter):
                         cap_def = get_capability(cap_name)
                         caps_list.append(cap_def.short_name if cap_def else cap_name)
 
-            grouped[api_format].append(
-                {
-                    "id": key.id,
-                    "name": key.name,
-                    "api_key_masked": masked_key,
-                    "internal_priority": key.internal_priority,
-                    "global_priority": key.global_priority,
-                    "rate_multiplier": key.rate_multiplier,
-                    "is_active": key.is_active,
-                    "circuit_breaker_open": key.circuit_breaker_open,
-                    "provider_name": provider.display_name or provider.name,
-                    "endpoint_base_url": endpoint.base_url,
-                    "api_format": api_format,
-                    "capabilities": caps_list,
-                    "success_rate": success_rate,
-                    "avg_response_time_ms": avg_response_time_ms,
-                    "request_count": key.request_count,
-                }
-            )
+            # 构建 Key 信息（基础数据）
+            key_info = {
+                "id": key.id,
+                "name": key.name,
+                "api_key_masked": masked_key,
+                "internal_priority": key.internal_priority,
+                "global_priority": key.global_priority,
+                "rate_multiplier": key.rate_multiplier,
+                "is_active": key.is_active,
+                "provider_name": provider.name,
+                "api_formats": api_formats,
+                "capabilities": caps_list,
+                "success_rate": success_rate,
+                "avg_response_time_ms": avg_response_time_ms,
+                "request_count": key.request_count,
+            }
+
+            # 将 Key 添加到每个支持的格式分组中，并附加格式特定的健康度数据
+            health_by_format = key.health_by_format or {}
+            circuit_by_format = key.circuit_breaker_by_format or {}
+            provider_id = str(provider.id)
+            for api_format in api_formats:
+                if api_format not in grouped:
+                    grouped[api_format] = []
+                # 为每个格式创建副本，设置当前格式
+                format_key_info = key_info.copy()
+                format_key_info["api_format"] = api_format
+                format_key_info["endpoint_base_url"] = endpoint_base_url_map.get(
+                    (provider_id, api_format)
+                )
+                # 添加格式特定的健康度数据
+                format_health = health_by_format.get(api_format, {})
+                format_circuit = circuit_by_format.get(api_format, {})
+                format_key_info["health_score"] = float(format_health.get("health_score") or 1.0)
+                format_key_info["circuit_breaker_open"] = bool(format_circuit.get("open", False))
+                grouped[api_format].append(format_key_info)
 
         # 直接返回分组对象，供前端使用
         return grouped
 
 
+# ========== Adapters ==========
+
+
+def _build_key_response(
+    key: ProviderAPIKey, api_key_plain: str | None = None
+) -> EndpointAPIKeyResponse:
+    """构建 Key 响应对象的辅助函数"""
+    try:
+        decrypted_key = crypto_service.decrypt(key.api_key)
+        masked_key = f"{decrypted_key[:8]}***{decrypted_key[-4:]}"
+    except Exception:
+        masked_key = "***ERROR***"
+
+    success_rate = key.success_count / key.request_count if key.request_count > 0 else 0.0
+    avg_response_time_ms = (
+        key.total_response_time_ms / key.success_count if key.success_count > 0 else 0.0
+    )
+
+    is_adaptive = key.rpm_limit is None
+    key_dict = key.__dict__.copy()
+    key_dict.pop("_sa_instance_state", None)
+
+    # 从 health_by_format 计算汇总字段（便于列表展示）
+    health_by_format = key.health_by_format or {}
+    circuit_by_format = key.circuit_breaker_by_format or {}
+
+    # 计算整体健康度（取所有格式中的最低值）
+    if health_by_format:
+        health_scores = [
+            float(h.get("health_score") or 1.0) for h in health_by_format.values()
+        ]
+        min_health_score = min(health_scores) if health_scores else 1.0
+        # 取最大的连续失败次数
+        max_consecutive = max(
+            (int(h.get("consecutive_failures") or 0) for h in health_by_format.values()),
+            default=0,
+        )
+        # 取最近的失败时间
+        failure_times = [
+            h.get("last_failure_at")
+            for h in health_by_format.values()
+            if h.get("last_failure_at")
+        ]
+        last_failure = max(failure_times) if failure_times else None
+    else:
+        min_health_score = 1.0
+        max_consecutive = 0
+        last_failure = None
+
+    # 检查是否有任何格式的熔断器打开
+    any_circuit_open = any(c.get("open", False) for c in circuit_by_format.values())
+
+    key_dict.update(
+        {
+            "api_key_masked": masked_key,
+            "api_key_plain": api_key_plain,
+            "success_rate": success_rate,
+            "avg_response_time_ms": round(avg_response_time_ms, 2),
+            "is_adaptive": is_adaptive,
+            "effective_limit": (
+                (key.learned_rpm_limit if key.learned_rpm_limit is not None else RPMDefaults.INITIAL_LIMIT)
+                if is_adaptive
+                else key.rpm_limit
+            ),
+            # 汇总字段
+            "health_score": min_health_score,
+            "consecutive_failures": max_consecutive,
+            "last_failure_at": last_failure,
+            "circuit_breaker_open": any_circuit_open,
+        }
+    )
+
+    # 防御性：确保 api_formats 存在（历史数据可能为空/缺失）
+    if "api_formats" not in key_dict or key_dict["api_formats"] is None:
+        key_dict["api_formats"] = []
+
+    return EndpointAPIKeyResponse(**key_dict)
+
+
 @dataclass
-class AdminBatchUpdateKeyPriorityAdapter(AdminApiAdapter):
-    endpoint_id: str
-    priority_data: BatchUpdateKeyPriorityRequest
+class AdminListProviderKeysAdapter(AdminApiAdapter):
+    """获取 Provider 的所有 Keys"""
+
+    provider_id: str
+    skip: int
+    limit: int
 
     async def handle(self, context):  # type: ignore[override]
         db = context.db
-        endpoint = (
-            db.query(ProviderEndpoint).filter(ProviderEndpoint.id == self.endpoint_id).first()
-        )
-        if not endpoint:
-            raise NotFoundException(f"Endpoint {self.endpoint_id} 不存在")
+        provider = db.query(Provider).filter(Provider.id == self.provider_id).first()
+        if not provider:
+            raise NotFoundException(f"Provider {self.provider_id} 不存在")
 
-        # 获取所有需要更新的 Key ID
-        key_ids = [item.key_id for item in self.priority_data.priorities]
-
-        # 验证所有 Key 都属于该 Endpoint
         keys = (
             db.query(ProviderAPIKey)
-            .filter(
-                ProviderAPIKey.id.in_(key_ids),
-                ProviderAPIKey.endpoint_id == self.endpoint_id,
-            )
+            .filter(ProviderAPIKey.provider_id == self.provider_id)
+            .order_by(ProviderAPIKey.internal_priority.asc(), ProviderAPIKey.created_at.asc())
+            .offset(self.skip)
+            .limit(self.limit)
             .all()
         )
 
-        if len(keys) != len(key_ids):
-            found_ids = {k.id for k in keys}
-            missing_ids = set(key_ids) - found_ids
-            raise InvalidRequestException(f"Keys 不属于该 Endpoint 或不存在: {missing_ids}")
+        return [_build_key_response(key) for key in keys]
 
-        # 批量更新优先级
-        key_map = {k.id: k for k in keys}
-        updated_count = 0
-        for item in self.priority_data.priorities:
-            key = key_map.get(item.key_id)
-            if key and key.internal_priority != item.internal_priority:
-                key.internal_priority = item.internal_priority
-                key.updated_at = datetime.now(timezone.utc)
-                updated_count += 1
 
+@dataclass
+class AdminCreateProviderKeyAdapter(AdminApiAdapter):
+    """为 Provider 添加 Key"""
+
+    provider_id: str
+    key_data: EndpointAPIKeyCreate
+
+    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(f"Provider {self.provider_id} 不存在")
+
+        # 验证 api_formats 必填
+        if not self.key_data.api_formats:
+            raise InvalidRequestException("api_formats 为必填字段")
+
+        # 允许同一个 API Key 在同一 Provider 下添加多次
+        # 用户可以为不同的 API 格式创建独立的配置记录，便于分开管理
+
+        encrypted_key = crypto_service.encrypt(self.key_data.api_key)
+        now = datetime.now(timezone.utc)
+
+        new_key = ProviderAPIKey(
+            id=str(uuid.uuid4()),
+            provider_id=self.provider_id,
+            api_formats=self.key_data.api_formats,
+            api_key=encrypted_key,
+            name=self.key_data.name,
+            note=self.key_data.note,
+            rate_multiplier=self.key_data.rate_multiplier,
+            rate_multipliers=self.key_data.rate_multipliers,  # 按 API 格式的成本倍率
+            internal_priority=self.key_data.internal_priority,
+            rpm_limit=self.key_data.rpm_limit,
+            allowed_models=self.key_data.allowed_models if self.key_data.allowed_models else None,
+            capabilities=self.key_data.capabilities if self.key_data.capabilities else None,
+            cache_ttl_minutes=self.key_data.cache_ttl_minutes,
+            max_probe_interval_minutes=self.key_data.max_probe_interval_minutes,
+            request_count=0,
+            success_count=0,
+            error_count=0,
+            total_response_time_ms=0,
+            health_by_format={},  # 按格式存储健康度
+            circuit_breaker_by_format={},  # 按格式存储熔断器状态
+            is_active=True,
+            last_used_at=None,
+            created_at=now,
+            updated_at=now,
+        )
+
+        db.add(new_key)
         db.commit()
+        db.refresh(new_key)
 
-        logger.info(f"[OK] 批量更新 Key 优先级: Endpoint={self.endpoint_id}, Updated={updated_count}/{len(key_ids)}")
-        return {"message": f"已更新 {updated_count} 个 Key 的优先级", "updated_count": updated_count}
+        logger.info(
+            f"[OK] 添加 Key: Provider={self.provider_id}, "
+            f"Formats={self.key_data.api_formats}, Key=***{self.key_data.api_key[-4:]}, ID={new_key.id}"
+        )
+
+        return _build_key_response(new_key, api_key_plain=self.key_data.api_key)

src/api/admin/endpoints/routes.py

@@ -45,7 +45,34 @@ async def list_provider_endpoints(
     limit: int = Query(100, ge=1, le=1000, description="返回的最大记录数"),
     db: Session = Depends(get_db),
 ) -> List[ProviderEndpointResponse]:
-    """获取指定 Provider 的所有 Endpoints"""
+    """
+    获取指定 Provider 的所有 Endpoints
+
+    获取指定 Provider 下的所有 Endpoint 列表，包括配置、统计信息等。
+    结果按创建时间倒序排列。
+
+    **路径参数**:
+    - `provider_id`: Provider ID
+
+    **查询参数**:
+    - `skip`: 跳过的记录数，用于分页（默认 0）
+    - `limit`: 返回的最大记录数（1-1000，默认 100）
+
+    **返回字段**:
+    - `id`: Endpoint ID
+    - `provider_id`: Provider ID
+    - `provider_name`: Provider 名称
+    - `api_format`: API 格式
+    - `base_url`: 基础 URL
+    - `custom_path`: 自定义路径
+    - `timeout`: 超时时间（秒）
+    - `max_retries`: 最大重试次数
+    - `is_active`: 是否活跃
+    - `total_keys`: Key 总数
+    - `active_keys`: 活跃 Key 数量
+    - `proxy`: 代理配置（密码已脱敏）
+    - 其他配置字段
+    """
     adapter = AdminListProviderEndpointsAdapter(
         provider_id=provider_id,
         skip=skip,
@@ -61,7 +88,29 @@ async def create_provider_endpoint(
     request: Request,
     db: Session = Depends(get_db),
 ) -> ProviderEndpointResponse:
-    """为 Provider 创建新的 Endpoint"""
+    """
+    为 Provider 创建新的 Endpoint
+
+    为指定 Provider 创建新的 Endpoint，每个 Provider 的每种 API 格式
+    只能创建一个 Endpoint。
+
+    **路径参数**:
+    - `provider_id`: Provider ID
+
+    **请求体字段**:
+    - `provider_id`: Provider ID（必须与路径参数一致）
+    - `api_format`: API 格式（如 claude、openai、gemini 等）
+    - `base_url`: 基础 URL
+    - `custom_path`: 自定义路径（可选）
+    - `headers`: 自定义请求头（可选）
+    - `timeout`: 超时时间（秒，默认 300）
+    - `max_retries`: 最大重试次数（默认 2）
+    - `config`: 额外配置（可选）
+    - `proxy`: 代理配置（可选）
+
+    **返回字段**:
+    - 包含完整的 Endpoint 信息
+    """
     adapter = AdminCreateProviderEndpointAdapter(
         provider_id=provider_id,
         endpoint_data=endpoint_data,
@@ -75,7 +124,29 @@ async def get_endpoint(
     request: Request,
     db: Session = Depends(get_db),
 ) -> ProviderEndpointResponse:
-    """获取 Endpoint 详情"""
+    """
+    获取 Endpoint 详情
+
+    获取指定 Endpoint 的详细信息，包括配置、统计信息等。
+
+    **路径参数**:
+    - `endpoint_id`: Endpoint ID
+
+    **返回字段**:
+    - `id`: Endpoint ID
+    - `provider_id`: Provider ID
+    - `provider_name`: Provider 名称
+    - `api_format`: API 格式
+    - `base_url`: 基础 URL
+    - `custom_path`: 自定义路径
+    - `timeout`: 超时时间（秒）
+    - `max_retries`: 最大重试次数
+    - `is_active`: 是否活跃
+    - `total_keys`: Key 总数
+    - `active_keys`: 活跃 Key 数量
+    - `proxy`: 代理配置（密码已脱敏）
+    - 其他配置字段
+    """
     adapter = AdminGetProviderEndpointAdapter(endpoint_id=endpoint_id)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -87,7 +158,27 @@ async def update_endpoint(
     request: Request,
     db: Session = Depends(get_db),
 ) -> ProviderEndpointResponse:
-    """更新 Endpoint"""
+    """
+    更新 Endpoint
+
+    更新指定 Endpoint 的配置。支持部分更新。
+
+    **路径参数**:
+    - `endpoint_id`: Endpoint ID
+
+    **请求体字段**（均为可选）:
+    - `base_url`: 基础 URL
+    - `custom_path`: 自定义路径
+    - `headers`: 自定义请求头
+    - `timeout`: 超时时间（秒）
+    - `max_retries`: 最大重试次数
+    - `is_active`: 是否活跃
+    - `config`: 额外配置
+    - `proxy`: 代理配置（设置为 null 可清除代理）
+
+    **返回字段**:
+    - 包含更新后的完整 Endpoint 信息
+    """
     adapter = AdminUpdateProviderEndpointAdapter(
         endpoint_id=endpoint_id,
         endpoint_data=endpoint_data,
@@ -101,7 +192,19 @@ async def delete_endpoint(
     request: Request,
     db: Session = Depends(get_db),
 ) -> dict:
-    """删除 Endpoint（级联删除所有关联的 Keys）"""
+    """
+    删除 Endpoint
+
+    删除指定的 Endpoint，会影响该 Provider 在该 API 格式下的路由能力。
+    Key 不会被删除，但包含该 API 格式的 Key 将无法被调度使用（直到重新创建该格式的 Endpoint）。
+
+    **路径参数**:
+    - `endpoint_id`: Endpoint ID
+
+    **返回字段**:
+    - `message`: 操作结果消息
+    - `affected_keys_count`: 受影响的 Key 数量（包含该 API 格式）
+    """
     adapter = AdminDeleteProviderEndpointAdapter(endpoint_id=endpoint_id)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -130,39 +233,33 @@ class AdminListProviderEndpointsAdapter(AdminApiAdapter):
             .all()
         )
 
-        endpoint_ids = [ep.id for ep in endpoints]
-        total_keys_map = {}
-        active_keys_map = {}
-        if endpoint_ids:
-            total_rows = (
-                db.query(ProviderAPIKey.endpoint_id, func.count(ProviderAPIKey.id).label("total"))
-                .filter(ProviderAPIKey.endpoint_id.in_(endpoint_ids))
-                .group_by(ProviderAPIKey.endpoint_id)
-                .all()
-            )
-            total_keys_map = {row.endpoint_id: row.total for row in total_rows}
-
-            active_rows = (
-                db.query(ProviderAPIKey.endpoint_id, func.count(ProviderAPIKey.id).label("active"))
-                .filter(
-                    and_(
-                        ProviderAPIKey.endpoint_id.in_(endpoint_ids),
-                        ProviderAPIKey.is_active.is_(True),
-                    )
-                )
-                .group_by(ProviderAPIKey.endpoint_id)
-                .all()
-            )
-            active_keys_map = {row.endpoint_id: row.active for row in active_rows}
+        # Key 是 Provider 级别资源：按 key.api_formats 归类到各 Endpoint.api_format 下
+        keys = (
+            db.query(ProviderAPIKey.api_formats, ProviderAPIKey.is_active)
+            .filter(ProviderAPIKey.provider_id == self.provider_id)
+            .all()
+        )
+        total_keys_map: dict[str, int] = {}
+        active_keys_map: dict[str, int] = {}
+        for api_formats, is_active in keys:
+            for fmt in (api_formats or []):
+                total_keys_map[fmt] = total_keys_map.get(fmt, 0) + 1
+                if is_active:
+                    active_keys_map[fmt] = active_keys_map.get(fmt, 0) + 1
 
         result: List[ProviderEndpointResponse] = []
         for endpoint in endpoints:
+            endpoint_format = (
+                endpoint.api_format
+                if isinstance(endpoint.api_format, str)
+                else endpoint.api_format.value
+            )
             endpoint_dict = {
                 **endpoint.__dict__,
                 "provider_name": provider.name,
                 "api_format": endpoint.api_format,
-                "total_keys": total_keys_map.get(endpoint.id, 0),
-                "active_keys": active_keys_map.get(endpoint.id, 0),
+                "total_keys": total_keys_map.get(endpoint_format, 0),
+                "active_keys": active_keys_map.get(endpoint_format, 0),
                 "proxy": mask_proxy_password(endpoint.proxy),
             }
             endpoint_dict.pop("_sa_instance_state", None)
@@ -210,8 +307,6 @@ class AdminCreateProviderEndpointAdapter(AdminApiAdapter):
             headers=self.endpoint_data.headers,
             timeout=self.endpoint_data.timeout,
             max_retries=self.endpoint_data.max_retries,
-            max_concurrent=self.endpoint_data.max_concurrent,
-            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,
@@ -256,19 +351,23 @@ class AdminGetProviderEndpointAdapter(AdminApiAdapter):
             raise NotFoundException(f"Endpoint {self.endpoint_id} 不存在")
 
         endpoint_obj, provider = endpoint
-        total_keys = (
-            db.query(ProviderAPIKey).filter(ProviderAPIKey.endpoint_id == self.endpoint_id).count()
+        endpoint_format = (
+            endpoint_obj.api_format
+            if isinstance(endpoint_obj.api_format, str)
+            else endpoint_obj.api_format.value
         )
-        active_keys = (
-            db.query(ProviderAPIKey)
-            .filter(
-                and_(
-                    ProviderAPIKey.endpoint_id == self.endpoint_id,
-                    ProviderAPIKey.is_active.is_(True),
-                )
-            )
-            .count()
+        keys = (
+            db.query(ProviderAPIKey.api_formats, ProviderAPIKey.is_active)
+            .filter(ProviderAPIKey.provider_id == endpoint_obj.provider_id)
+            .all()
         )
+        total_keys = 0
+        active_keys = 0
+        for api_formats, is_active in keys:
+            if endpoint_format in (api_formats or []):
+                total_keys += 1
+                if is_active:
+                    active_keys += 1
 
         endpoint_dict = {
             k: v
@@ -320,19 +419,21 @@ class AdminUpdateProviderEndpointAdapter(AdminApiAdapter):
         provider = db.query(Provider).filter(Provider.id == endpoint.provider_id).first()
         logger.info(f"[OK] 更新 Endpoint: ID={self.endpoint_id}, Updates={list(update_data.keys())}")
 
-        total_keys = (
-            db.query(ProviderAPIKey).filter(ProviderAPIKey.endpoint_id == self.endpoint_id).count()
+        endpoint_format = (
+            endpoint.api_format if isinstance(endpoint.api_format, str) else endpoint.api_format.value
         )
-        active_keys = (
-            db.query(ProviderAPIKey)
-            .filter(
-                and_(
-                    ProviderAPIKey.endpoint_id == self.endpoint_id,
-                    ProviderAPIKey.is_active.is_(True),
-                )
-            )
-            .count()
+        keys = (
+            db.query(ProviderAPIKey.api_formats, ProviderAPIKey.is_active)
+            .filter(ProviderAPIKey.provider_id == endpoint.provider_id)
+            .all()
         )
+        total_keys = 0
+        active_keys = 0
+        for api_formats, is_active in keys:
+            if endpoint_format in (api_formats or []):
+                total_keys += 1
+                if is_active:
+                    active_keys += 1
 
         endpoint_dict = {
             k: v
@@ -361,12 +462,26 @@ class AdminDeleteProviderEndpointAdapter(AdminApiAdapter):
         if not endpoint:
             raise NotFoundException(f"Endpoint {self.endpoint_id} 不存在")
 
-        keys_count = (
-            db.query(ProviderAPIKey).filter(ProviderAPIKey.endpoint_id == self.endpoint_id).count()
+        endpoint_format = (
+            endpoint.api_format if isinstance(endpoint.api_format, str) else endpoint.api_format.value
+        )
+        keys = (
+            db.query(ProviderAPIKey.api_formats)
+            .filter(ProviderAPIKey.provider_id == endpoint.provider_id)
+            .all()
+        )
+        affected_keys_count = sum(
+            1 for (api_formats,) in keys if endpoint_format in (api_formats or [])
         )
         db.delete(endpoint)
         db.commit()
 
-        logger.warning(f"[DELETE] 删除 Endpoint: ID={self.endpoint_id}, 同时删除了 {keys_count} 个 Keys")
+        logger.warning(
+            f"[DELETE] 删除 Endpoint: ID={self.endpoint_id}, Format={endpoint_format}, "
+            f"AffectedKeys={affected_keys_count}"
+        )
 
-        return {"message": f"Endpoint {self.endpoint_id} 已删除", "deleted_keys_count": keys_count}
+        return {
+            "message": f"Endpoint {self.endpoint_id} 已删除",
+            "affected_keys_count": affected_keys_count,
+        }

src/api/admin/ldap.py

@@ -166,21 +166,95 @@ class LDAPConfigTest(BaseModel):
 
 @router.get("/config")
 async def get_ldap_config(request: Request, db: Session = Depends(get_db)) -> Any:
-    """获取LDAP配置（管理员）"""
+    """
+    获取 LDAP 配置
+
+    获取系统当前的 LDAP 认证配置信息，用于管理界面显示和编辑。
+    密码字段不会返回原文，仅返回是否已设置的标志。
+
+    **返回字段**:
+    - `server_url`: LDAP 服务器地址（如：ldap://ldap.example.com:389）
+    - `bind_dn`: 绑定 DN（如：cn=admin,dc=example,dc=com）
+    - `base_dn`: 搜索基准 DN（如：ou=users,dc=example,dc=com）
+    - `has_bind_password`: 是否已设置绑定密码（布尔值）
+    - `user_search_filter`: 用户搜索过滤器（默认：(uid={username})）
+    - `username_attr`: 用户名属性（默认：uid）
+    - `email_attr`: 邮箱属性（默认：mail）
+    - `display_name_attr`: 显示名称属性（默认：cn）
+    - `is_enabled`: 是否启用 LDAP 认证
+    - `is_exclusive`: 是否仅允许 LDAP 登录（独占模式）
+    - `use_starttls`: 是否使用 STARTTLS 加密连接
+    - `connect_timeout`: 连接超时时间（秒，1-60）
+    """
     adapter = AdminGetLDAPConfigAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
 @router.put("/config")
 async def update_ldap_config(request: Request, db: Session = Depends(get_db)) -> Any:
-    """更新LDAP配置（管理员）"""
+    """
+    更新 LDAP 配置
+
+    更新系统的 LDAP 认证配置。支持完整配置更新，包括连接参数、
+    搜索过滤器、属性映射等。提供多重安全校验，防止误锁定管理员。
+
+    **请求体字段**:
+    - `server_url`: LDAP 服务器地址（必填，1-255字符）
+    - `bind_dn`: 绑定 DN（必填，1-255字符）
+    - `bind_password`: 绑定密码（可选，设为空字符串可清除密码）
+    - `base_dn`: 搜索基准 DN（必填，1-255字符）
+    - `user_search_filter`: 用户搜索过滤器（必须包含 {username} 占位符，默认：(uid={username})）
+    - `username_attr`: 用户名属性（默认：uid）
+    - `email_attr`: 邮箱属性（默认：mail）
+    - `display_name_attr`: 显示名称属性（默认：cn）
+    - `is_enabled`: 是否启用 LDAP 认证
+    - `is_exclusive`: 是否仅允许 LDAP 登录（需先启用 LDAP）
+    - `use_starttls`: 是否使用 STARTTLS 加密连接
+    - `connect_timeout`: 连接超时时间（秒，1-60，默认 10）
+
+    **安全校验**:
+    - 启用 LDAP 时必须设置有效的绑定密码
+    - 启用独占模式前会检查是否有至少 1 个有效的本地管理员账户
+    - 独占模式要求先启用 LDAP 认证
+    - 搜索过滤器必须包含 {username} 占位符且括号匹配
+    - 搜索过滤器嵌套层数不超过 5 层，长度不超过 200 字符
+
+    **返回字段**:
+    - `message`: 操作结果消息
+    """
     adapter = AdminUpdateLDAPConfigAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
 @router.post("/test")
 async def test_ldap_connection(request: Request, db: Session = Depends(get_db)) -> Any:
-    """测试LDAP连接（管理员）"""
+    """
+    测试 LDAP 连接
+
+    在保存配置前测试 LDAP 服务器连接是否正常。支持使用已保存的配置，
+    也支持通过请求体覆盖任意配置项进行临时测试，而不影响已保存的配置。
+
+    **请求体字段**（均为可选，用于临时覆盖）:
+    - `server_url`: LDAP 服务器地址（覆盖已保存的配置）
+    - `bind_dn`: 绑定 DN（覆盖已保存的配置）
+    - `bind_password`: 绑定密码（覆盖已保存的密码）
+    - `base_dn`: 搜索基准 DN（覆盖已保存的配置）
+    - `user_search_filter`: 用户搜索过滤器（覆盖已保存的配置）
+    - `username_attr`: 用户名属性（覆盖已保存的配置）
+    - `email_attr`: 邮箱属性（覆盖已保存的配置）
+    - `display_name_attr`: 显示名称属性（覆盖已保存的配置）
+    - `use_starttls`: 是否使用 STARTTLS（覆盖已保存的配置）
+    - `connect_timeout`: 连接超时时间（覆盖已保存的配置）
+
+    **测试逻辑**:
+    - 未提供的字段使用已保存的配置值
+    - `bind_password` 优先使用请求体中的值，否则使用已保存的加密密码
+    - 测试时会尝试连接 LDAP 服务器并验证绑定 DN
+
+    **返回字段**:
+    - `success`: 测试是否成功（布尔值）
+    - `message`: 测试结果消息（成功或失败原因）
+    """
     adapter = AdminTestLDAPConnectionAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 

src/api/admin/management_tokens/__init__.py

@@ -0,0 +1,10 @@
+"""Management Token 管理员路由模块"""
+
+from fastapi import APIRouter
+
+from .routes import router as management_tokens_router
+
+router = APIRouter()
+router.include_router(management_tokens_router)
+
+__all__ = ["router"]

src/api/admin/management_tokens/routes.py

@@ -0,0 +1,300 @@
+"""管理员 Management Token 管理端点"""
+
+from dataclasses import dataclass
+from typing import Optional
+
+from fastapi import APIRouter, Depends, HTTPException, Query, Request
+from fastapi.responses import JSONResponse
+from sqlalchemy.orm import Session
+
+from src.api.base.admin_adapter import AdminApiAdapter
+from src.api.base.context import ApiRequestContext
+from src.api.base.pipeline import ApiRequestPipeline
+from src.core.exceptions import NotFoundException
+from src.database import get_db
+from src.models.database import AuditEventType, ManagementToken, User
+from src.services.management_token import ManagementTokenService, token_to_dict
+
+router = APIRouter(prefix="/api/admin/management-tokens", tags=["Admin - Management Tokens"])
+pipeline = ApiRequestPipeline()
+
+
+# ============== 安全基类 ==============
+
+
+class AdminManagementTokenApiAdapter(AdminApiAdapter):
+    """管理员 Management Token 管理 API 的基类
+
+    安全限制：禁止使用 Management Token 调用这些接口。
+    """
+
+    def authorize(self, context: ApiRequestContext) -> None:
+        # 先调用父类的认证和权限检查
+        super().authorize(context)
+
+        # 禁止使用 Management Token 调用 management-tokens 相关接口
+        if context.management_token is not None:
+            raise HTTPException(
+                status_code=403,
+                detail="不允许使用 Management Token 管理其他 Token，请使用 Web 界面或 JWT 认证",
+            )
+
+
+# ============== 路由 ==============
+
+
+@router.get("")
+async def list_all_management_tokens(
+    request: Request,
+    user_id: Optional[str] = Query(None, description="筛选用户 ID"),
+    is_active: Optional[bool] = Query(None, description="筛选激活状态"),
+    skip: int = Query(0, ge=0),
+    limit: int = Query(50, ge=1, le=100),
+    db: Session = Depends(get_db),
+):
+    """列出所有 Management Tokens（管理员）
+
+    管理员查看所有用户的 Management Tokens，支持筛选和分页。
+
+    **查询参数**
+    - user_id (Optional[str]): 筛选指定用户 ID 的 tokens
+    - is_active (Optional[bool]): 筛选激活状态（true/false）
+    - skip (int): 分页偏移量，默认 0
+    - limit (int): 每页数量，范围 1-100，默认 50
+
+    **返回字段**
+    - items (List[dict]): Token 列表
+        - id (str): Token ID
+        - user_id (str): 所属用户 ID
+        - user (dict): 用户信息（包含 id, username, email 等）
+        - name (str): Token 名称
+        - description (Optional[str]): 描述
+        - token_hash (str): Token 哈希值（不返回明文）
+        - is_active (bool): 是否激活
+        - allowed_ips (Optional[List[str]]): IP 白名单
+        - expires_at (Optional[str]): 过期时间（ISO 8601 格式）
+        - last_used_at (Optional[str]): 最后使用时间
+        - created_at (str): 创建时间
+        - updated_at (str): 更新时间
+    - total (int): 总数量
+    - skip (int): 当前偏移量
+    - limit (int): 当前每页数量
+    """
+    adapter = AdminListManagementTokensAdapter(
+        user_id=user_id, is_active=is_active, skip=skip, limit=limit
+    )
+    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
+
+
+@router.get("/{token_id}")
+async def get_management_token(
+    token_id: str,
+    request: Request,
+    db: Session = Depends(get_db),
+):
+    """获取 Management Token 详情（管理员）
+
+    管理员查看任意 Management Token 的详细信息。
+
+    **路径参数**
+    - token_id (str): Token ID
+
+    **返回字段**
+    - id (str): Token ID
+    - user_id (str): 所属用户 ID
+    - user (dict): 用户信息（包含 id, username, email 等）
+    - name (str): Token 名称
+    - description (Optional[str]): 描述
+    - token_hash (str): Token 哈希值（不返回明文）
+    - is_active (bool): 是否激活
+    - allowed_ips (Optional[List[str]]): IP 白名单
+    - expires_at (Optional[str]): 过期时间（ISO 8601 格式）
+    - last_used_at (Optional[str]): 最后使用时间
+    - created_at (str): 创建时间
+    - updated_at (str): 更新时间
+    """
+    adapter = AdminGetManagementTokenAdapter(token_id=token_id)
+    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
+
+
+@router.delete("/{token_id}")
+async def delete_management_token(
+    token_id: str, request: Request, db: Session = Depends(get_db)
+):
+    """删除任意 Management Token（管理员）
+
+    管理员可以删除任意用户的 Management Token。
+
+    **路径参数**
+    - token_id (str): 要删除的 Token ID
+
+    **返回字段**
+    - message (str): 操作结果消息
+    """
+    adapter = AdminDeleteManagementTokenAdapter(token_id=token_id)
+    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
+
+
+@router.patch("/{token_id}/status")
+async def toggle_management_token(
+    token_id: str, request: Request, db: Session = Depends(get_db)
+):
+    """切换任意 Management Token 状态（管理员）
+
+    管理员可以启用/禁用任意用户的 Management Token。
+
+    **路径参数**
+    - token_id (str): Token ID
+
+    **返回字段**
+    - message (str): 操作结果消息（"Token 已启用" 或 "Token 已禁用"）
+    - data (dict): 更新后的 Token 信息
+        - id (str): Token ID
+        - user_id (str): 所属用户 ID
+        - user (dict): 用户信息
+        - name (str): Token 名称
+        - description (Optional[str]): 描述
+        - token_hash (str): Token 哈希值
+        - is_active (bool): 是否激活（已切换后的状态）
+        - allowed_ips (Optional[List[str]]): IP 白名单
+        - expires_at (Optional[str]): 过期时间
+        - last_used_at (Optional[str]): 最后使用时间
+        - created_at (str): 创建时间
+        - updated_at (str): 更新时间
+    """
+    adapter = AdminToggleManagementTokenAdapter(token_id=token_id)
+    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
+
+
+# ============== 适配器 ==============
+
+
+@dataclass
+class AdminListManagementTokensAdapter(AdminManagementTokenApiAdapter):
+    """列出所有 Management Tokens"""
+
+    name: str = "admin_list_management_tokens"
+    user_id: Optional[str] = None
+    is_active: Optional[bool] = None
+    skip: int = 0
+    limit: int = 50
+
+    async def handle(self, context: ApiRequestContext):
+        # 构建查询
+        query = context.db.query(ManagementToken)
+
+        if self.user_id:
+            query = query.filter(ManagementToken.user_id == self.user_id)
+        if self.is_active is not None:
+            query = query.filter(ManagementToken.is_active == self.is_active)
+
+        total = query.count()
+        tokens = (
+            query.order_by(ManagementToken.created_at.desc())
+            .offset(self.skip)
+            .limit(self.limit)
+            .all()
+        )
+
+        # 预加载用户信息
+        user_ids = list(set(t.user_id for t in tokens))
+        users = {u.id: u for u in context.db.query(User).filter(User.id.in_(user_ids)).all()}
+        for token in tokens:
+            token.user = users.get(token.user_id)
+
+        return JSONResponse(
+            content={
+                "items": [token_to_dict(t, include_user=True) for t in tokens],
+                "total": total,
+                "skip": self.skip,
+                "limit": self.limit,
+            }
+        )
+
+
+@dataclass
+class AdminGetManagementTokenAdapter(AdminManagementTokenApiAdapter):
+    """获取 Management Token 详情"""
+
+    name: str = "admin_get_management_token"
+    token_id: str = ""
+
+    async def handle(self, context: ApiRequestContext):
+        token = ManagementTokenService.get_token_by_id(
+            db=context.db, token_id=self.token_id
+        )
+
+        if not token:
+            raise NotFoundException("Management Token 不存在")
+
+        # 加载用户信息
+        token.user = context.db.query(User).filter(User.id == token.user_id).first()
+
+        return JSONResponse(content=token_to_dict(token, include_user=True))
+
+
+@dataclass
+class AdminDeleteManagementTokenAdapter(AdminManagementTokenApiAdapter):
+    """删除 Management Token"""
+
+    name: str = "admin_delete_management_token"
+    token_id: str = ""
+    audit_success_event = AuditEventType.MANAGEMENT_TOKEN_DELETED
+
+    async def handle(self, context: ApiRequestContext):
+        # 先获取 token 信息用于审计
+        token = ManagementTokenService.get_token_by_id(
+            db=context.db, token_id=self.token_id
+        )
+
+        if not token:
+            raise NotFoundException("Management Token 不存在")
+
+        context.add_audit_metadata(
+            token_id=token.id,
+            token_name=token.name,
+            owner_user_id=token.user_id,
+        )
+
+        success = ManagementTokenService.delete_token(
+            db=context.db, token_id=self.token_id
+        )
+
+        if not success:
+            raise NotFoundException("Management Token 不存在")
+
+        return JSONResponse(content={"message": "删除成功"})
+
+
+@dataclass
+class AdminToggleManagementTokenAdapter(AdminManagementTokenApiAdapter):
+    """切换 Management Token 状态"""
+
+    name: str = "admin_toggle_management_token"
+    token_id: str = ""
+    audit_success_event = AuditEventType.MANAGEMENT_TOKEN_UPDATED
+
+    async def handle(self, context: ApiRequestContext):
+        token = ManagementTokenService.toggle_status(
+            db=context.db, token_id=self.token_id
+        )
+
+        if not token:
+            raise NotFoundException("Management Token 不存在")
+
+        # 加载用户信息
+        token.user = context.db.query(User).filter(User.id == token.user_id).first()
+
+        context.add_audit_metadata(
+            token_id=token.id,
+            token_name=token.name,
+            owner_user_id=token.user_id,
+            is_active=token.is_active,
+        )
+
+        return JSONResponse(
+            content={
+                "message": f"Token 已{'启用' if token.is_active else '禁用'}",
+                "data": token_to_dict(token, include_user=True),
+            }
+        )

src/api/admin/models/__init__.py

@@ -8,7 +8,7 @@ from .catalog import router as catalog_router
 from .external import router as external_router
 from .global_models import router as global_models_router
 
-router = APIRouter(prefix="/api/admin/models", tags=["Admin - Model Management"])
+router = APIRouter(prefix="/api/admin/models", tags=["Admin - Models"])
 
 # 挂载子路由
 router.include_router(catalog_router)

src/api/admin/models/catalog.py

@@ -31,6 +31,22 @@ async def get_model_catalog(
     request: Request,
     db: Session = Depends(get_db),
 ) -> ModelCatalogResponse:
+    """
+    获取统一模型目录
+
+    基于 GlobalModel 聚合所有活跃模型及其关联提供商的信息，返回完整的模型目录视图。
+
+    **返回字段**:
+    - `models`: 模型列表，每个模型包含：
+      - `global_model_name`: GlobalModel 名称
+      - `display_name`: 显示名称
+      - `description`: 模型描述
+      - `providers`: 提供商列表，包含提供商名称、价格、能力等详细信息
+      - `price_range`: 价格区间（基于 GlobalModel 第一阶梯价格）
+      - `total_providers`: 关联提供商数量
+      - `capabilities`: 模型能力标志（视觉、函数调用、流式输出）
+    - `total`: 模型总数
+    """
     adapter = AdminGetModelCatalogAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -109,7 +125,6 @@ class AdminGetModelCatalogAdapter(AdminApiAdapter):
                     ModelCatalogProviderDetail(
                         provider_id=provider.id,
                         provider_name=provider.name,
-                        provider_display_name=provider.display_name,
                         model_id=model.id,
                         target_model=model.provider_model_name,
                         # 显示有效价格

src/api/admin/models/external.py

@@ -82,9 +82,21 @@ def _mark_official_providers(data: dict[str, Any]) -> dict[str, Any]:
 @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
+    获取外部模型数据
+
+    从 models.dev 获取第三方模型数据，用于导入新模型或参考定价信息。
+    该接口作为代理请求解决跨域问题，并提供缓存优化。
+
+    **功能特性**:
+    - 代理 models.dev API，解决前端跨域问题
+    - 使用 Redis 缓存 15 分钟，多 worker 共享缓存
+    - 自动标记官方提供商（official 字段），前端可据此过滤第三方转售商
+
+    **返回字段**:
+    - 键为提供商 ID（如 "anthropic"、"openai"）
+    - 值为提供商详细信息，包含：
+      - `official`: 是否为官方提供商（true/false）
+      - 其他 models.dev 提供的原始字段（模型列表、定价等）
     """
     # 检查缓存
     cached = await _get_cached_data()
@@ -130,7 +142,16 @@ async def get_external_models(_: User = Depends(require_admin)) -> JSONResponse:
 
 @router.delete("/external/cache")
 async def clear_external_models_cache(_: User = Depends(require_admin)) -> dict:
-    """清除 models.dev 缓存"""
+    """
+    清除外部模型数据缓存
+
+    手动清除 models.dev 的 Redis 缓存，强制下次请求重新获取最新数据。
+    通常用于需要立即更新外部模型数据的场景。
+
+    **返回字段**:
+    - `cleared`: 是否成功清除缓存（true/false）
+    - `message`: 提示信息（仅在 Redis 未启用时返回）
+    """
     redis = await get_redis_client()
     if redis is None:
         return {"cleared": False, "message": "Redis 未启用"}

src/api/admin/models/global_models.py

@@ -40,7 +40,27 @@ async def list_global_models(
     search: Optional[str] = Query(None),
     db: Session = Depends(get_db),
 ) -> GlobalModelListResponse:
-    """获取 GlobalModel 列表"""
+    """
+    获取 GlobalModel 列表
+
+    查询系统中的全局模型列表，支持分页、过滤和搜索功能。
+
+    **查询参数**:
+    - `skip`: 跳过记录数，用于分页（默认 0）
+    - `limit`: 返回记录数，用于分页（默认 100，最大 1000）
+    - `is_active`: 过滤活跃状态（true/false/null，null 表示不过滤）
+    - `search`: 搜索关键词，支持按名称或显示名称模糊搜索
+
+    **返回字段**:
+    - `models`: GlobalModel 列表，每个包含：
+      - `id`: GlobalModel ID
+      - `name`: 模型名称（唯一）
+      - `display_name`: 显示名称
+      - `is_active`: 是否活跃
+      - `provider_count`: 关联提供商数量
+      - 定价和能力配置等其他字段
+    - `total`: 返回的模型总数
+    """
     adapter = AdminListGlobalModelsAdapter(
         skip=skip,
         limit=limit,
@@ -56,7 +76,21 @@ async def get_global_model(
     global_model_id: str,
     db: Session = Depends(get_db),
 ) -> GlobalModelWithStats:
-    """获取单个 GlobalModel 详情（含统计信息）"""
+    """
+    获取单个 GlobalModel 详情
+
+    查询指定 GlobalModel 的详细信息，包含关联的提供商和价格统计数据。
+
+    **路径参数**:
+    - `global_model_id`: GlobalModel ID
+
+    **返回字段**:
+    - 基础字段：`id`, `name`, `display_name`, `is_active` 等
+    - 统计字段：
+      - `total_models`: 关联的 Model 实现数量
+      - `total_providers`: 关联的提供商数量
+      - `price_range`: 价格区间统计（最低/最高输入输出价格）
+    """
     adapter = AdminGetGlobalModelAdapter(global_model_id=global_model_id)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -67,7 +101,24 @@ async def create_global_model(
     payload: GlobalModelCreate,
     db: Session = Depends(get_db),
 ) -> GlobalModelResponse:
-    """创建 GlobalModel"""
+    """
+    创建 GlobalModel
+
+    创建一个新的全局模型定义，作为多个提供商实现的统一抽象。
+
+    **请求体字段**:
+    - `name`: 模型名称（唯一标识，如 "claude-3-5-sonnet-20241022"）
+    - `display_name`: 显示名称（如 "Claude 3.5 Sonnet"）
+    - `is_active`: 是否活跃（默认 true）
+    - `default_price_per_request`: 默认按次计费价格（可选）
+    - `default_tiered_pricing`: 默认阶梯定价配置（包含多个价格阶梯）
+    - `supported_capabilities`: 支持的能力标志（vision、function_calling、streaming）
+    - `config`: 额外配置（JSON 格式，如 description、context_window 等）
+
+    **返回字段**:
+    - `id`: 创建的 GlobalModel ID
+    - 其他请求体中的所有字段
+    """
     adapter = AdminCreateGlobalModelAdapter(payload=payload)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -79,7 +130,26 @@ async def update_global_model(
     payload: GlobalModelUpdate,
     db: Session = Depends(get_db),
 ) -> GlobalModelResponse:
-    """更新 GlobalModel"""
+    """
+    更新 GlobalModel
+
+    更新指定 GlobalModel 的配置信息，支持部分字段更新。
+    更新后会自动失效相关缓存。
+
+    **路径参数**:
+    - `global_model_id`: GlobalModel ID
+
+    **请求体字段**（均为可选）:
+    - `display_name`: 显示名称
+    - `is_active`: 是否活跃
+    - `default_price_per_request`: 默认按次计费价格
+    - `default_tiered_pricing`: 默认阶梯定价配置
+    - `supported_capabilities`: 支持的能力标志
+    - `config`: 额外配置
+
+    **返回字段**:
+    - 更新后的完整 GlobalModel 信息
+    """
     adapter = AdminUpdateGlobalModelAdapter(global_model_id=global_model_id, payload=payload)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -90,7 +160,18 @@ async def delete_global_model(
     global_model_id: str,
     db: Session = Depends(get_db),
 ):
-    """删除 GlobalModel（级联删除所有关联的 Provider 模型实现）"""
+    """
+    删除 GlobalModel
+
+    删除指定的 GlobalModel，会级联删除所有关联的 Provider 模型实现。
+    删除后会自动失效相关缓存。
+
+    **路径参数**:
+    - `global_model_id`: GlobalModel ID
+
+    **返回**:
+    - 成功删除返回 204 状态码，无响应体
+    """
     adapter = AdminDeleteGlobalModelAdapter(global_model_id=global_model_id)
     await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
     return None
@@ -105,7 +186,29 @@ async def batch_assign_to_providers(
     payload: BatchAssignToProvidersRequest,
     db: Session = Depends(get_db),
 ) -> BatchAssignToProvidersResponse:
-    """批量为多个 Provider 添加 GlobalModel 实现"""
+    """
+    批量为提供商添加模型实现
+
+    为指定的 GlobalModel 批量创建多个 Provider 的模型实现（Model 记录）。
+    用于快速将一个统一模型分配给多个提供商。
+
+    **路径参数**:
+    - `global_model_id`: GlobalModel ID
+
+    **请求体字段**:
+    - `provider_ids`: 提供商 ID 列表
+    - `create_models`: Model 创建配置列表，每个包含：
+      - `provider_id`: 提供商 ID
+      - `provider_model_name`: 提供商侧的模型名称（如 "claude-3-5-sonnet-20241022"）
+      - 其他可选字段（价格覆盖、能力覆盖等）
+
+    **返回字段**:
+    - `success`: 成功创建的 Model 列表
+    - `errors`: 失败的提供商及错误信息列表
+    - `total_requested`: 请求处理的总数
+    - `total_success`: 成功创建的数量
+    - `total_errors`: 失败的数量
+    """
     adapter = AdminBatchAssignToProvidersAdapter(global_model_id=global_model_id, payload=payload)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -116,7 +219,27 @@ async def get_global_model_providers(
     global_model_id: str,
     db: Session = Depends(get_db),
 ) -> GlobalModelProvidersResponse:
-    """获取 GlobalModel 的所有关联提供商（包括非活跃的）"""
+    """
+    获取 GlobalModel 的关联提供商
+
+    查询指定 GlobalModel 的所有关联提供商及其模型实现详情，包括非活跃的提供商。
+    用于查看某个统一模型在各个提供商上的具体配置。
+
+    **路径参数**:
+    - `global_model_id`: GlobalModel ID
+
+    **返回字段**:
+    - `providers`: 提供商列表，每个包含：
+      - `provider_id`: 提供商 ID
+      - `provider_name`: 提供商名称
+      - `provider_display_name`: 提供商显示名称
+      - `model_id`: Model 实现 ID
+      - `target_model`: 提供商侧的模型名称
+      - 价格信息（input_price_per_1m、output_price_per_1m 等）
+      - 能力标志（supports_vision、supports_function_calling、supports_streaming）
+      - `is_active`: 是否活跃
+    - `total`: 关联提供商总数
+    """
     adapter = AdminGetGlobalModelProvidersAdapter(global_model_id=global_model_id)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -329,7 +452,6 @@ class AdminGetGlobalModelProvidersAdapter(AdminApiAdapter):
                 ModelCatalogProviderDetail(
                     provider_id=provider.id,
                     provider_name=provider.name,
-                    provider_display_name=provider.display_name,
                     model_id=model.id,
                     target_model=model.provider_model_name,
                     input_price_per_1m=model.get_effective_input_price(),

src/api/admin/monitoring/audit.py

@@ -39,6 +39,34 @@ async def get_audit_logs(
     offset: int = Query(0, description="偏移量"),
     db: Session = Depends(get_db),
 ):
+    """
+    获取审计日志
+
+    获取系统审计日志列表，支持按用户、事件类型、时间范围筛选。需要管理员权限。
+
+    **查询参数**:
+    - `user_id`: 可选，用户 ID 筛选（UUID 格式）
+    - `event_type`: 可选，事件类型筛选
+    - `days`: 查询最近多少天的日志，默认 7 天
+    - `limit`: 返回数量限制，默认 100
+    - `offset`: 分页偏移量，默认 0
+
+    **返回字段**:
+    - `items`: 审计日志列表，每条日志包含：
+      - `id`: 日志 ID
+      - `event_type`: 事件类型
+      - `user_id`: 用户 ID
+      - `user_email`: 用户邮箱
+      - `user_username`: 用户名
+      - `description`: 事件描述
+      - `ip_address`: IP 地址
+      - `status_code`: HTTP 状态码
+      - `error_message`: 错误信息
+      - `metadata`: 事件元数据
+      - `created_at`: 创建时间
+    - `meta`: 分页元数据（total, limit, offset, count）
+    - `filters`: 筛选条件
+    """
     adapter = AdminGetAuditLogsAdapter(
         user_id=user_id,
         event_type=event_type,
@@ -51,6 +79,19 @@ async def get_audit_logs(
 
 @router.get("/system-status")
 async def get_system_status(request: Request, db: Session = Depends(get_db)):
+    """
+    获取系统状态
+
+    获取系统当前的运行状态和关键指标。需要管理员权限。
+
+    **返回字段**:
+    - `timestamp`: 当前时间戳
+    - `users`: 用户统计（total: 总用户数, active: 活跃用户数）
+    - `providers`: 提供商统计（total: 总提供商数, active: 活跃提供商数）
+    - `api_keys`: API Key 统计（total: 总数, active: 活跃数）
+    - `today_stats`: 今日统计（requests: 请求数, tokens: token 数, cost_usd: 成本）
+    - `recent_errors`: 最近 1 小时内的错误数
+    """
     adapter = AdminSystemStatusAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -61,6 +102,26 @@ async def get_suspicious_activities(
     hours: int = Query(24, description="时间范围（小时）"),
     db: Session = Depends(get_db),
 ):
+    """
+    获取可疑活动记录
+
+    获取系统检测到的可疑活动记录。需要管理员权限。
+
+    **查询参数**:
+    - `hours`: 时间范围（小时），默认 24 小时
+
+    **返回字段**:
+    - `activities`: 可疑活动列表，每条记录包含：
+      - `id`: 记录 ID
+      - `event_type`: 事件类型
+      - `user_id`: 用户 ID
+      - `description`: 事件描述
+      - `ip_address`: IP 地址
+      - `metadata`: 事件元数据
+      - `created_at`: 创建时间
+    - `count`: 活动总数
+    - `time_range_hours`: 查询的时间范围（小时）
+    """
     adapter = AdminSuspiciousActivitiesAdapter(hours=hours)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -72,19 +133,56 @@ async def analyze_user_behavior(
     days: int = Query(30, description="分析天数"),
     db: Session = Depends(get_db),
 ):
+    """
+    分析用户行为
+
+    分析指定用户的行为模式和使用情况。需要管理员权限。
+
+    **路径参数**:
+    - `user_id`: 用户 ID
+
+    **查询参数**:
+    - `days`: 分析最近多少天的数据，默认 30 天
+
+    **返回字段**:
+    - 用户行为分析结果，包括活动频率、使用模式、异常行为等
+    """
     adapter = AdminUserBehaviorAdapter(user_id=user_id, days=days)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
 @router.get("/resilience-status")
 async def get_resilience_status(request: Request, db: Session = Depends(get_db)):
+    """
+    获取韧性系统状态
+
+    获取系统韧性管理的当前状态，包括错误统计、熔断器状态等。需要管理员权限。
+
+    **返回字段**:
+    - `timestamp`: 当前时间戳
+    - `health_score`: 健康评分（0-100）
+    - `status`: 系统状态（healthy: 健康，degraded: 降级，critical: 严重）
+    - `error_statistics`: 错误统计信息
+    - `recent_errors`: 最近的错误列表（最多 10 条）
+    - `recommendations`: 系统建议
+    """
     adapter = AdminResilienceStatusAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
 @router.delete("/resilience/error-stats")
 async def reset_error_stats(request: Request, db: Session = Depends(get_db)):
-    """Reset resilience error statistics"""
+    """
+    重置错误统计
+
+    重置韧性系统的错误统计数据。需要管理员权限。
+
+    **返回字段**:
+    - `message`: 操作结果信息
+    - `previous_stats`: 重置前的统计数据
+    - `reset_by`: 执行重置的管理员邮箱
+    - `reset_at`: 重置时间
+    """
     adapter = AdminResetErrorStatsAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -95,6 +193,18 @@ async def get_circuit_history(
     limit: int = Query(50, ge=1, le=200),
     db: Session = Depends(get_db),
 ):
+    """
+    获取熔断器历史记录
+
+    获取熔断器的状态变更历史记录。需要管理员权限。
+
+    **查询参数**:
+    - `limit`: 返回数量限制，默认 50，最大 200
+
+    **返回字段**:
+    - `items`: 熔断器历史记录列表
+    - `count`: 记录总数
+    """
     adapter = AdminCircuitHistoryAdapter(limit=limit)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 

src/api/admin/monitoring/cache.py

@@ -117,12 +117,21 @@ async def get_cache_stats(
     """
     获取缓存亲和性统计信息
 
-    返回:
-    - 缓存命中率
-    - 缓存用户数
-    - Provider切换次数
-    - Key切换次数
-    - 缓存预留配置
+    获取缓存调度器的运行统计数据，包括命中率、切换次数、调度器配置等。
+    用于监控缓存亲和性功能的运行状态和性能指标。
+
+    **返回字段**:
+    - `status`: 状态（ok）
+    - `data`: 统计数据对象
+      - `scheduler`: 调度器名称（cache_aware 或 random）
+      - `total_affinities`: 总缓存亲和性数量
+      - `cache_hit_rate`: 缓存命中率（0.0-1.0）
+      - `provider_switches`: Provider 切换次数
+      - `key_switches`: Key 切换次数
+      - `cache_hits`: 缓存命中次数
+      - `cache_misses`: 缓存未命中次数
+      - `scheduler_metrics`: 调度器详细指标
+      - `affinity_stats`: 亲和性统计数据
     """
     adapter = AdminCacheStatsAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
@@ -137,16 +146,33 @@ async def get_user_affinity(
     """
     查询指定用户的所有缓存亲和性
 
-    参数:
-    - user_identifier: 用户标识符，支持以下格式：
-      * 用户名 (username)，如: yuanhonghu
-      * 邮箱 (email)，如: user@example.com
-      * 用户UUID (user_id)，如: 550e8400-e29b-41d4-a716-446655440000
-      * API Key ID，如: 660e8400-e29b-41d4-a716-446655440000
+    根据用户标识符查询该用户在各个端点上的缓存亲和性记录。
+    支持多种标识符格式的自动识别和解析。
 
-    返回:
-    - 用户信息
-    - 所有端点的缓存亲和性列表（每个端点一条记录）
+    **路径参数**:
+    - `user_identifier`: 用户标识符，支持以下格式：
+      - 用户名（username），如：yuanhonghu
+      - 邮箱（email），如：user@example.com
+      - 用户 UUID（user_id），如：550e8400-e29b-41d4-a716-446655440000
+      - API Key ID，如：660e8400-e29b-41d4-a716-446655440000
+
+    **返回字段**:
+    - `status`: 状态（ok 或 not_found）
+    - `message`: 提示消息（当无缓存时）
+    - `user_info`: 用户信息
+      - `user_id`: 用户 ID
+      - `username`: 用户名
+      - `email`: 邮箱
+    - `affinities`: 缓存亲和性列表
+      - `provider_id`: Provider ID
+      - `endpoint_id`: Endpoint ID
+      - `key_id`: Key ID
+      - `api_format`: API 格式
+      - `model_name`: 模型名称（global_model_id）
+      - `created_at`: 创建时间
+      - `expire_at`: 过期时间
+      - `request_count`: 请求计数
+    - `total_endpoints`: 缓存的端点数量
     """
     adapter = AdminGetUserAffinityAdapter(user_identifier=user_identifier)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
@@ -161,10 +187,50 @@ async def list_affinities(
     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
+    获取所有缓存亲和性列表
 
-    参数:
-    - keyword: 可选，支持用户名/邮箱/User ID/API Key ID 或模糊匹配
+    查询系统中所有的缓存亲和性记录，支持按关键词过滤和分页。
+    返回详细的用户、Provider、Endpoint、Key 信息。
+
+    **查询参数**:
+    - `keyword`: 可选，支持以下过滤方式（可选）
+      - 用户名/邮箱/User ID/API Key ID（精确匹配）
+      - 任意字段的模糊匹配（affinity_key、user_id、username、email、provider_id、key_id）
+    - `limit`: 返回数量限制（1-1000，默认 100）
+    - `offset`: 偏移量（用于分页，默认 0）
+
+    **返回字段**:
+    - `status`: 状态（ok）
+    - `data`: 分页数据对象
+      - `items`: 缓存亲和性列表
+        - `affinity_key`: API Key ID（用于缓存键）
+        - `user_api_key_name`: 用户 API Key 名称
+        - `user_api_key_prefix`: 脱敏后的用户 API Key
+        - `is_standalone`: 是否为独立 API Key
+        - `user_id`: 用户 ID
+        - `username`: 用户名
+        - `email`: 邮箱
+        - `provider_id`: Provider ID
+        - `provider_name`: Provider 显示名称
+        - `endpoint_id`: Endpoint ID
+        - `endpoint_api_format`: Endpoint API 格式
+        - `endpoint_url`: Endpoint 基础 URL
+        - `key_id`: Key ID
+        - `key_name`: Key 名称
+        - `key_prefix`: 脱敏后的 Provider Key
+        - `rate_multiplier`: 速率倍数
+        - `global_model_id`: GlobalModel ID
+        - `model_name`: 模型名称
+        - `model_display_name`: 模型显示名称
+        - `api_format`: API 格式
+        - `created_at`: 创建时间
+        - `expire_at`: 过期时间
+        - `request_count`: 请求计数
+      - `meta`: 分页元数据
+        - `count`: 总数量
+        - `limit`: 每页数量
+        - `offset`: 当前偏移量
+      - `matched_user_id`: 匹配到的用户 ID（当关键词为用户标识时）
     """
     adapter = AdminListAffinitiesAdapter(keyword=keyword, limit=limit, offset=offset)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
@@ -177,10 +243,27 @@ async def clear_user_cache(
     db: Session = Depends(get_db),
 ) -> Any:
     """
-    Clear cache affinity for a specific user
+    清除指定用户的缓存亲和性
 
-    Parameters:
-    - user_identifier: User identifier (username, email, user_id, or API Key ID)
+    清除指定用户或 API Key 的所有缓存亲和性记录。
+    支持按用户维度或单个 API Key 维度清除。
+
+    **路径参数**:
+    - `user_identifier`: 用户标识符，支持以下格式：
+      - 用户名（username）
+      - 邮箱（email）
+      - 用户 UUID（user_id）
+      - API Key ID（清除该 API Key 的缓存）
+
+    **返回字段**:
+    - `status`: 状态（ok）
+    - `message`: 操作结果消息
+    - `user_info`: 用户信息
+      - `user_id`: 用户 ID
+      - `username`: 用户名
+      - `email`: 邮箱
+      - `api_key_id`: API Key ID（当清除单个 API Key 时）
+      - `api_key_name`: API Key 名称（当清除单个 API Key 时）
     """
     adapter = AdminClearUserCacheAdapter(user_identifier=user_identifier)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
@@ -196,13 +279,23 @@ async def clear_single_affinity(
     db: Session = Depends(get_db),
 ) -> Any:
     """
-    Clear a single cache affinity entry
+    清除单条缓存亲和性记录
 
-    Parameters:
-    - affinity_key: API Key ID
-    - endpoint_id: Endpoint ID
-    - model_id: Model ID (GlobalModel ID)
-    - api_format: API format (claude/openai)
+    根据精确的缓存键（affinity_key + endpoint_id + model_id + api_format）
+    清除单条缓存亲和性记录。用于精确控制缓存清除。
+
+    **路径参数**:
+    - `affinity_key`: API Key ID（用于缓存的键）
+    - `endpoint_id`: Endpoint ID
+    - `model_id`: GlobalModel ID
+    - `api_format`: API 格式（如：claude、openai、gemini）
+
+    **返回字段**:
+    - `status`: 状态（ok）
+    - `message`: 操作结果消息
+    - `affinity_key`: API Key ID
+    - `endpoint_id`: Endpoint ID
+    - `model_id`: GlobalModel ID
     """
     adapter = AdminClearSingleAffinityAdapter(
         affinity_key=affinity_key, endpoint_id=endpoint_id, model_id=model_id, api_format=api_format
@@ -216,9 +309,17 @@ async def clear_all_cache(
     db: Session = Depends(get_db),
 ) -> Any:
     """
-    Clear all cache affinities
+    清除所有缓存亲和性
 
-    Warning: This affects all users, use with caution
+    清除系统中所有用户的缓存亲和性记录。此操作会影响所有用户，
+    下次请求时将重新建立缓存亲和性。请谨慎使用。
+
+    **警告**: 此操作影响所有用户，使用前请确认
+
+    **返回字段**:
+    - `status`: 状态（ok）
+    - `message`: 操作结果消息
+    - `count`: 清除的缓存数量
     """
     adapter = AdminClearAllCacheAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
@@ -231,10 +332,19 @@ async def clear_provider_cache(
     db: Session = Depends(get_db),
 ) -> Any:
     """
-    Clear cache affinities for a specific provider
+    清除指定 Provider 的缓存亲和性
 
-    Parameters:
-    - provider_id: Provider ID
+    清除与指定 Provider 相关的所有缓存亲和性记录。
+    当 Provider 配置变更或下线时使用。
+
+    **路径参数**:
+    - `provider_id`: Provider ID
+
+    **返回字段**:
+    - `status`: 状态（ok）
+    - `message`: 操作结果消息
+    - `provider_id`: Provider ID
+    - `count`: 清除的缓存数量
     """
     adapter = AdminClearProviderCacheAdapter(provider_id=provider_id)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
@@ -248,9 +358,25 @@ async def get_cache_config(
     """
     获取缓存相关配置
 
-    返回:
-    - 缓存TTL
-    - 缓存预留比例
+    获取缓存亲和性功能的配置参数，包括缓存 TTL、预留比例、
+    动态预留机制配置等。
+
+    **返回字段**:
+    - `status`: 状态（ok）
+    - `data`: 配置数据
+      - `cache_ttl_seconds`: 缓存亲和性有效期（秒）
+      - `cache_reservation_ratio`: 静态预留比例（已被动态预留替代）
+      - `dynamic_reservation`: 动态预留机制配置
+        - `enabled`: 是否启用
+        - `config`: 配置参数
+          - `probe_phase_requests`: 探测阶段请求数阈值
+          - `probe_reservation`: 探测阶段预留比例
+          - `stable_min_reservation`: 稳定阶段最小预留比例
+          - `stable_max_reservation`: 稳定阶段最大预留比例
+          - `low_load_threshold`: 低负载阈值
+          - `high_load_threshold`: 高负载阈值
+        - `description`: 各参数说明
+      - `description`: 配置说明
     """
     adapter = AdminCacheConfigAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
@@ -262,7 +388,31 @@ async def get_cache_metrics(
     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。
+    获取缓存调度指标（Prometheus 格式）
+
+    以 Prometheus 文本格式输出缓存调度器的监控指标，
+    方便接入 Prometheus/Grafana 等监控系统。
+
+    **返回格式**: Prometheus 文本格式（Content-Type: text/plain）
+
+    **指标列表**:
+    - `cache_scheduler_total_batches`: 总批次数
+    - `cache_scheduler_last_batch_size`: 最后一批候选数
+    - `cache_scheduler_total_candidates`: 总候选数
+    - `cache_scheduler_last_candidate_count`: 最后一批候选计数
+    - `cache_scheduler_cache_hits`: 缓存命中次数
+    - `cache_scheduler_cache_misses`: 缓存未命中次数
+    - `cache_scheduler_cache_hit_rate`: 缓存命中率
+    - `cache_scheduler_concurrency_denied`: 并发拒绝次数
+    - `cache_scheduler_avg_candidates_per_batch`: 平均每批候选数
+    - `cache_affinity_total`: 总缓存亲和性数量
+    - `cache_affinity_hits`: 亲和性命中次数
+    - `cache_affinity_misses`: 亲和性未命中次数
+    - `cache_affinity_hit_rate`: 亲和性命中率
+    - `cache_affinity_invalidations`: 亲和性失效次数
+    - `cache_affinity_provider_switches`: Provider 切换次数
+    - `cache_affinity_key_switches`: Key 切换次数
+    - `cache_scheduler_info`: 调度器信息（label: scheduler）
     """
     adapter = AdminCacheMetricsAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
@@ -669,7 +819,7 @@ class AdminListAffinitiesAdapter(AdminApiAdapter):
                 "username": user.username if user else None,
                 "email": user.email if user else None,
                 "provider_id": provider_id,
-                "provider_name": provider.display_name if provider else None,
+                "provider_name": provider.name if provider else None,
                 "endpoint_id": endpoint_id,
                 "endpoint_api_format": (
                     endpoint.api_format if endpoint and endpoint.api_format else None
@@ -998,10 +1148,39 @@ async def get_model_mapping_cache_stats(
     """
     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
+    获取模型解析缓存的详细统计信息，包括各类型缓存键数量、
+    映射关系列表、Provider 级别的模型映射缓存等。
+
+    **返回字段**:
+    - `status`: 状态（ok）
+    - `data`: 统计数据
+      - `available`: Redis 是否可用
+      - `message`: 提示消息（当 Redis 未启用时）
+      - `ttl_seconds`: 缓存 TTL（秒）
+      - `total_keys`: 总缓存键数量
+      - `breakdown`: 各类型缓存键数量分解
+        - `model_by_id`: Model ID 缓存数量
+        - `model_by_provider_global`: Provider-GlobalModel 缓存数量
+        - `global_model_by_id`: GlobalModel ID 缓存数量
+        - `global_model_by_name`: GlobalModel 名称缓存数量
+        - `global_model_resolve`: GlobalModel 解析缓存数量
+      - `mappings`: 模型映射列表（最多 100 条）
+        - `mapping_name`: 映射名称（别名）
+        - `global_model_name`: GlobalModel 名称
+        - `global_model_display_name`: GlobalModel 显示名称
+        - `providers`: 使用该映射的 Provider 列表
+        - `ttl`: 缓存剩余 TTL（秒）
+      - `provider_model_mappings`: Provider 级别的模型映射（最多 100 条）
+        - `provider_id`: Provider ID
+        - `provider_name`: Provider 名称
+        - `global_model_id`: GlobalModel ID
+        - `global_model_name`: GlobalModel 名称
+        - `global_model_display_name`: GlobalModel 显示名称
+        - `provider_model_name`: Provider 侧的模型名称
+        - `aliases`: 别名列表
+        - `ttl`: 缓存剩余 TTL（秒）
+        - `hit_count`: 缓存命中次数
+      - `unmapped`: 未映射或无效的缓存条目
     """
     adapter = AdminModelMappingCacheStatsAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
@@ -1015,7 +1194,15 @@ async def clear_all_model_mapping_cache(
     """
     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
+    清除系统中所有的模型映射缓存，包括 Model、GlobalModel、
+    模型解析等所有相关缓存。下次请求时将重新从数据库查询。
+
+    **警告**: 此操作会影响所有模型解析，请谨慎使用
+
+    **返回字段**:
+    - `status`: 状态（ok）
+    - `message`: 操作结果消息
+    - `deleted_count`: 删除的缓存键数量
     """
     adapter = AdminClearAllModelMappingCacheAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
@@ -1030,8 +1217,17 @@ async def clear_model_mapping_cache_by_name(
     """
     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）
+    根据模型名称清除相关的映射缓存，包括 resolve 缓存和 name 缓存。
+    用于更新单个模型的配置后刷新缓存。
+
+    **路径参数**:
+    - `model_name`: 模型名称（可以是 GlobalModel.name 或映射名称）
+
+    **返回字段**:
+    - `status`: 状态（ok）
+    - `message`: 操作结果消息
+    - `model_name`: 模型名称
+    - `deleted_keys`: 删除的缓存键列表
     """
     adapter = AdminClearModelMappingCacheByNameAdapter(model_name=model_name)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
@@ -1047,9 +1243,19 @@ async def clear_provider_model_mapping_cache(
     """
     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
+    清除特定 Provider 和 GlobalModel 组合的映射缓存及其命中次数统计。
+    用于 Provider 模型配置更新后刷新缓存。
+
+    **路径参数**:
+    - `provider_id`: Provider ID
+    - `global_model_id`: GlobalModel ID
+
+    **返回字段**:
+    - `status`: 状态（ok）
+    - `message`: 操作结果消息
+    - `provider_id`: Provider ID
+    - `global_model_id`: GlobalModel ID
+    - `deleted_keys`: 删除的缓存键列表
     """
     adapter = AdminClearProviderModelMappingCacheAdapter(
         provider_id=provider_id, global_model_id=global_model_id
@@ -1163,9 +1369,7 @@ class AdminModelMappingCacheStatsAdapter(AdminApiAdapter):
                                     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
-                                            )
+                                            provider_names.append(provider.name)
                                             continue
                                         # 检查是否在映射列表中
                                         if model.provider_model_mappings:
@@ -1175,9 +1379,7 @@ class AdminModelMappingCacheStatsAdapter(AdminApiAdapter):
                                                 if isinstance(a, dict)
                                             ]
                                             if mapping_name in mapping_list:
-                                                provider_names.append(
-                                                    provider.display_name or provider.name
-                                                )
+                                                provider_names.append(provider.name)
                                     provider_names = sorted(list(set(provider_names)))
 
                                 mappings.append({
@@ -1267,7 +1469,7 @@ class AdminModelMappingCacheStatsAdapter(AdminApiAdapter):
 
                                     provider_model_mappings.append({
                                         "provider_id": provider_id,
-                                        "provider_name": provider.display_name or provider.name,
+                                        "provider_name": provider.name,
                                         "global_model_id": global_model_id,
                                         "global_model_name": global_model.name,
                                         "global_model_display_name": global_model.display_name,

src/api/admin/monitoring/trace.py

@@ -71,7 +71,47 @@ async def get_request_trace(
     request: Request,
     db: Session = Depends(get_db),
 ):
-    """获取特定请求的完整追踪信息"""
+    """
+    获取请求的完整追踪信息
+
+    获取指定请求的完整链路追踪信息，包括所有候选（candidates）的执行情况。
+
+    **路径参数**:
+    - `request_id`: 请求 ID
+
+    **返回字段**:
+    - `request_id`: 请求 ID
+    - `total_candidates`: 候选总数
+    - `final_status`: 最终状态（success: 成功，failed: 失败，streaming: 流式传输中，pending: 等待中）
+    - `total_latency_ms`: 总延迟（毫秒）
+    - `candidates`: 候选列表，每个候选包含：
+      - `id`: 候选 ID
+      - `request_id`: 请求 ID
+      - `candidate_index`: 候选索引
+      - `retry_index`: 重试序号
+      - `provider_id`: 提供商 ID
+      - `provider_name`: 提供商名称
+      - `provider_website`: 提供商官网
+      - `endpoint_id`: 端点 ID
+      - `endpoint_name`: 端点名称（API 格式）
+      - `key_id`: 密钥 ID
+      - `key_name`: 密钥名称
+      - `key_preview`: 密钥脱敏预览
+      - `key_capabilities`: 密钥支持的能力
+      - `required_capabilities`: 请求需要的能力标签
+      - `status`: 状态（pending, success, failed, skipped）
+      - `skip_reason`: 跳过原因
+      - `is_cached`: 是否缓存命中
+      - `status_code`: HTTP 状态码
+      - `error_type`: 错误类型
+      - `error_message`: 错误信息
+      - `latency_ms`: 延迟（毫秒）
+      - `concurrent_requests`: 并发请求数
+      - `extra_data`: 额外数据
+      - `created_at`: 创建时间
+      - `started_at`: 开始时间
+      - `finished_at`: 完成时间
+    """
 
     adapter = AdminGetRequestTraceAdapter(request_id=request_id)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
@@ -85,9 +125,23 @@ async def get_provider_failure_rate(
     db: Session = Depends(get_db),
 ):
     """
-    获取某个 Provider 的失败率统计
+    获取提供商的失败率统计
 
-    需要管理员权限
+    获取指定提供商最近的失败率统计信息。需要管理员权限。
+
+    **路径参数**:
+    - `provider_id`: 提供商 ID
+
+    **查询参数**:
+    - `limit`: 统计最近的尝试数量，默认 100，最大 1000
+
+    **返回字段**:
+    - `provider_id`: 提供商 ID
+    - `total_attempts`: 总尝试次数
+    - `success_count`: 成功次数
+    - `failed_count`: 失败次数
+    - `failure_rate`: 失败率（百分比）
+    - `avg_latency_ms`: 平均延迟（毫秒）
     """
     adapter = AdminProviderFailureRateAdapter(provider_id=provider_id, limit=limit)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)

src/api/admin/provider_query.py

@@ -13,10 +13,11 @@ 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.config.constants import TimeoutDefaults
 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, ProviderEndpoint, User
+from src.models.database import Provider, ProviderAPIKey, ProviderEndpoint, User
 from src.utils.auth_utils import get_current_user
 
 router = APIRouter(prefix="/api/admin/provider-query", tags=["Provider Query"])
@@ -81,10 +82,13 @@ async def query_available_models(
     Returns:
         所有端点的模型列表（合并）
     """
-    # 获取提供商及其端点
+    # 获取提供商及其端点和 API Keys
     provider = (
         db.query(Provider)
-        .options(joinedload(Provider.endpoints).joinedload(ProviderEndpoint.api_keys))
+        .options(
+            joinedload(Provider.endpoints),
+            joinedload(Provider.api_keys),
+        )
         .filter(Provider.id == request.provider_id)
         .first()
     )
@@ -95,49 +99,70 @@ async def query_available_models(
     # 收集所有活跃端点的配置
     endpoint_configs: list[dict] = []
 
+    # 构建 api_format -> endpoint 映射
+    format_to_endpoint: dict[str, ProviderEndpoint] = {}
+    for endpoint in provider.endpoints:
+        if endpoint.is_active:
+            format_to_endpoint[endpoint.api_format] = endpoint
+
     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:
-                    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 endpoint_configs:
-                break
+        # 指定了特定的 API Key（从 provider.api_keys 查找）
+        api_key = next(
+            (key for key in provider.api_keys if key.id == request.api_key_id),
+            None
+        )
+
+        if not api_key:
+            raise HTTPException(status_code=404, detail="API Key not found")
+
+        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")
+
+        # 根据 Key 的 api_formats 找对应的 Endpoint
+        key_formats = api_key.api_formats or []
+        for fmt in key_formats:
+            endpoint = format_to_endpoint.get(fmt)
+            if endpoint:
+                endpoint_configs.append({
+                    "api_key": api_key_value,
+                    "base_url": endpoint.base_url,
+                    "api_format": fmt,
+                    "extra_headers": endpoint.headers,
+                })
 
         if not endpoint_configs:
-            raise HTTPException(status_code=404, detail="API Key not found")
+            raise HTTPException(
+                status_code=400,
+                detail="No matching endpoint found for this API Key's formats"
+            )
     else:
-        # 遍历所有活跃端点，每个端点取第一个可用的 Key
+        # 遍历所有活跃端点，为每个端点找一个支持该格式的 Key
         for endpoint in provider.endpoints:
-            if not endpoint.is_active or not endpoint.api_keys:
+            if not endpoint.is_active:
                 continue
 
-            # 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
+            # 找第一个支持该格式的可用 Key
+            for api_key in provider.api_keys:
+                if not api_key.is_active:
+                    continue
+                key_formats = api_key.api_formats or []
+                if endpoint.api_format not in key_formats:
+                    continue
+                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
+                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")
@@ -214,7 +239,6 @@ async def query_available_models(
         "provider": {
             "id": provider.id,
             "name": provider.name,
-            "display_name": provider.display_name,
         },
     }
 
@@ -229,17 +253,14 @@ async def test_model(
     测试模型连接性
 
     向指定提供商的指定模型发送测试请求，验证模型是否可用
-
-    Args:
-        request: 测试请求
-
-    Returns:
-        测试结果
     """
-    # 获取提供商及其端点
+    # 获取提供商及其端点和 Keys
     provider = (
         db.query(Provider)
-        .options(joinedload(Provider.endpoints).joinedload(ProviderEndpoint.api_keys))
+        .options(
+            joinedload(Provider.endpoints),
+            joinedload(Provider.api_keys),
+        )
         .filter(Provider.id == request.provider_id)
         .first()
     )
@@ -247,28 +268,38 @@ async def test_model(
     if not provider:
         raise HTTPException(status_code=404, detail="Provider not found")
 
-    # 找到合适的端点和API Key
-    endpoint_config = None
+    # 构建 api_format -> endpoint 映射
+    format_to_endpoint: dict[str, ProviderEndpoint] = {}
+    for ep in provider.endpoints:
+        if ep.is_active:
+            format_to_endpoint[ep.api_format] = ep
+
+    # 找到合适的端点和 API Key
     endpoint = None
     api_key = None
 
     if request.api_key_id:
-        # 使用指定的API Key
-        for ep in provider.endpoints:
-            for key in ep.api_keys:
-                if key.id == request.api_key_id and key.is_active and ep.is_active:
-                    endpoint = ep
-                    api_key = key
+        # 使用指定的 API Key
+        api_key = next(
+            (key for key in provider.api_keys if key.id == request.api_key_id and key.is_active),
+            None
+        )
+        if api_key:
+            # 找到该 Key 支持的第一个活跃 Endpoint
+            for fmt in (api_key.api_formats or []):
+                if fmt in format_to_endpoint:
+                    endpoint = format_to_endpoint[fmt]
                     break
-            if endpoint:
-                break
     else:
         # 使用第一个可用的端点和密钥
         for ep in provider.endpoints:
-            if not ep.is_active or not ep.api_keys:
+            if not ep.is_active:
                 continue
-            for key in ep.api_keys:
-                if key.is_active:
+            # 找支持该格式的第一个可用 Key
+            for key in provider.api_keys:
+                if not key.is_active:
+                    continue
+                if ep.api_format in (key.api_formats or []):
                     endpoint = ep
                     api_key = key
                     break
@@ -284,14 +315,14 @@ async def test_model(
         logger.error(f"[test-model] Failed to decrypt API key: {e}")
         raise HTTPException(status_code=500, detail="Failed to decrypt API key")
 
-    # 构建请求配置
+    # 构建请求配置（timeout 从 Provider 读取）
     endpoint_config = {
         "api_key": api_key_value,
         "api_key_id": api_key.id,  # 添加API Key ID用于用量记录
         "base_url": endpoint.base_url,
         "api_format": endpoint.api_format,
         "extra_headers": endpoint.headers,
-        "timeout": endpoint.timeout or 30.0,
+        "timeout": provider.timeout or TimeoutDefaults.HTTP_REQUEST,
     }
 
     try:
@@ -304,7 +335,6 @@ async def test_model(
                 "provider": {
                     "id": provider.id,
                     "name": provider.name,
-                    "display_name": provider.display_name,
                 },
                 "model": request.model_name,
             }
@@ -325,7 +355,6 @@ async def test_model(
                     "provider": {
                         "id": provider.id,
                         "name": provider.name,
-                        "display_name": provider.display_name,
                     },
                     "model": request.model_name,
                 }
@@ -415,7 +444,6 @@ async def test_model(
                 "provider": {
                     "id": provider.id,
                     "name": provider.name,
-                    "display_name": provider.display_name,
                 },
                 "model": request.model_name,
                 "endpoint": {
@@ -433,7 +461,6 @@ async def test_model(
             "provider": {
                 "id": provider.id,
                 "name": provider.name,
-                "display_name": provider.display_name,
             },
             "model": request.model_name,
             "endpoint": {

src/api/admin/provider_strategy.py

@@ -39,6 +39,31 @@ async def update_provider_billing(
     request: Request,
     db: Session = Depends(get_db),
 ):
+    """
+    更新提供商计费配置
+
+    更新指定提供商的计费策略、配额设置和优先级配置。
+
+    **路径参数**:
+    - `provider_id`: 提供商 ID
+
+    **请求体字段**:
+    - `billing_type`: 计费类型（pay_as_you_go、subscription、prepaid、monthly_quota）
+    - `monthly_quota_usd`: 月度配额（美元），可选
+    - `quota_reset_day`: 配额重置周期（天数，1-365），默认 30
+    - `quota_last_reset_at`: 当前周期开始时间，可选（设置后会自动同步该周期内的历史使用量）
+    - `quota_expires_at`: 配额过期时间，可选
+    - `rpm_limit`: 每分钟请求数限制，可选
+    - `provider_priority`: 提供商优先级（0-200），默认 100
+
+    **返回字段**:
+    - `message`: 操作结果信息
+    - `provider`: 更新后的提供商信息
+      - `id`: 提供商 ID
+      - `name`: 提供商名称
+      - `billing_type`: 计费类型
+      - `provider_priority`: 提供商优先级
+    """
     adapter = AdminProviderBillingAdapter(provider_id=provider_id)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -50,6 +75,35 @@ async def get_provider_stats(
     hours: int = 24,
     db: Session = Depends(get_db),
 ):
+    """
+    获取提供商统计数据
+
+    获取指定提供商的计费信息和使用统计数据。
+
+    **路径参数**:
+    - `provider_id`: 提供商 ID
+
+    **查询参数**:
+    - `hours`: 统计时间范围（小时），默认 24
+
+    **返回字段**:
+    - `provider_id`: 提供商 ID
+    - `provider_name`: 提供商名称
+    - `period_hours`: 统计时间范围
+    - `billing_info`: 计费信息
+      - `billing_type`: 计费类型
+      - `monthly_quota_usd`: 月度配额
+      - `monthly_used_usd`: 月度已使用
+      - `quota_remaining_usd`: 剩余配额
+      - `quota_expires_at`: 配额过期时间
+    - `usage_stats`: 使用统计
+      - `total_requests`: 总请求数
+      - `successful_requests`: 成功请求数
+      - `failed_requests`: 失败请求数
+      - `success_rate`: 成功率
+      - `avg_response_time_ms`: 平均响应时间（毫秒）
+      - `total_cost_usd`: 总成本（美元）
+    """
     adapter = AdminProviderStatsAdapter(provider_id=provider_id, hours=hours)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -67,6 +121,20 @@ async def reset_provider_quota(
 
 @router.get("/strategies")
 async def list_available_strategies(request: Request, db: Session = Depends(get_db)):
+    """
+    获取可用负载均衡策略列表
+
+    列出系统中所有已注册的负载均衡策略插件。
+
+    **返回字段**:
+    - `strategies`: 策略列表
+      - `name`: 策略名称
+      - `priority`: 策略优先级
+      - `version`: 策略版本
+      - `description`: 策略描述
+      - `author`: 策略作者
+    - `total`: 策略总数
+    """
     adapter = AdminListStrategiesAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -93,7 +161,6 @@ class AdminProviderBillingAdapter(AdminApiAdapter):
         provider.billing_type = config.billing_type
         provider.monthly_quota_usd = config.monthly_quota_usd
         provider.quota_reset_day = config.quota_reset_day
-        provider.rpm_limit = config.rpm_limit
         provider.provider_priority = config.provider_priority
 
         from dateutil import parser
@@ -190,13 +257,6 @@ class AdminProviderStatsAdapter(AdminApiAdapter):
                         provider.quota_expires_at.isoformat() if provider.quota_expires_at else None
                     ),
                 },
-                "rpm_info": {
-                    "rpm_limit": provider.rpm_limit,
-                    "rpm_used": provider.rpm_used,
-                    "rpm_reset_at": (
-                        provider.rpm_reset_at.isoformat() if provider.rpm_reset_at else None
-                    ),
-                },
                 "usage_stats": {
                     "total_requests": total_requests,
                     "successful_requests": total_success,
@@ -224,8 +284,6 @@ class AdminProviderResetQuotaAdapter(AdminApiAdapter):
 
         old_used = provider.monthly_used_usd
         provider.monthly_used_usd = 0.0
-        provider.rpm_used = 0
-        provider.rpm_reset_at = None
         db.commit()
 
         logger.info(f"Manually reset quota for provider {provider.name}")

src/api/admin/providers/models.py

@@ -49,7 +49,36 @@ async def list_provider_models(
     limit: int = 100,
     db: Session = Depends(get_db),
 ) -> List[ModelResponse]:
-    """获取提供商的所有模型（管理员）"""
+    """
+    获取提供商的所有模型
+
+    获取指定提供商的模型列表，支持分页和状态过滤。
+
+    **路径参数**:
+    - `provider_id`: 提供商 ID
+
+    **查询参数**:
+    - `is_active`: 可选的活跃状态过滤，true 仅返回活跃模型，false 返回禁用模型，不传则返回全部
+    - `skip`: 跳过的记录数，默认为 0
+    - `limit`: 返回的最大记录数，默认为 100
+
+    **返回字段**（数组，每项包含）:
+    - `id`: 模型 ID
+    - `provider_id`: 提供商 ID
+    - `global_model_id`: 全局模型 ID
+    - `provider_model_name`: 提供商模型名称
+    - `is_active`: 是否启用
+    - `input_price_per_1m`: 输入价格（每百万 token）
+    - `output_price_per_1m`: 输出价格（每百万 token）
+    - `cache_creation_price_per_1m`: 缓存创建价格（每百万 token）
+    - `cache_read_price_per_1m`: 缓存读取价格（每百万 token）
+    - `price_per_request`: 每次请求价格
+    - `supports_vision`: 是否支持视觉
+    - `supports_function_calling`: 是否支持函数调用
+    - `supports_streaming`: 是否支持流式输出
+    - `created_at`: 创建时间
+    - `updated_at`: 更新时间
+    """
     adapter = AdminListProviderModelsAdapter(
         provider_id=provider_id,
         is_active=is_active,
@@ -66,7 +95,29 @@ async def create_provider_model(
     request: Request,
     db: Session = Depends(get_db),
 ) -> ModelResponse:
-    """创建模型（管理员）"""
+    """
+    创建模型
+
+    为指定提供商创建一个新的模型配置。
+
+    **路径参数**:
+    - `provider_id`: 提供商 ID
+
+    **请求体字段**:
+    - `provider_model_name`: 提供商模型名称（必填）
+    - `global_model_id`: 全局模型 ID（可选，关联到全局模型）
+    - `is_active`: 是否启用（默认 true）
+    - `input_price_per_1m`: 输入价格（每百万 token）（可选）
+    - `output_price_per_1m`: 输出价格（每百万 token）（可选）
+    - `cache_creation_price_per_1m`: 缓存创建价格（每百万 token）（可选）
+    - `cache_read_price_per_1m`: 缓存读取价格（每百万 token）（可选）
+    - `price_per_request`: 每次请求价格（可选）
+    - `supports_vision`: 是否支持视觉（可选）
+    - `supports_function_calling`: 是否支持函数调用（可选）
+    - `supports_streaming`: 是否支持流式输出（可选）
+
+    **返回字段**: 返回创建的模型详细信息（与 GET 单个模型接口返回格式相同）
+    """
     adapter = AdminCreateProviderModelAdapter(provider_id=provider_id, model_data=model_data)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -78,7 +129,32 @@ async def get_provider_model(
     request: Request,
     db: Session = Depends(get_db),
 ) -> ModelResponse:
-    """获取模型详情（管理员）"""
+    """
+    获取模型详情
+
+    获取指定模型的详细配置信息。
+
+    **路径参数**:
+    - `provider_id`: 提供商 ID
+    - `model_id`: 模型 ID
+
+    **返回字段**:
+    - `id`: 模型 ID
+    - `provider_id`: 提供商 ID
+    - `global_model_id`: 全局模型 ID
+    - `provider_model_name`: 提供商模型名称
+    - `is_active`: 是否启用
+    - `input_price_per_1m`: 输入价格（每百万 token）
+    - `output_price_per_1m`: 输出价格（每百万 token）
+    - `cache_creation_price_per_1m`: 缓存创建价格（每百万 token）
+    - `cache_read_price_per_1m`: 缓存读取价格（每百万 token）
+    - `price_per_request`: 每次请求价格
+    - `supports_vision`: 是否支持视觉
+    - `supports_function_calling`: 是否支持函数调用
+    - `supports_streaming`: 是否支持流式输出
+    - `created_at`: 创建时间
+    - `updated_at`: 更新时间
+    """
     adapter = AdminGetProviderModelAdapter(provider_id=provider_id, model_id=model_id)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -91,7 +167,30 @@ async def update_provider_model(
     request: Request,
     db: Session = Depends(get_db),
 ) -> ModelResponse:
-    """更新模型（管理员）"""
+    """
+    更新模型配置
+
+    更新指定模型的配置信息。只需传入需要更新的字段，未传入的字段保持不变。
+
+    **路径参数**:
+    - `provider_id`: 提供商 ID
+    - `model_id`: 模型 ID
+
+    **请求体字段**（所有字段可选）:
+    - `provider_model_name`: 提供商模型名称
+    - `global_model_id`: 全局模型 ID
+    - `is_active`: 是否启用
+    - `input_price_per_1m`: 输入价格（每百万 token）
+    - `output_price_per_1m`: 输出价格（每百万 token）
+    - `cache_creation_price_per_1m`: 缓存创建价格（每百万 token）
+    - `cache_read_price_per_1m`: 缓存读取价格（每百万 token）
+    - `price_per_request`: 每次请求价格
+    - `supports_vision`: 是否支持视觉
+    - `supports_function_calling`: 是否支持函数调用
+    - `supports_streaming`: 是否支持流式输出
+
+    **返回字段**: 返回更新后的模型详细信息（与 GET 单个模型接口返回格式相同）
+    """
     adapter = AdminUpdateProviderModelAdapter(
         provider_id=provider_id,
         model_id=model_id,
@@ -107,7 +206,18 @@ async def delete_provider_model(
     request: Request,
     db: Session = Depends(get_db),
 ):
-    """删除模型（管理员）"""
+    """
+    删除模型
+
+    删除指定的模型配置。注意：此操作不可逆。
+
+    **路径参数**:
+    - `provider_id`: 提供商 ID
+    - `model_id`: 模型 ID
+
+    **返回字段**:
+    - `message`: 删除成功提示信息
+    """
     adapter = AdminDeleteProviderModelAdapter(provider_id=provider_id, model_id=model_id)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -119,7 +229,29 @@ async def batch_create_provider_models(
     request: Request,
     db: Session = Depends(get_db),
 ) -> List[ModelResponse]:
-    """批量创建模型（管理员）"""
+    """
+    批量创建模型
+
+    为指定提供商批量创建多个模型配置。
+
+    **路径参数**:
+    - `provider_id`: 提供商 ID
+
+    **请求体**: 模型数据数组，每项包含：
+    - `provider_model_name`: 提供商模型名称（必填）
+    - `global_model_id`: 全局模型 ID（可选）
+    - `is_active`: 是否启用（默认 true）
+    - `input_price_per_1m`: 输入价格（每百万 token）（可选）
+    - `output_price_per_1m`: 输出价格（每百万 token）（可选）
+    - `cache_creation_price_per_1m`: 缓存创建价格（每百万 token）（可选）
+    - `cache_read_price_per_1m`: 缓存读取价格（每百万 token）（可选）
+    - `price_per_request`: 每次请求价格（可选）
+    - `supports_vision`: 是否支持视觉（可选）
+    - `supports_function_calling`: 是否支持函数调用（可选）
+    - `supports_streaming`: 是否支持流式输出（可选）
+
+    **返回字段**: 返回创建的模型列表（与 GET 模型列表接口返回格式相同）
+    """
     adapter = AdminBatchCreateModelsAdapter(provider_id=provider_id, models_data=models_data)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -134,10 +266,23 @@ async def get_provider_available_source_models(
     db: Session = Depends(get_db),
 ):
     """
-    获取该 Provider 支持的所有统一模型名（source_model）
+    获取提供商支持的可用源模型
 
-    包括：
-    1. 直连模型（Model.provider_model_name 直接作为统一模型名）
+    获取该提供商支持的所有统一模型名（source_model），包含价格和能力信息。
+
+    **路径参数**:
+    - `provider_id`: 提供商 ID
+
+    **返回字段**:
+    - `models`: 可用源模型数组，每项包含：
+      - `global_model_name`: 全局模型名称
+      - `display_name`: 显示名称
+      - `provider_model_name`: 提供商模型名称
+      - `model_id`: 模型 ID
+      - `price`: 价格信息（包含 input_price_per_1m, output_price_per_1m, cache_creation_price_per_1m, cache_read_price_per_1m, price_per_request）
+      - `capabilities`: 能力信息（包含 supports_vision, supports_function_calling, supports_streaming）
+      - `is_active`: 是否启用
+    - `total`: 总数
     """
     adapter = AdminGetProviderAvailableSourceModelsAdapter(provider_id=provider_id)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
@@ -153,7 +298,27 @@ async def batch_assign_global_models_to_provider(
     request: Request,
     db: Session = Depends(get_db),
 ) -> BatchAssignModelsToProviderResponse:
-    """批量为 Provider 关联 GlobalModels（自动继承价格和能力配置）"""
+    """
+    批量关联全局模型
+
+    批量为提供商关联全局模型，自动继承全局模型的价格和能力配置。
+
+    **路径参数**:
+    - `provider_id`: 提供商 ID
+
+    **请求体字段**:
+    - `global_model_ids`: 全局模型 ID 数组（必填）
+
+    **返回字段**:
+    - `success`: 成功关联的模型数组，每项包含：
+      - `global_model_id`: 全局模型 ID
+      - `global_model_name`: 全局模型名称
+      - `model_id`: 新创建的模型 ID
+    - `errors`: 失败的模型数组，每项包含：
+      - `global_model_id`: 全局模型 ID
+      - `global_model_name`: 全局模型名称（如果可用）
+      - `error`: 错误信息
+    """
     adapter = AdminBatchAssignModelsToProviderAdapter(
         provider_id=provider_id, payload=payload
     )
@@ -173,10 +338,32 @@ async def import_models_from_upstream(
     """
     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
+    从上游提供商导入模型列表。导入的模型作为独立的 ProviderModel 存储，
+    不会自动创建 GlobalModel。后续需要手动关联 GlobalModel 才能参与路由。
+
+    **流程说明**:
+    1. 检查模型是否已存在于当前 Provider（按 provider_model_name 匹配）
+    2. 创建新的 ProviderModel（global_model_id = NULL）
+    3. 支持设置价格覆盖（tiered_pricing, price_per_request）
+
+    **路径参数**:
+    - `provider_id`: 提供商 ID
+
+    **请求体字段**:
+    - `model_ids`: 模型 ID 数组（必填，每个 ID 长度 1-100 字符）
+    - `tiered_pricing`: 可选的阶梯计费配置（应用于所有导入的模型）
+    - `price_per_request`: 可选的按次计费价格（应用于所有导入的模型）
+
+    **返回字段**:
+    - `success`: 成功导入的模型数组，每项包含：
+      - `model_id`: 模型 ID
+      - `provider_model_id`: 提供商模型 ID
+      - `global_model_id`: 全局模型 ID（如果已关联）
+      - `global_model_name`: 全局模型名称（如果已关联）
+      - `created_global_model`: 是否新创建了全局模型（始终为 false）
+    - `errors`: 失败的模型数组，每项包含：
+      - `model_id`: 模型 ID
+      - `error`: 错误信息
     """
     adapter = AdminImportFromUpstreamAdapter(provider_id=provider_id, payload=payload)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
@@ -453,7 +640,7 @@ class AdminBatchAssignModelsToProviderAdapter(AdminApiAdapter):
 
 @dataclass
 class AdminImportFromUpstreamAdapter(AdminApiAdapter):
-    """从上游提供商导入模型"""
+    """从上游提供商导入模型（不创建 GlobalModel，作为独立 ProviderModel）"""
 
     provider_id: str
     payload: ImportFromUpstreamRequest
@@ -467,16 +654,13 @@ class AdminImportFromUpstreamAdapter(AdminApiAdapter):
         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,
-                }
-            ]
-        }
+        # 获取价格覆盖配置
+        tiered_pricing = None
+        price_per_request = None
+        if hasattr(self.payload, 'tiered_pricing') and self.payload.tiered_pricing:
+            tiered_pricing = self.payload.tiered_pricing
+        if hasattr(self.payload, 'price_per_request') and self.payload.price_per_request is not None:
+            price_per_request = self.payload.price_per_request
 
         for model_id in self.payload.model_ids:
             # 输入验证：检查 model_id 长度
@@ -493,56 +677,37 @@ class AdminImportFromUpstreamAdapter(AdminApiAdapter):
                 # 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
+                    # 1. 检查是否已存在同名的 ProviderModel
                     existing = (
                         db.query(Model)
                         .filter(
                             Model.provider_id == self.provider_id,
-                            Model.global_model_id == global_model.id,
+                            Model.provider_model_name == model_id,
                         )
                         .first()
                     )
                     if existing:
-                        # 已存在关联，提交 savepoint 并记录成功
+                        # 已存在，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,
+                                global_model_id=existing.global_model_id or "",
+                                global_model_name=existing.global_model.name if existing.global_model else "",
                                 provider_model_id=existing.id,
-                                created_global_model=created_global_model,
+                                created_global_model=False,
                             )
                         )
                         continue
 
-                    # 4. 创建新的 Model 记录
+                    # 2. 创建新的 Model 记录（不关联 GlobalModel）
                     new_model = Model(
                         provider_id=self.provider_id,
-                        global_model_id=global_model.id,
-                        provider_model_name=global_model.name,
+                        global_model_id=None,  # 独立模型，不关联 GlobalModel
+                        provider_model_name=model_id,
                         is_active=True,
+                        tiered_pricing=tiered_pricing,
+                        price_per_request=price_per_request,
                     )
                     db.add(new_model)
                     db.flush()
@@ -552,12 +717,15 @@ class AdminImportFromUpstreamAdapter(AdminApiAdapter):
                     success.append(
                         ImportFromUpstreamSuccessItem(
                             model_id=model_id,
-                            global_model_id=global_model.id,
-                            global_model_name=global_model.name,
+                            global_model_id="",  # 未关联
+                            global_model_name="",  # 未关联
                             provider_model_id=new_model.id,
-                            created_global_model=created_global_model,
+                            created_global_model=False,
                         )
                     )
+                    logger.info(
+                        f"Created independent ProviderModel: {model_id} for provider {provider.name}"
+                    )
                 except Exception as e:
                     # 回滚到 savepoint
                     savepoint.rollback()
@@ -568,11 +736,9 @@ class AdminImportFromUpstreamAdapter(AdminApiAdapter):
 
         db.commit()
         logger.info(
-            f"Imported {len(success)} models from upstream to provider {provider.name} by {context.user.username}"
+            f"Imported {len(success)} independent models to provider {provider.name} by {context.user.username}"
         )
 
-        # 清除 /v1/models 列表缓存
-        if success:
-            await invalidate_models_list_cache()
+        # 不需要清除 /v1/models 缓存，因为独立模型不参与路由
 
         return ImportFromUpstreamResponse(success=success, errors=errors)

src/api/admin/providers/routes.py

@@ -11,9 +11,11 @@ from src.api.base.admin_adapter import AdminApiAdapter
 from src.api.base.pipeline import ApiRequestPipeline
 from src.core.enums import ProviderBillingType
 from src.core.exceptions import InvalidRequestException, NotFoundException
+from src.core.logger import logger
 from src.database import get_db
 from src.models.admin_requests import CreateProviderRequest, UpdateProviderRequest
 from src.models.database import Provider
+from src.services.cache.provider_cache import ProviderCacheService
 
 router = APIRouter(tags=["Provider CRUD"])
 pipeline = ApiRequestPipeline()
@@ -27,24 +29,114 @@ async def list_providers(
     is_active: Optional[bool] = None,
     db: Session = Depends(get_db),
 ):
+    """
+    获取提供商列表
+
+    获取所有提供商的基本信息列表，支持分页和状态过滤。
+
+    **查询参数**:
+    - `skip`: 跳过的记录数，用于分页，默认为 0
+    - `limit`: 返回的最大记录数，范围 1-500，默认为 100
+    - `is_active`: 可选的活跃状态过滤，true 仅返回活跃提供商，false 返回禁用提供商，不传则返回全部
+
+    **返回字段**:
+    - `id`: 提供商 ID
+    - `name`: 提供商名称（唯一）
+    - `api_format`: API 格式（如 claude、openai、gemini 等）
+    - `base_url`: API 基础 URL
+    - `api_key`: API 密钥（脱敏显示）
+    - `priority`: 优先级
+    - `is_active`: 是否活跃
+    - `created_at`: 创建时间
+    - `updated_at`: 更新时间
+    """
     adapter = AdminListProvidersAdapter(skip=skip, limit=limit, is_active=is_active)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
 @router.post("/")
 async def create_provider(request: Request, db: Session = Depends(get_db)):
+    """
+    创建新提供商
+
+    创建一个新的 AI 模型提供商配置。
+
+    **请求体字段**:
+    - `name`: 提供商名称（必填，唯一）
+    - `description`: 描述信息（可选）
+    - `website`: 官网地址（可选）
+    - `billing_type`: 计费类型（可选，pay_as_you_go/subscription/prepaid，默认 pay_as_you_go）
+    - `monthly_quota_usd`: 月度配额（美元）（可选）
+    - `quota_reset_day`: 配额重置日期（1-31）（可选）
+    - `quota_last_reset_at`: 上次配额重置时间（可选）
+    - `quota_expires_at`: 配额过期时间（可选）
+    - `provider_priority`: 提供商优先级（数字越小优先级越高，默认 100）
+    - `is_active`: 是否启用（默认 true）
+    - `concurrent_limit`: 并发限制（可选）
+    - `timeout`: 请求超时（秒，可选）
+    - `max_retries`: 最大重试次数（可选）
+    - `proxy`: 代理配置（可选）
+    - `config`: 额外配置信息（JSON，可选）
+
+    **返回字段**:
+    - `id`: 新创建的提供商 ID
+    - `name`: 提供商名称
+    - `message`: 成功提示信息
+    """
     adapter = AdminCreateProviderAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
 @router.put("/{provider_id}")
 async def update_provider(provider_id: str, request: Request, db: Session = Depends(get_db)):
+    """
+    更新提供商配置
+
+    更新指定提供商的配置信息。只需传入需要更新的字段，未传入的字段保持不变。
+
+    **路径参数**:
+    - `provider_id`: 提供商 ID
+
+    **请求体字段**（所有字段可选）:
+    - `name`: 提供商名称
+    - `description`: 描述信息
+    - `website`: 官网地址
+    - `billing_type`: 计费类型（pay_as_you_go/subscription/prepaid）
+    - `monthly_quota_usd`: 月度配额（美元）
+    - `quota_reset_day`: 配额重置日期（1-31）
+    - `quota_last_reset_at`: 上次配额重置时间
+    - `quota_expires_at`: 配额过期时间
+    - `provider_priority`: 提供商优先级
+    - `is_active`: 是否启用
+    - `concurrent_limit`: 并发限制
+    - `timeout`: 请求超时（秒）
+    - `max_retries`: 最大重试次数
+    - `proxy`: 代理配置
+    - `config`: 额外配置信息（JSON）
+
+    **返回字段**:
+    - `id`: 提供商 ID
+    - `name`: 提供商名称
+    - `is_active`: 是否启用
+    - `message`: 成功提示信息
+    """
     adapter = AdminUpdateProviderAdapter(provider_id=provider_id)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
 @router.delete("/{provider_id}")
 async def delete_provider(provider_id: str, request: Request, db: Session = Depends(get_db)):
+    """
+    删除提供商
+
+    删除指定的提供商。注意：此操作会级联删除关联的端点、密钥和模型配置。
+
+    **路径参数**:
+    - `provider_id`: 提供商 ID
+
+    **返回字段**:
+    - `message`: 删除成功提示信息
+    """
     adapter = AdminDeleteProviderAdapter(provider_id=provider_id)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -73,7 +165,6 @@ class AdminListProvidersAdapter(AdminApiAdapter):
                 {
                     "id": provider.id,
                     "name": provider.name,
-                    "display_name": provider.display_name,
                     "api_format": api_format.value if api_format else None,
                     "base_url": base_url,
                     "api_key": "***" if api_key else None,
@@ -125,7 +216,6 @@ class AdminCreateProviderAdapter(AdminApiAdapter):
             # 创建 Provider 对象
             provider = Provider(
                 name=validated_data.name,
-                display_name=validated_data.display_name,
                 description=validated_data.description,
                 website=validated_data.website,
                 billing_type=billing_type,
@@ -133,11 +223,12 @@ class AdminCreateProviderAdapter(AdminApiAdapter):
                 quota_reset_day=validated_data.quota_reset_day,
                 quota_last_reset_at=validated_data.quota_last_reset_at,
                 quota_expires_at=validated_data.quota_expires_at,
-                rpm_limit=validated_data.rpm_limit,
                 provider_priority=validated_data.provider_priority,
                 is_active=validated_data.is_active,
-                rate_limit=validated_data.rate_limit,
                 concurrent_limit=validated_data.concurrent_limit,
+                timeout=validated_data.timeout,
+                max_retries=validated_data.max_retries,
+                proxy=validated_data.proxy.model_dump() if validated_data.proxy else None,
                 config=validated_data.config,
             )
 
@@ -157,7 +248,6 @@ class AdminCreateProviderAdapter(AdminApiAdapter):
             return {
                 "id": provider.id,
                 "name": provider.name,
-                "display_name": provider.display_name,
                 "message": "提供商创建成功",
             }
         except InvalidRequestException:
@@ -200,6 +290,9 @@ class AdminUpdateProviderAdapter(AdminApiAdapter):
                 if field == "billing_type" and value is not None:
                     # billing_type 需要转换为枚举
                     setattr(provider, field, ProviderBillingType(value))
+                elif field == "proxy" and value is not None:
+                    # proxy 需要转换为 dict（如果是 Pydantic 模型）
+                    setattr(provider, field, value if isinstance(value, dict) else value.model_dump())
                 else:
                     setattr(provider, field, value)
 
@@ -207,6 +300,11 @@ class AdminUpdateProviderAdapter(AdminApiAdapter):
             db.commit()
             db.refresh(provider)
 
+            # 如果更新了 billing_type，清除缓存
+            if "billing_type" in update_data:
+                await ProviderCacheService.invalidate_provider_cache(provider.id)
+                logger.debug(f"已清除 Provider 缓存: {provider.id}")
+
             context.add_audit_metadata(
                 action="update_provider",
                 provider_id=provider.id,

src/api/admin/providers/summary.py

@@ -40,7 +40,40 @@ async def get_providers_summary(
     request: Request,
     db: Session = Depends(get_db),
 ) -> List[ProviderWithEndpointsSummary]:
-    """获取所有 Providers 的摘要信息(包含 Endpoints 和 Keys 统计)"""
+    """
+    获取所有提供商摘要信息
+
+    获取所有提供商的详细摘要信息，包含端点、密钥、模型统计和健康状态。
+
+    **返回字段**（数组，每项包含）:
+    - `id`: 提供商 ID
+    - `name`: 提供商名称
+    - `description`: 描述信息
+    - `website`: 官网地址
+    - `provider_priority`: 优先级
+    - `is_active`: 是否启用
+    - `billing_type`: 计费类型
+    - `monthly_quota_usd`: 月度配额（美元）
+    - `monthly_used_usd`: 本月已使用金额（美元）
+    - `quota_reset_day`: 配额重置日期
+    - `quota_last_reset_at`: 上次配额重置时间
+    - `quota_expires_at`: 配额过期时间
+    - `timeout`: 默认请求超时（秒）
+    - `max_retries`: 默认最大重试次数
+    - `proxy`: 默认代理配置
+    - `total_endpoints`: 端点总数
+    - `active_endpoints`: 活跃端点数
+    - `total_keys`: 密钥总数
+    - `active_keys`: 活跃密钥数
+    - `total_models`: 模型总数
+    - `active_models`: 活跃模型数
+    - `avg_health_score`: 平均健康分数（0-1）
+    - `unhealthy_endpoints`: 不健康端点数（健康分数 < 0.5）
+    - `api_formats`: 支持的 API 格式列表
+    - `endpoint_health_details`: 端点健康详情（包含 api_format, health_score, is_active, active_keys）
+    - `created_at`: 创建时间
+    - `updated_at`: 更新时间
+    """
     adapter = AdminProviderSummaryAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -51,7 +84,43 @@ async def get_provider_summary(
     request: Request,
     db: Session = Depends(get_db),
 ) -> ProviderWithEndpointsSummary:
-    """获取单个 Provider 的摘要信息(包含 Endpoints 和 Keys 统计)"""
+    """
+    获取单个提供商摘要信息
+
+    获取指定提供商的详细摘要信息，包含端点、密钥、模型统计和健康状态。
+
+    **路径参数**:
+    - `provider_id`: 提供商 ID
+
+    **返回字段**:
+    - `id`: 提供商 ID
+    - `name`: 提供商名称
+    - `description`: 描述信息
+    - `website`: 官网地址
+    - `provider_priority`: 优先级
+    - `is_active`: 是否启用
+    - `billing_type`: 计费类型
+    - `monthly_quota_usd`: 月度配额（美元）
+    - `monthly_used_usd`: 本月已使用金额（美元）
+    - `quota_reset_day`: 配额重置日期
+    - `quota_last_reset_at`: 上次配额重置时间
+    - `quota_expires_at`: 配额过期时间
+    - `timeout`: 默认请求超时（秒）
+    - `max_retries`: 默认最大重试次数
+    - `proxy`: 默认代理配置
+    - `total_endpoints`: 端点总数
+    - `active_endpoints`: 活跃端点数
+    - `total_keys`: 密钥总数
+    - `active_keys`: 活跃密钥数
+    - `total_models`: 模型总数
+    - `active_models`: 活跃模型数
+    - `avg_health_score`: 平均健康分数（0-1）
+    - `unhealthy_endpoints`: 不健康端点数（健康分数 < 0.5）
+    - `api_formats`: 支持的 API 格式列表
+    - `endpoint_health_details`: 端点健康详情（包含 api_format, health_score, is_active, active_keys）
+    - `created_at`: 创建时间
+    - `updated_at`: 更新时间
+    """
     provider = db.query(Provider).filter(Provider.id == provider_id).first()
     if not provider:
         raise NotFoundException(f"Provider {provider_id} not found")
@@ -67,7 +136,34 @@ async def get_provider_health_monitor(
     per_endpoint_limit: int = Query(48, ge=10, le=200, description="每个端点的事件数量"),
     db: Session = Depends(get_db),
 ) -> ProviderEndpointHealthMonitorResponse:
-    """获取 Provider 下所有端点的健康监控时间线"""
+    """
+    获取提供商健康监控数据
+
+    获取指定提供商下所有端点的健康监控时间线，包含请求成功率、延迟、错误信息等。
+
+    **路径参数**:
+    - `provider_id`: 提供商 ID
+
+    **查询参数**:
+    - `lookback_hours`: 回溯的小时数，范围 1-72，默认为 6
+    - `per_endpoint_limit`: 每个端点返回的事件数量，范围 10-200，默认为 48
+
+    **返回字段**:
+    - `provider_id`: 提供商 ID
+    - `provider_name`: 提供商名称
+    - `generated_at`: 生成时间
+    - `endpoints`: 端点健康监控数据数组，每项包含：
+      - `endpoint_id`: 端点 ID
+      - `api_format`: API 格式
+      - `is_active`: 是否活跃
+      - `total_attempts`: 总请求次数
+      - `success_count`: 成功次数
+      - `failed_count`: 失败次数
+      - `skipped_count`: 跳过次数
+      - `success_rate`: 成功率（0-1）
+      - `last_event_at`: 最后事件时间
+      - `events`: 事件详情数组（包含 timestamp, status, status_code, latency_ms, error_type, error_message）
+    """
 
     adapter = AdminProviderHealthMonitorAdapter(
         provider_id=provider_id,
@@ -84,7 +180,30 @@ async def update_provider_settings(
     request: Request,
     db: Session = Depends(get_db),
 ) -> ProviderWithEndpointsSummary:
-    """更新 Provider 基础配置（display_name, description, priority, weight 等）"""
+    """
+    更新提供商基础配置
+
+    更新提供商的基础配置信息，如名称、描述、优先级等。只需传入需要更新的字段。
+
+    **路径参数**:
+    - `provider_id`: 提供商 ID
+
+    **请求体字段**（所有字段可选）:
+    - `name`: 提供商名称
+    - `description`: 描述信息
+    - `website`: 官网地址
+    - `provider_priority`: 优先级
+    - `is_active`: 是否启用
+    - `billing_type`: 计费类型
+    - `monthly_quota_usd`: 月度配额（美元）
+    - `quota_reset_day`: 配额重置日期
+    - `quota_expires_at`: 配额过期时间
+    - `timeout`: 默认请求超时（秒）
+    - `max_retries`: 默认最大重试次数
+    - `proxy`: 默认代理配置
+
+    **返回字段**: 返回更新后的提供商摘要信息（与 GET /summary 接口返回格式相同）
+    """
 
     adapter = AdminUpdateProviderSettingsAdapter(provider_id=provider_id, update_data=update_data)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
@@ -95,18 +214,18 @@ def _build_provider_summary(db: Session, provider: Provider) -> ProviderWithEndp
 
     total_endpoints = len(endpoints)
     active_endpoints = sum(1 for e in endpoints if e.is_active)
-    endpoint_ids = [e.id for e in endpoints]
 
     # Key 统计（合并为单个查询）
-    total_keys = 0
-    active_keys = 0
-    if endpoint_ids:
-        key_stats = db.query(
+    key_stats = (
+        db.query(
             func.count(ProviderAPIKey.id).label("total"),
             func.sum(case((ProviderAPIKey.is_active == True, 1), else_=0)).label("active"),
-        ).filter(ProviderAPIKey.endpoint_id.in_(endpoint_ids)).first()
-        total_keys = key_stats.total or 0
-        active_keys = int(key_stats.active or 0)
+        )
+        .filter(ProviderAPIKey.provider_id == provider.id)
+        .first()
+    )
+    total_keys = key_stats.total or 0
+    active_keys = int(key_stats.active or 0)
 
     # Model 统计（合并为单个查询）
     model_stats = db.query(
@@ -118,25 +237,34 @@ def _build_provider_summary(db: Session, provider: Provider) -> ProviderWithEndp
 
     api_formats = [e.api_format for e in endpoints]
 
-    # 优化: 一次性加载所有 endpoint 的 keys，避免 N+1 查询
-    all_keys = []
-    if endpoint_ids:
-        all_keys = (
-            db.query(ProviderAPIKey).filter(ProviderAPIKey.endpoint_id.in_(endpoint_ids)).all()
-        )
+    # 优化: 一次性加载 Provider 的 keys，避免 N+1 查询
+    all_keys = db.query(ProviderAPIKey).filter(ProviderAPIKey.provider_id == provider.id).all()
 
-    # 按 endpoint_id 分组 keys
-    keys_by_endpoint: dict[str, list[ProviderAPIKey]] = {}
+    # 按 api_formats 分组 keys（通过 api_formats 关联）
+    format_to_endpoint_id: dict[str, str] = {e.api_format: e.id for e in endpoints}
+    keys_by_endpoint: dict[str, list[ProviderAPIKey]] = {e.id: [] for e in endpoints}
     for key in all_keys:
-        if key.endpoint_id not in keys_by_endpoint:
-            keys_by_endpoint[key.endpoint_id] = []
-        keys_by_endpoint[key.endpoint_id].append(key)
+        formats = key.api_formats or []
+        for fmt in formats:
+            endpoint_id = format_to_endpoint_id.get(fmt)
+            if endpoint_id:
+                keys_by_endpoint[endpoint_id].append(key)
 
     endpoint_health_map: dict[str, float] = {}
     for endpoint in endpoints:
         keys = keys_by_endpoint.get(endpoint.id, [])
         if keys:
-            health_scores = [k.health_score for k in keys if k.health_score is not None]
+            # 从 health_by_format 获取对应格式的健康度
+            api_fmt = endpoint.api_format
+            health_scores = []
+            for k in keys:
+                health_by_format = k.health_by_format or {}
+                if api_fmt in health_by_format:
+                    score = health_by_format[api_fmt].get("health_score")
+                    if score is not None:
+                        health_scores.append(float(score))
+                else:
+                    health_scores.append(1.0)  # 默认健康度
             avg_health = sum(health_scores) / len(health_scores) if health_scores else 1.0
             endpoint_health_map[endpoint.id] = avg_health
         else:
@@ -164,7 +292,6 @@ def _build_provider_summary(db: Session, provider: Provider) -> ProviderWithEndp
     return ProviderWithEndpointsSummary(
         id=provider.id,
         name=provider.name,
-        display_name=provider.display_name,
         description=provider.description,
         website=provider.website,
         provider_priority=provider.provider_priority,
@@ -175,9 +302,9 @@ def _build_provider_summary(db: Session, provider: Provider) -> ProviderWithEndp
         quota_reset_day=provider.quota_reset_day,
         quota_last_reset_at=provider.quota_last_reset_at,
         quota_expires_at=provider.quota_expires_at,
-        rpm_limit=provider.rpm_limit,
-        rpm_used=provider.rpm_used,
-        rpm_reset_at=provider.rpm_reset_at,
+        timeout=provider.timeout,
+        max_retries=provider.max_retries,
+        proxy=provider.proxy,
         total_endpoints=total_endpoints,
         active_endpoints=active_endpoints,
         total_keys=total_keys,
@@ -221,7 +348,7 @@ class AdminProviderHealthMonitorAdapter(AdminApiAdapter):
         if not endpoint_ids:
             response = ProviderEndpointHealthMonitorResponse(
                 provider_id=provider.id,
-                provider_name=provider.display_name or provider.name,
+                provider_name=provider.name,
                 generated_at=now,
                 endpoints=[],
             )
@@ -296,7 +423,7 @@ class AdminProviderHealthMonitorAdapter(AdminApiAdapter):
 
         response = ProviderEndpointHealthMonitorResponse(
             provider_id=provider.id,
-            provider_name=provider.display_name or provider.name,
+            provider_name=provider.name,
             generated_at=now,
             endpoints=endpoint_monitors,
         )

src/api/admin/security/ip_management.py

@@ -18,7 +18,7 @@ from src.core.logger import logger
 from src.database import get_db
 from src.services.rate_limit.ip_limiter import IPRateLimiter
 
-router = APIRouter(prefix="/api/admin/security/ip", tags=["IP Security"])
+router = APIRouter(prefix="/api/admin/security/ip", tags=["Admin - Security"])
 pipeline = ApiRequestPipeline()
 
 
@@ -56,42 +56,110 @@ class RemoveIPFromWhitelistRequest(BaseModel):
 
 @router.post("/blacklist")
 async def add_to_blacklist(request: Request, db: Session = Depends(get_db)):
-    """Add IP to blacklist"""
+    """
+    添加 IP 到黑名单
+
+    将指定 IP 地址添加到黑名单，被加入黑名单的 IP 将无法访问系统。需要管理员权限。
+
+    **请求体字段**:
+    - `ip_address`: IP 地址
+    - `reason`: 加入黑名单的原因
+    - `ttl`: 可选，过期时间（秒），不指定表示永久
+
+    **返回字段**:
+    - `success`: 是否成功
+    - `message`: 操作结果信息
+    - `reason`: 加入黑名单的原因
+    - `ttl`: 过期时间（秒或"永久"）
+    """
     adapter = AddToBlacklistAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=ApiMode.ADMIN)
 
 
 @router.delete("/blacklist/{ip_address}")
 async def remove_from_blacklist(ip_address: str, request: Request, db: Session = Depends(get_db)):
-    """Remove IP from blacklist"""
+    """
+    从黑名单移除 IP
+
+    将指定 IP 地址从黑名单中移除。需要管理员权限。
+
+    **路径参数**:
+    - `ip_address`: IP 地址
+
+    **返回字段**:
+    - `success`: 是否成功
+    - `message`: 操作结果信息
+    """
     adapter = RemoveFromBlacklistAdapter(ip_address=ip_address)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=ApiMode.ADMIN)
 
 
 @router.get("/blacklist/stats")
 async def get_blacklist_stats(request: Request, db: Session = Depends(get_db)):
-    """Get blacklist statistics"""
+    """
+    获取黑名单统计信息
+
+    获取黑名单的统计信息和列表。需要管理员权限。
+
+    **返回字段**:
+    - `total`: 黑名单总数
+    - `items`: 黑名单列表，每个项包含：
+      - `ip`: IP 地址
+      - `reason`: 加入原因
+      - `added_at`: 添加时间
+      - `ttl`: 剩余有效时间（秒）
+    """
     adapter = GetBlacklistStatsAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=ApiMode.ADMIN)
 
 
 @router.post("/whitelist")
 async def add_to_whitelist(request: Request, db: Session = Depends(get_db)):
-    """Add IP to whitelist"""
+    """
+    添加 IP 到白名单
+
+    将指定 IP 地址或 CIDR 网段添加到白名单，白名单中的 IP 将跳过速率限制检查。需要管理员权限。
+
+    **请求体字段**:
+    - `ip_address`: IP 地址或 CIDR 格式（如 192.168.1.0/24）
+
+    **返回字段**:
+    - `success`: 是否成功
+    - `message`: 操作结果信息
+    """
     adapter = AddToWhitelistAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=ApiMode.ADMIN)
 
 
 @router.delete("/whitelist/{ip_address}")
 async def remove_from_whitelist(ip_address: str, request: Request, db: Session = Depends(get_db)):
-    """Remove IP from whitelist"""
+    """
+    从白名单移除 IP
+
+    将指定 IP 地址从白名单中移除。需要管理员权限。
+
+    **路径参数**:
+    - `ip_address`: IP 地址
+
+    **返回字段**:
+    - `success`: 是否成功
+    - `message`: 操作结果信息
+    """
     adapter = RemoveFromWhitelistAdapter(ip_address=ip_address)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=ApiMode.ADMIN)
 
 
 @router.get("/whitelist")
 async def get_whitelist(request: Request, db: Session = Depends(get_db)):
-    """Get whitelist"""
+    """
+    获取白名单
+
+    获取当前的 IP 白名单列表。需要管理员权限。
+
+    **返回字段**:
+    - `whitelist`: 白名单 IP 地址列表
+    - `total`: 白名单总数
+    """
     adapter = GetWhitelistAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=ApiMode.ADMIN)
 

src/api/admin/system.py

@@ -42,21 +42,157 @@ def _get_version_from_git() -> str | None:
     return None
 
 
-@router.get("/version")
-async def get_system_version():
-    """获取系统版本信息"""
-    # 优先从 git 获取
+def _get_current_version() -> str:
+    """获取当前版本号"""
     version = _get_version_from_git()
     if version:
-        return {"version": version}
-
-    # 回退到静态版本文件
+        return version
     try:
         from src._version import __version__
 
-        return {"version": __version__}
+        return __version__
     except ImportError:
-        return {"version": "unknown"}
+        return "unknown"
+
+
+def _parse_version(version_str: str) -> tuple:
+    """解析版本号为可比较的元组，支持 3-4 段版本号
+
+    例如:
+    - '0.2.5' -> (0, 2, 5, 0)
+    - '0.2.5.1' -> (0, 2, 5, 1)
+    - 'v0.2.5-4-g1234567' -> (0, 2, 5, 0)
+    """
+    import re
+
+    version_str = version_str.lstrip("v")
+    main_version = re.split(r"[-+]", version_str)[0]
+    try:
+        parts = main_version.split(".")
+        # 标准化为 4 段，便于比较
+        int_parts = [int(p) for p in parts]
+        while len(int_parts) < 4:
+            int_parts.append(0)
+        return tuple(int_parts[:4])
+    except ValueError:
+        return (0, 0, 0, 0)
+
+
+@router.get("/version")
+async def get_system_version():
+    """
+    获取系统版本信息
+
+    获取当前系统的版本号。优先从 git describe 获取，回退到静态版本文件。
+
+    **返回字段**:
+    - `version`: 版本号字符串
+    """
+    return {"version": _get_current_version()}
+
+
+@router.get("/check-update")
+async def check_update():
+    """
+    检查系统更新
+
+    从 GitHub Tags 获取最新版本并与当前版本对比。
+
+    **返回字段**:
+    - `current_version`: 当前版本号
+    - `latest_version`: 最新版本号
+    - `has_update`: 是否有更新可用
+    - `release_url`: 最新版本的 GitHub 页面链接
+    """
+    import httpx
+
+    from src.clients.http_client import HTTPClientPool
+
+    current_version = _get_current_version()
+    github_repo = "Aethersailor/Aether"
+    github_tags_url = f"https://api.github.com/repos/{github_repo}/tags"
+
+    try:
+        async with HTTPClientPool.get_temp_client(
+            timeout=httpx.Timeout(connect=5.0, read=10.0, write=5.0, pool=5.0)
+        ) as client:
+            response = await client.get(
+                github_tags_url,
+                headers={
+                    "Accept": "application/vnd.github.v3+json",
+                    "User-Agent": f"Aether/{current_version}",
+                },
+                params={"per_page": 10},
+            )
+
+            if response.status_code != 200:
+                return {
+                    "current_version": current_version,
+                    "latest_version": None,
+                    "has_update": False,
+                    "release_url": None,
+                    "error": f"GitHub API 返回错误: {response.status_code}",
+                }
+
+            tags = response.json()
+            if not tags:
+                return {
+                    "current_version": current_version,
+                    "latest_version": None,
+                    "has_update": False,
+                    "release_url": None,
+                    "error": None,
+                }
+
+            # 找到最新的版本 tag（按版本号排序，而非时间）
+            version_tags = []
+            for tag in tags:
+                tag_name = tag.get("name", "")
+                if tag_name.startswith("v") or tag_name[0].isdigit():
+                    version_tags.append((tag_name, _parse_version(tag_name)))
+
+            if not version_tags:
+                return {
+                    "current_version": current_version,
+                    "latest_version": None,
+                    "has_update": False,
+                    "release_url": None,
+                    "error": None,
+                }
+
+            # 按版本号排序，取最大的
+            version_tags.sort(key=lambda x: x[1], reverse=True)
+            latest_tag = version_tags[0][0]
+            latest_version = latest_tag.lstrip("v")
+
+            current_tuple = _parse_version(current_version)
+            latest_tuple = _parse_version(latest_version)
+            has_update = latest_tuple > current_tuple
+
+            return {
+                "current_version": current_version,
+                "latest_version": latest_version,
+                "has_update": has_update,
+                "release_url": f"https://github.com/{github_repo}/releases/tag/{latest_tag}",
+                "error": None,
+            }
+
+    except httpx.TimeoutException:
+        return {
+            "current_version": current_version,
+            "latest_version": None,
+            "has_update": False,
+            "release_url": None,
+            "error": "检查更新超时",
+        }
+    except Exception as e:
+        return {
+            "current_version": current_version,
+            "latest_version": None,
+            "has_update": False,
+            "release_url": None,
+            "error": f"检查更新失败: {str(e)}",
+        }
 
 
 pipeline = ApiRequestPipeline()
@@ -64,7 +200,16 @@ pipeline = ApiRequestPipeline()
 
 @router.get("/settings")
 async def get_system_settings(request: Request, db: Session = Depends(get_db)):
-    """获取系统设置（管理员）"""
+    """
+    获取系统设置
+
+    获取系统的全局设置信息。需要管理员权限。
+
+    **返回字段**:
+    - `default_provider`: 默认提供商名称
+    - `default_model`: 默认模型名称
+    - `enable_usage_tracking`: 是否启用使用情况追踪
+    """
 
     adapter = AdminGetSystemSettingsAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
@@ -72,7 +217,19 @@ async def get_system_settings(request: Request, db: Session = Depends(get_db)):
 
 @router.put("/settings")
 async def update_system_settings(http_request: Request, db: Session = Depends(get_db)):
-    """更新系统设置（管理员）"""
+    """
+    更新系统设置
+
+    更新系统的全局设置。需要管理员权限。
+
+    **请求体字段**:
+    - `default_provider`: 可选，默认提供商名称（空字符串表示清除设置）
+    - `default_model`: 可选，默认模型名称（空字符串表示清除设置）
+    - `enable_usage_tracking`: 可选，是否启用使用情况追踪
+
+    **返回字段**:
+    - `message`: 操作结果信息
+    """
 
     adapter = AdminUpdateSystemSettingsAdapter()
     return await pipeline.run(adapter=adapter, http_request=http_request, db=db, mode=adapter.mode)
@@ -80,7 +237,14 @@ async def update_system_settings(http_request: Request, db: Session = Depends(ge
 
 @router.get("/configs")
 async def get_all_system_configs(request: Request, db: Session = Depends(get_db)):
-    """获取所有系统配置（管理员）"""
+    """
+    获取所有系统配置
+
+    获取系统中所有的配置项。需要管理员权限。
+
+    **返回字段**:
+    - 配置项的键值对字典
+    """
 
     adapter = AdminGetAllConfigsAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
@@ -88,7 +252,19 @@ async def get_all_system_configs(request: Request, db: Session = Depends(get_db)
 
 @router.get("/configs/{key}")
 async def get_system_config(key: str, request: Request, db: Session = Depends(get_db)):
-    """获取特定系统配置（管理员）"""
+    """
+    获取特定系统配置
+
+    获取指定配置项的值。需要管理员权限。
+
+    **路径参数**:
+    - `key`: 配置项键名
+
+    **返回字段**:
+    - `key`: 配置项键名
+    - `value`: 配置项的值（敏感配置项不返回实际值）
+    - `is_set`: 可选，对于敏感配置项，指示是否已设置
+    """
 
     adapter = AdminGetSystemConfigAdapter(key=key)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
@@ -100,7 +276,24 @@ async def set_system_config(
     request: Request,
     db: Session = Depends(get_db),
 ):
-    """设置系统配置（管理员）"""
+    """
+    设置系统配置
+
+    设置或更新指定配置项的值。需要管理员权限。
+
+    **路径参数**:
+    - `key`: 配置项键名
+
+    **请求体字段**:
+    - `value`: 配置项的值
+    - `description`: 可选，配置项描述
+
+    **返回字段**:
+    - `key`: 配置项键名
+    - `value`: 配置项的值（敏感配置项显示为 ********）
+    - `description`: 配置项描述
+    - `updated_at`: 更新时间
+    """
 
     adapter = AdminSetSystemConfigAdapter(key=key)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
@@ -108,7 +301,17 @@ async def set_system_config(
 
 @router.delete("/configs/{key}")
 async def delete_system_config(key: str, request: Request, db: Session = Depends(get_db)):
-    """删除系统配置（管理员）"""
+    """
+    删除系统配置
+
+    删除指定的配置项。需要管理员权限。
+
+    **路径参数**:
+    - `key`: 配置项键名
+
+    **返回字段**:
+    - `message`: 操作结果信息
+    """
 
     adapter = AdminDeleteSystemConfigAdapter(key=key)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
@@ -116,20 +319,54 @@ async def delete_system_config(key: str, request: Request, db: Session = Depends
 
 @router.get("/stats")
 async def get_system_stats(request: Request, db: Session = Depends(get_db)):
+    """
+    获取系统统计信息
+
+    获取系统的整体统计数据。需要管理员权限。
+
+    **返回字段**:
+    - `users`: 用户统计（total: 总用户数, active: 活跃用户数）
+    - `providers`: 提供商统计（total: 总提供商数, active: 活跃提供商数）
+    - `api_keys`: API Key 总数
+    - `requests`: 请求总数
+    """
     adapter = AdminSystemStatsAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
 @router.post("/cleanup")
 async def trigger_cleanup(request: Request, db: Session = Depends(get_db)):
-    """Manually trigger usage record cleanup task"""
+    """
+    手动触发清理任务
+
+    手动触发使用记录清理任务，清理过期的请求/响应数据。需要管理员权限。
+
+    **返回字段**:
+    - `message`: 操作结果信息
+    - `stats`: 清理统计信息
+      - `total_records`: 总记录数统计（before, after, deleted）
+      - `body_fields`: 请求/响应体字段清理统计（before, after, cleaned）
+      - `header_fields`: 请求/响应头字段清理统计（before, after, cleaned）
+    - `timestamp`: 清理完成时间
+    """
     adapter = AdminTriggerCleanupAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
 @router.get("/api-formats")
 async def get_api_formats(request: Request, db: Session = Depends(get_db)):
-    """获取所有可用的API格式列表"""
+    """
+    获取所有可用的 API 格式列表
+
+    获取系统支持的所有 API 格式及其元数据。需要管理员权限。
+
+    **返回字段**:
+    - `formats`: API 格式列表，每个格式包含：
+      - `value`: 格式值
+      - `label`: 显示名称
+      - `default_path`: 默认路径
+      - `aliases`: 别名列表
+    """
     adapter = AdminGetApiFormatsAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -493,36 +730,6 @@ class AdminExportConfigAdapter(AdminApiAdapter):
             )
             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,
@@ -530,12 +737,44 @@ class AdminExportConfigAdapter(AdminApiAdapter):
                         "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,
+                        "proxy": ep.proxy,
+                    }
+                )
+
+            # 导出 Provider Keys（按 provider_id 归属，包含 api_formats）
+            keys = (
+                db.query(ProviderAPIKey)
+                .filter(ProviderAPIKey.provider_id == provider.id)
+                .order_by(ProviderAPIKey.internal_priority.asc(), ProviderAPIKey.created_at.asc())
+                .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,
+                        "api_formats": key.api_formats or [],
+                        "rate_multiplier": key.rate_multiplier,
+                        "rate_multipliers": key.rate_multipliers,
+                        "internal_priority": key.internal_priority,
+                        "global_priority": key.global_priority,
+                        "rpm_limit": key.rpm_limit,
+                        "allowed_models": key.allowed_models,
+                        "capabilities": key.capabilities,
+                        "cache_ttl_minutes": key.cache_ttl_minutes,
+                        "max_probe_interval_minutes": key.max_probe_interval_minutes,
+                        "is_active": key.is_active,
                     }
                 )
 
@@ -567,25 +806,26 @@ class AdminExportConfigAdapter(AdminApiAdapter):
             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,
+                    "timeout": provider.timeout,
+                    "max_retries": provider.max_retries,
+                    "proxy": provider.proxy,
                     "config": provider.config,
                     "endpoints": endpoints_data,
+                    "api_keys": keys_data,
                     "models": models_data,
                 }
             )
 
         return {
-            "version": "1.0",
+            "version": "2.0",
             "exported_at": datetime.now(timezone.utc).isoformat(),
             "global_models": global_models_data,
             "providers": providers_data,
@@ -614,7 +854,7 @@ class AdminImportConfigAdapter(AdminApiAdapter):
 
         # 验证配置版本
         version = payload.get("version")
-        if version != "1.0":
+        if version != "2.0":
             raise InvalidRequestException(f"不支持的配置版本: {version}")
 
         # 获取导入选项
@@ -703,8 +943,8 @@ class AdminImportConfigAdapter(AdminApiAdapter):
                         )
                     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.name = prov_data.get(
+                            "name", existing_provider.name
                         )
                         existing_provider.description = prov_data.get("description")
                         existing_provider.website = prov_data.get("website")
@@ -718,15 +958,18 @@ class AdminImportConfigAdapter(AdminApiAdapter):
                         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.timeout = prov_data.get("timeout", existing_provider.timeout)
+                        existing_provider.max_retries = prov_data.get(
+                            "max_retries", existing_provider.max_retries
+                        )
+                        existing_provider.proxy = prov_data.get("proxy", existing_provider.proxy)
                         existing_provider.config = prov_data.get("config")
                         existing_provider.updated_at = datetime.now(timezone.utc)
                         stats["providers"]["updated"] += 1
@@ -739,17 +982,17 @@ class AdminImportConfigAdapter(AdminApiAdapter):
                     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"),
+                        timeout=prov_data.get("timeout"),
+                        max_retries=prov_data.get("max_retries"),
+                        proxy=prov_data.get("proxy"),
                         config=prov_data.get("config"),
                     )
                     db.add(new_provider)
@@ -769,7 +1012,6 @@ class AdminImportConfigAdapter(AdminApiAdapter):
                     )
 
                     if existing_ep:
-                        endpoint_id = existing_ep.id
                         if merge_mode == "skip":
                             stats["endpoints"]["skipped"] += 1
                         elif merge_mode == "error":
@@ -782,12 +1024,11 @@ class AdminImportConfigAdapter(AdminApiAdapter):
                             )
                             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.max_retries = ep_data.get("max_retries", 2)
                             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.proxy = ep_data.get("proxy")
                             existing_ep.updated_at = datetime.now(timezone.utc)
                             stats["endpoints"]["updated"] += 1
                     else:
@@ -798,69 +1039,107 @@ class AdminImportConfigAdapter(AdminApiAdapter):
                             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"),
+                            max_retries=ep_data.get("max_retries", 2),
                             is_active=ep_data.get("is_active", True),
                             custom_path=ep_data.get("custom_path"),
                             config=ep_data.get("config"),
+                            proxy=ep_data.get("proxy"),
                         )
                         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
+                # 导入 Provider Keys（按 provider_id 归属）
+                endpoint_format_rows = (
+                    db.query(ProviderEndpoint.api_format)
+                    .filter(ProviderEndpoint.provider_id == provider_id)
+                    .all()
+                )
+                endpoint_formats: set[str] = set()
+                for (api_format,) in endpoint_format_rows:
+                    fmt = api_format.value if hasattr(api_format, "value") else str(api_format)
+                    endpoint_formats.add(fmt.strip().upper())
+                existing_keys = (
+                    db.query(ProviderAPIKey)
+                    .filter(ProviderAPIKey.provider_id == provider_id)
+                    .all()
+                )
+                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),
+                for key_data in prov_data.get("api_keys", []):
+                    if not key_data.get("api_key"):
+                        stats["errors"].append(
+                            f"跳过空 API Key (Provider: {prov_data['name']})"
                         )
-                        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
+                        continue
+
+                    plaintext_key = key_data["api_key"]
+                    if plaintext_key in existing_key_values:
+                        stats["keys"]["skipped"] += 1
+                        continue
+
+                    raw_formats = key_data.get("api_formats") or []
+                    if not isinstance(raw_formats, list) or len(raw_formats) == 0:
+                        stats["errors"].append(
+                            f"跳过无 api_formats 的 Key (Provider: {prov_data['name']})"
+                        )
+                        continue
+
+                    normalized_formats: list[str] = []
+                    seen: set[str] = set()
+                    missing_formats: list[str] = []
+                    for fmt in raw_formats:
+                        if not isinstance(fmt, str):
+                            continue
+                        fmt_upper = fmt.strip().upper()
+                        if not fmt_upper or fmt_upper in seen:
+                            continue
+                        seen.add(fmt_upper)
+                        if endpoint_formats and fmt_upper not in endpoint_formats:
+                            missing_formats.append(fmt_upper)
+                            continue
+                        normalized_formats.append(fmt_upper)
+
+                    if missing_formats:
+                        stats["errors"].append(
+                            f"Key (Provider: {prov_data['name']}) 的 api_formats 未配置对应 Endpoint，已跳过: {missing_formats}"
+                        )
+
+                    if len(normalized_formats) == 0:
+                        stats["keys"]["skipped"] += 1
+                        continue
+
+                    encrypted_key = crypto_service.encrypt(plaintext_key)
+
+                    new_key = ProviderAPIKey(
+                        id=str(uuid.uuid4()),
+                        provider_id=provider_id,
+                        api_formats=normalized_formats,
+                        api_key=encrypted_key,
+                        name=key_data.get("name") or "Imported Key",
+                        note=key_data.get("note"),
+                        rate_multiplier=key_data.get("rate_multiplier", 1.0),
+                        rate_multipliers=key_data.get("rate_multipliers"),
+                        internal_priority=key_data.get("internal_priority", 50),
+                        global_priority=key_data.get("global_priority"),
+                        rpm_limit=key_data.get("rpm_limit"),
+                        allowed_models=key_data.get("allowed_models"),
+                        capabilities=key_data.get("capabilities"),
+                        cache_ttl_minutes=key_data.get("cache_ttl_minutes", 5),
+                        max_probe_interval_minutes=key_data.get("max_probe_interval_minutes", 32),
+                        is_active=key_data.get("is_active", True),
+                        health_by_format={},
+                        circuit_breaker_by_format={},
+                    )
+                    db.add(new_key)
+                    existing_key_values.add(plaintext_key)
+                    stats["keys"]["created"] += 1
 
                 # 导入 Models
                 for model_data in prov_data.get("models", []):
@@ -1001,7 +1280,6 @@ class AdminExportUsersAdapter(AdminApiAdapter):
                 "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,
@@ -1038,7 +1316,7 @@ class AdminExportUsersAdapter(AdminApiAdapter):
                     "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_api_formats": user.allowed_api_formats,
                     "allowed_models": user.allowed_models,
                     "model_capability_settings": user.model_capability_settings,
                     "quota_usd": user.quota_usd,
@@ -1130,7 +1408,6 @@ class AdminImportUsersAdapter(AdminApiAdapter):
                 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"),
@@ -1174,7 +1451,7 @@ class AdminImportUsersAdapter(AdminApiAdapter):
                         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_api_formats = user_data.get("allowed_api_formats")
                         existing_user.allowed_models = user_data.get("allowed_models")
                         existing_user.model_capability_settings = user_data.get(
                             "model_capability_settings"
@@ -1198,7 +1475,7 @@ class AdminImportUsersAdapter(AdminApiAdapter):
                         password_hash=user_data.get("password_hash", ""),
                         role=role,
                         allowed_providers=user_data.get("allowed_providers"),
-                        allowed_endpoints=user_data.get("allowed_endpoints"),
+                        allowed_api_formats=user_data.get("allowed_api_formats"),
                         allowed_models=user_data.get("allowed_models"),
                         model_capability_settings=user_data.get("model_capability_settings"),
                         quota_usd=user_data.get("quota_usd"),

src/api/admin/usage/routes.py

@@ -39,12 +39,21 @@ async def get_usage_aggregation(
     db: Session = Depends(get_db),
 ):
     """
-    Get usage aggregation by specified dimension.
+    获取使用情况聚合统计
 
-    - group_by=model: Aggregate by model
-    - group_by=user: Aggregate by user
-    - group_by=provider: Aggregate by provider
-    - group_by=api_format: Aggregate by API format
+    按指定维度聚合使用情况统计数据。
+
+    **查询参数**:
+    - `group_by`: 必需，聚合维度，可选值：model（按模型）、user（按用户）、provider（按提供商）、api_format（按 API 格式）
+    - `start_date`: 可选，开始日期（ISO 格式）
+    - `end_date`: 可选，结束日期（ISO 格式）
+    - `limit`: 返回数量限制，默认 20，最大 100
+
+    **返回字段**:
+    - 按模型聚合时：model, request_count, total_tokens, total_cost, actual_cost
+    - 按用户聚合时：user_id, email, username, request_count, total_tokens, total_cost
+    - 按提供商聚合时：provider_id, provider, request_count, total_tokens, total_cost, actual_cost, avg_response_time_ms, success_rate, error_count
+    - 按 API 格式聚合时：api_format, request_count, total_tokens, total_cost, actual_cost, avg_response_time_ms
     """
     if group_by == "model":
         adapter = AdminUsageByModelAdapter(start_date=start_date, end_date=end_date, limit=limit)
@@ -69,6 +78,25 @@ async def get_usage_stats(
     end_date: Optional[datetime] = None,
     db: Session = Depends(get_db),
 ):
+    """
+    获取使用情况总体统计
+
+    获取指定时间范围内的使用情况总体统计数据。
+
+    **查询参数**:
+    - `start_date`: 可选，开始日期（ISO 格式）
+    - `end_date`: 可选，结束日期（ISO 格式）
+
+    **返回字段**:
+    - `total_requests`: 总请求数
+    - `total_tokens`: 总 token 数
+    - `total_cost`: 总成本（美元）
+    - `total_actual_cost`: 实际总成本（美元）
+    - `avg_response_time`: 平均响应时间（秒）
+    - `error_count`: 错误请求数
+    - `error_rate`: 错误率（百分比）
+    - `cache_stats`: 缓存统计信息（cache_creation_tokens, cache_read_tokens, cache_creation_cost, cache_read_cost）
+    """
     adapter = AdminUsageStatsAdapter(start_date=start_date, end_date=end_date)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -79,9 +107,12 @@ async def get_activity_heatmap(
     db: Session = Depends(get_db),
 ):
     """
-    Get activity heatmap data for the past 365 days.
+    获取活动热力图数据
 
-    This endpoint is cached for 5 minutes to reduce database load.
+    获取过去 365 天的活动热力图数据。此接口缓存 5 分钟以减少数据库负载。
+
+    **返回字段**:
+    - 按日期聚合的请求数、token 数、成本等统计数据
     """
     adapter = AdminActivityHeatmapAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
@@ -102,6 +133,33 @@ async def get_usage_records(
     offset: int = Query(0, ge=0),
     db: Session = Depends(get_db),
 ):
+    """
+    获取使用记录列表
+
+    获取详细的使用记录列表，支持多种筛选条件。
+
+    **查询参数**:
+    - `start_date`: 可选，开始日期（ISO 格式）
+    - `end_date`: 可选，结束日期（ISO 格式）
+    - `search`: 可选，通用搜索关键词（支持用户名、密钥名、模型名、提供商名模糊搜索，多个关键词用空格分隔）
+    - `user_id`: 可选，用户 ID 筛选
+    - `username`: 可选，用户名模糊搜索
+    - `model`: 可选，模型名模糊搜索
+    - `provider`: 可选，提供商名称搜索
+    - `status`: 可选，状态筛选（stream: 流式请求，standard: 标准请求，error: 错误请求，pending: 等待中，streaming: 流式中，completed: 已完成，failed: 失败，active: 活跃请求）
+    - `limit`: 返回数量限制，默认 100，最大 500
+    - `offset`: 分页偏移量，默认 0
+
+    **返回字段**:
+    - `records`: 使用记录列表，包含 id, user_id, user_email, username, api_key, provider, model, target_model,
+      input_tokens, output_tokens, cache_creation_input_tokens, cache_read_input_tokens, total_tokens,
+      cost, actual_cost, rate_multiplier, response_time_ms, first_byte_time_ms, created_at, is_stream,
+      input_price_per_1m, output_price_per_1m, cache_creation_price_per_1m, cache_read_price_per_1m,
+      status_code, error_message, status, has_fallback, api_format, api_key_name, request_metadata
+    - `total`: 符合条件的总记录数
+    - `limit`: 当前分页限制
+    - `offset`: 当前分页偏移量
+    """
     adapter = AdminUsageRecordsAdapter(
         start_date=start_date,
         end_date=end_date,
@@ -124,10 +182,19 @@ async def get_active_requests(
     db: Session = Depends(get_db),
 ):
     """
-    获取活跃请求的状态（轻量级接口，用于前端轮询）
+    获取活跃请求的状态
 
+    获取当前活跃（pending/streaming 状态）请求的状态信息。这是一个轻量级接口，适合前端轮询。
+
+    **查询参数**:
+    - `ids`: 可选，逗号分隔的请求 ID 列表，用于查询特定请求的状态
+
+    **行为说明**:
     - 如果提供 ids 参数，只返回这些 ID 对应请求的最新状态
     - 如果不提供 ids，返回所有 pending/streaming 状态的请求
+
+    **返回字段**:
+    - `requests`: 活跃请求列表，包含请求状态信息
     """
     adapter = AdminActiveRequestsAdapter(ids=ids)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
@@ -142,9 +209,49 @@ async def get_usage_detail(
     db: Session = Depends(get_db),
 ):
     """
-    Get detailed information of a specific usage record.
+    获取使用记录详情
 
-    Includes request/response headers and body.
+    获取指定使用记录的详细信息，包括请求/响应的头部和正文。
+
+    **路径参数**:
+    - `usage_id`: 使用记录 ID
+
+    **返回字段**:
+    - `id`: 记录 ID
+    - `request_id`: 请求 ID
+    - `user`: 用户信息（id, username, email）
+    - `api_key`: API Key 信息（id, name, display）
+    - `provider`: 提供商名称
+    - `api_format`: API 格式
+    - `model`: 请求的模型名称
+    - `target_model`: 映射后的目标模型名称
+    - `tokens`: Token 统计（input, output, total）
+    - `cost`: 成本统计（input, output, total）
+    - `cache_creation_input_tokens`: 缓存创建输入 token 数
+    - `cache_read_input_tokens`: 缓存读取输入 token 数
+    - `cache_creation_cost`: 缓存创建成本
+    - `cache_read_cost`: 缓存读取成本
+    - `request_cost`: 请求成本
+    - `input_price_per_1m`: 输入价格（每百万 token）
+    - `output_price_per_1m`: 输出价格（每百万 token）
+    - `cache_creation_price_per_1m`: 缓存创建价格（每百万 token）
+    - `cache_read_price_per_1m`: 缓存读取价格（每百万 token）
+    - `price_per_request`: 每请求价格
+    - `request_type`: 请求类型
+    - `is_stream`: 是否为流式请求
+    - `status_code`: HTTP 状态码
+    - `error_message`: 错误信息
+    - `response_time_ms`: 响应时间（毫秒）
+    - `first_byte_time_ms`: 首字节时间（TTFB，毫秒）
+    - `created_at`: 创建时间
+    - `request_headers`: 请求头
+    - `request_body`: 请求体
+    - `provider_request_headers`: 提供商请求头
+    - `response_headers`: 提供商响应头
+    - `client_response_headers`: 返回给客户端的响应头
+    - `response_body`: 响应体
+    - `metadata`: 提供商响应元数据
+    - `tiered_pricing`: 阶梯计费信息（如适用）
     """
     adapter = AdminUsageDetailAdapter(usage_id=usage_id)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
@@ -247,8 +354,8 @@ class AdminUsageByModelAdapter(AdminApiAdapter):
         )
         # 过滤掉 pending/streaming 状态的请求（尚未完成的请求不应计入统计）
         query = query.filter(Usage.status.notin_(["pending", "streaming"]))
-        # 过滤掉 unknown/pending provider（请求未到达任何提供商）
-        query = query.filter(Usage.provider.notin_(["unknown", "pending"]))
+        # 过滤掉 unknown/pending provider_name（请求未到达任何提供商）
+        query = query.filter(Usage.provider_name.notin_(["unknown", "pending"]))
 
         if self.start_date:
             query = query.filter(Usage.created_at >= self.start_date)
@@ -459,8 +566,8 @@ class AdminUsageByApiFormatAdapter(AdminApiAdapter):
         )
         # 过滤掉 pending/streaming 状态的请求
         query = query.filter(Usage.status.notin_(["pending", "streaming"]))
-        # 过滤掉 unknown/pending provider
-        query = query.filter(Usage.provider.notin_(["unknown", "pending"]))
+        # 过滤掉 unknown/pending provider_name
+        query = query.filter(Usage.provider_name.notin_(["unknown", "pending"]))
         # 只统计有 api_format 的记录
         query = query.filter(Usage.api_format.isnot(None))
 
@@ -659,8 +766,8 @@ class AdminUsageRecordsAdapter(AdminApiAdapter):
                 float(usage.rate_multiplier) if usage.rate_multiplier is not None else 1.0
             )
 
-            # 提供商名称优先级：关联的 Provider 表 > usage.provider 字段
-            provider_name = usage.provider
+            # 提供商名称优先级：关联的 Provider 表 > usage.provider_name 字段
+            provider_name = usage.provider_name
             if usage.provider_id and str(usage.provider_id) in provider_map:
                 provider_name = provider_map[str(usage.provider_id)]
 
@@ -775,7 +882,7 @@ class AdminUsageDetailAdapter(AdminApiAdapter):
                 "name": api_key.name if api_key else None,
                 "display": api_key.get_display_key() if api_key else None,
             },
-            "provider": usage_record.provider,
+            "provider": usage_record.provider_name,
             "api_format": usage_record.api_format,
             "model": usage_record.model,
             "target_model": usage_record.target_model,
@@ -810,6 +917,7 @@ class AdminUsageDetailAdapter(AdminApiAdapter):
             "request_body": usage_record.get_request_body(),
             "provider_request_headers": usage_record.provider_request_headers,
             "response_headers": usage_record.response_headers,
+            "client_response_headers": usage_record.client_response_headers,
             "response_body": usage_record.get_response_body(),
             "metadata": usage_record.request_metadata,
             "tiered_pricing": tiered_pricing_info,
@@ -828,7 +936,7 @@ class AdminUsageDetailAdapter(AdminApiAdapter):
         # 尝试获取模型的阶梯配置（带来源信息）
         cost_service = ModelCostService(db)
         pricing_result = await cost_service.get_tiered_pricing_with_source_async(
-            usage_record.provider, usage_record.model
+            usage_record.provider_name, usage_record.model
         )
 
         if not pricing_result:

src/api/admin/users/routes.py

@@ -26,6 +26,18 @@ pipeline = ApiRequestPipeline()
 # 管理员端点
 @router.post("")
 async def create_user_endpoint(request: Request, db: Session = Depends(get_db)):
+    """
+    创建用户
+
+    创建新用户账号（管理员专用）。
+
+    **请求体**:
+    - `email`: 邮箱地址
+    - `username`: 用户名
+    - `password`: 密码
+    - `role`: 角色（user/admin）
+    - `quota_usd`: 配额（USD）
+    """
     adapter = AdminCreateUserAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -33,18 +45,33 @@ async def create_user_endpoint(request: Request, db: Session = Depends(get_db)):
 @router.get("")
 async def list_users(
     request: Request,
-    skip: int = Query(0, ge=0),
-    limit: int = Query(100, ge=1, le=1000),
-    role: Optional[str] = None,
-    is_active: Optional[bool] = None,
+    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"),
+    role: Optional[str] = Query(None, description="按角色筛选（user/admin）"),
+    is_active: Optional[bool] = Query(None, description="按状态筛选"),
     db: Session = Depends(get_db),
 ):
+    """
+    获取用户列表
+
+    分页获取用户列表，支持按角色和状态筛选。
+
+    **返回字段**: id, email, username, role, quota_usd, used_usd, is_active, created_at 等
+    """
     adapter = AdminListUsersAdapter(skip=skip, limit=limit, role=role, is_active=is_active)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
 @router.get("/{user_id}")
-async def get_user(user_id: str, request: Request, db: Session = Depends(get_db)):  # UUID
+async def get_user(user_id: str, request: Request, db: Session = Depends(get_db)):
+    """
+    获取用户详情
+
+    获取指定用户的详细信息。
+
+    **路径参数**:
+    - `user_id`: 用户 ID (UUID)
+    """
     adapter = AdminGetUserAdapter(user_id=user_id)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -55,19 +82,51 @@ async def update_user(
     request: Request,
     db: Session = Depends(get_db),
 ):
+    """
+    更新用户信息
+
+    更新指定用户的信息，包括角色、配额、权限等。
+
+    **路径参数**:
+    - `user_id`: 用户 ID (UUID)
+
+    **请求体** (均为可选):
+    - `email`: 邮箱地址
+    - `username`: 用户名
+    - `role`: 角色
+    - `quota_usd`: 配额
+    - `is_active`: 是否启用
+    - `allowed_providers`: 允许的提供商列表
+    - `allowed_models`: 允许的模型列表
+    """
     adapter = AdminUpdateUserAdapter(user_id=user_id)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
 @router.delete("/{user_id}")
-async def delete_user(user_id: str, request: Request, db: Session = Depends(get_db)):  # UUID
+async def delete_user(user_id: str, request: Request, db: Session = Depends(get_db)):
+    """
+    删除用户
+
+    永久删除指定用户。不能删除最后一个管理员账户。
+
+    **路径参数**:
+    - `user_id`: 用户 ID (UUID)
+    """
     adapter = AdminDeleteUserAdapter(user_id=user_id)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
 @router.patch("/{user_id}/quota")
 async def reset_user_quota(user_id: str, request: Request, db: Session = Depends(get_db)):
-    """Reset user quota (set used_usd to 0)"""
+    """
+    重置用户配额
+
+    将用户的已用配额（used_usd）重置为 0。
+
+    **路径参数**:
+    - `user_id`: 用户 ID (UUID)
+    """
     adapter = AdminResetUserQuotaAdapter(user_id=user_id)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -76,10 +135,17 @@ async def reset_user_quota(user_id: str, request: Request, db: Session = Depends
 async def get_user_api_keys(
     user_id: str,
     request: Request,
-    is_active: Optional[bool] = None,
+    is_active: Optional[bool] = Query(None, description="按状态筛选"),
     db: Session = Depends(get_db),
 ):
-    """获取用户的所有API Keys（不包括独立Keys）"""
+    """
+    获取用户的 API 密钥列表
+
+    获取指定用户的所有 API 密钥（不包括独立密钥）。
+
+    **路径参数**:
+    - `user_id`: 用户 ID (UUID)
+    """
     adapter = AdminGetUserKeysAdapter(user_id=user_id, is_active=is_active)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -90,7 +156,23 @@ async def create_user_api_key(
     request: Request,
     db: Session = Depends(get_db),
 ):
-    """为用户创建API Key"""
+    """
+    为用户创建 API 密钥
+
+    为指定用户创建新的 API 密钥。
+
+    **路径参数**:
+    - `user_id`: 用户 ID (UUID)
+
+    **请求体**:
+    - `name`: 密钥名称
+    - `allowed_providers`: 允许的提供商（可选）
+    - `allowed_models`: 允许的模型（可选）
+    - `rate_limit`: 速率限制（可选）
+    - `expire_days`: 过期天数（可选）
+
+    **返回**: 包含完整密钥值的响应（仅此一次显示）
+    """
     adapter = AdminCreateUserKeyAdapter(user_id=user_id)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -102,7 +184,15 @@ async def delete_user_api_key(
     request: Request,
     db: Session = Depends(get_db),
 ):
-    """删除用户的API Key"""
+    """
+    删除用户的 API 密钥
+
+    删除指定用户的指定 API 密钥。
+
+    **路径参数**:
+    - `user_id`: 用户 ID (UUID)
+    - `key_id`: 密钥 ID
+    """
     adapter = AdminDeleteUserKeyAdapter(user_id=user_id, key_id=key_id)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -156,7 +246,7 @@ class AdminCreateUserAdapter(AdminApiAdapter):
             "username": user.username,
             "role": user.role.value,
             "allowed_providers": user.allowed_providers,
-            "allowed_endpoints": user.allowed_endpoints,
+            "allowed_api_formats": user.allowed_api_formats,
             "allowed_models": user.allowed_models,
             "quota_usd": user.quota_usd,
             "used_usd": user.used_usd,
@@ -183,6 +273,9 @@ class AdminListUsersAdapter(AdminApiAdapter):
                 "email": u.email,
                 "username": u.username,
                 "role": u.role.value,
+                "allowed_providers": u.allowed_providers,
+                "allowed_api_formats": u.allowed_api_formats,
+                "allowed_models": u.allowed_models,
                 "quota_usd": u.quota_usd,
                 "used_usd": u.used_usd,
                 "total_usd": getattr(u, "total_usd", 0),
@@ -216,7 +309,7 @@ class AdminGetUserAdapter(AdminApiAdapter):
             "username": user.username,
             "role": user.role.value,
             "allowed_providers": user.allowed_providers,
-            "allowed_endpoints": user.allowed_endpoints,
+            "allowed_api_formats": user.allowed_api_formats,
             "allowed_models": user.allowed_models,
             "quota_usd": user.quota_usd,
             "used_usd": user.used_usd,
@@ -282,7 +375,7 @@ class AdminUpdateUserAdapter(AdminApiAdapter):
             "username": user.username,
             "role": user.role.value,
             "allowed_providers": user.allowed_providers,
-            "allowed_endpoints": user.allowed_endpoints,
+            "allowed_api_formats": user.allowed_api_formats,
             "allowed_models": user.allowed_models,
             "quota_usd": user.quota_usd,
             "used_usd": user.used_usd,

src/api/announcements/routes.py

@@ -35,7 +35,32 @@ async def list_announcements(
     offset: int = Query(0, description="偏移量"),
     db: Session = Depends(get_db),
 ):
-    """获取公告列表（包含已读状态）"""
+    """
+    获取公告列表
+
+    获取公告列表，支持分页和筛选。如果用户已登录，返回包含已读状态。
+
+    **查询参数**:
+    - `active_only`: 是否只返回有效公告，默认 true
+    - `limit`: 返回数量限制，默认 50
+    - `offset`: 分页偏移量，默认 0
+
+    **返回字段**:
+    - `items`: 公告列表，每条公告包含：
+      - `id`: 公告 ID
+      - `title`: 标题
+      - `content`: 内容
+      - `type`: 类型（info/warning/error/success）
+      - `priority`: 优先级
+      - `is_pinned`: 是否置顶
+      - `is_read`: 是否已读（仅登录用户）
+      - `author`: 作者信息
+      - `start_time`: 生效开始时间
+      - `end_time`: 生效结束时间
+      - `created_at`: 创建时间
+    - `total`: 总数
+    - `unread_count`: 未读数量（仅登录用户）
+    """
     adapter = ListAnnouncementsAdapter(active_only=active_only, limit=limit, offset=offset)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -45,7 +70,16 @@ async def get_active_announcements(
     request: Request,
     db: Session = Depends(get_db),
 ):
-    """获取当前有效的公告（首页展示）"""
+    """
+    获取当前有效的公告
+
+    获取当前时间范围内有效的公告列表，用于首页展示。
+
+    **返回字段**:
+    - `items`: 有效公告列表
+    - `total`: 有效公告总数
+    - `unread_count`: 未读数量（仅登录用户）
+    """
     adapter = GetActiveAnnouncementsAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -56,7 +90,27 @@ async def get_announcement(
     request: Request,
     db: Session = Depends(get_db),
 ):
-    """获取单个公告详情"""
+    """
+    获取单个公告详情
+
+    获取指定公告的详细信息。
+
+    **路径参数**:
+    - `announcement_id`: 公告 ID（UUID）
+
+    **返回字段**:
+    - `id`: 公告 ID
+    - `title`: 标题
+    - `content`: 内容
+    - `type`: 类型（info/warning/error/success）
+    - `priority`: 优先级
+    - `is_pinned`: 是否置顶
+    - `author`: 作者信息（id, username）
+    - `start_time`: 生效开始时间
+    - `end_time`: 生效结束时间
+    - `created_at`: 创建时间
+    - `updated_at`: 更新时间
+    """
     adapter = GetAnnouncementAdapter(announcement_id=announcement_id)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -67,7 +121,17 @@ async def mark_announcement_as_read(
     request: Request,
     db: Session = Depends(get_db),
 ):
-    """Mark announcement as read"""
+    """
+    标记公告为已读
+
+    将指定公告标记为当前用户已读。需要登录。
+
+    **路径参数**:
+    - `announcement_id`: 公告 ID（UUID）
+
+    **返回字段**:
+    - `message`: 操作结果信息
+    """
     adapter = MarkAnnouncementReadAdapter(announcement_id=announcement_id)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -80,7 +144,25 @@ async def create_announcement(
     request: Request,
     db: Session = Depends(get_db),
 ):
-    """创建公告（管理员权限）"""
+    """
+    创建公告
+
+    创建新的系统公告。需要管理员权限。
+
+    **请求体字段**:
+    - `title`: 公告标题（必填）
+    - `content`: 公告内容（必填）
+    - `type`: 公告类型（info/warning/error/success），默认 info
+    - `priority`: 优先级（0-100），默认 0
+    - `is_pinned`: 是否置顶，默认 false
+    - `start_time`: 生效开始时间（可选）
+    - `end_time`: 生效结束时间（可选）
+
+    **返回字段**:
+    - `id`: 新创建的公告 ID
+    - `title`: 公告标题
+    - `message`: 操作结果信息
+    """
     adapter = CreateAnnouncementAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -91,7 +173,27 @@ async def update_announcement(
     request: Request,
     db: Session = Depends(get_db),
 ):
-    """更新公告（管理员权限）"""
+    """
+    更新公告
+
+    更新指定公告的信息。需要管理员权限。
+
+    **路径参数**:
+    - `announcement_id`: 公告 ID（UUID）
+
+    **请求体字段（均为可选）**:
+    - `title`: 公告标题
+    - `content`: 公告内容
+    - `type`: 公告类型（info/warning/error/success）
+    - `priority`: 优先级（0-100）
+    - `is_active`: 是否启用
+    - `is_pinned`: 是否置顶
+    - `start_time`: 生效开始时间
+    - `end_time`: 生效结束时间
+
+    **返回字段**:
+    - `message`: 操作结果信息
+    """
     adapter = UpdateAnnouncementAdapter(announcement_id=announcement_id)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -102,7 +204,17 @@ async def delete_announcement(
     request: Request,
     db: Session = Depends(get_db),
 ):
-    """删除公告（管理员权限）"""
+    """
+    删除公告
+
+    删除指定的公告。需要管理员权限。
+
+    **路径参数**:
+    - `announcement_id`: 公告 ID（UUID）
+
+    **返回字段**:
+    - `message`: 操作结果信息
+    """
     adapter = DeleteAnnouncementAdapter(announcement_id=announcement_id)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -115,7 +227,14 @@ async def get_my_unread_announcement_count(
     request: Request,
     db: Session = Depends(get_db),
 ):
-    """获取我的未读公告数量"""
+    """
+    获取我的未读公告数量
+
+    获取当前用户的未读公告数量。需要登录。
+
+    **返回字段**:
+    - `unread_count`: 未读公告数量
+    """
     adapter = UnreadAnnouncementCountAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 

src/api/auth/routes.py

@@ -95,72 +95,142 @@ pipeline = ApiRequestPipeline()
 # API端点
 @router.get("/registration-settings", response_model=RegistrationSettingsResponse)
 async def registration_settings(request: Request, db: Session = Depends(get_db)):
-    """公开获取注册相关配置"""
+    """
+    获取注册相关配置
+
+    返回系统注册配置，包括是否开放注册、是否需要邮箱验证等。
+    此接口为公开接口，无需认证。
+    """
     adapter = AuthRegistrationSettingsAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
 @router.get("/settings")
 async def auth_settings(request: Request, db: Session = Depends(get_db)):
-    """公开获取认证设置（用于前端判断显示哪些登录选项）"""
+    """
+    获取认证设置
+
+    返回系统支持的认证方式，如本地认证、LDAP 认证等。
+    前端据此判断显示哪些登录选项。此接口为公开接口，无需认证。
+    """
     adapter = AuthSettingsAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
 @router.post("/login", response_model=LoginResponse)
 async def login(request: Request, db: Session = Depends(get_db)):
+    """
+    用户登录
+
+    使用邮箱和密码登录，成功后返回 JWT access_token 和 refresh_token。
+
+    - **access_token**: 用于后续 API 调用，有效期 24 小时
+    - **refresh_token**: 用于刷新 access_token
+
+    速率限制: 5次/分钟/IP
+    """
     adapter = AuthLoginAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
 @router.post("/refresh", response_model=RefreshTokenResponse)
 async def refresh_token(request: Request, db: Session = Depends(get_db)):
+    """
+    刷新访问令牌
+
+    使用 refresh_token 获取新的 access_token 和 refresh_token。
+    原 refresh_token 刷新后失效。
+    """
     adapter = AuthRefreshAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
 @router.post("/register", response_model=RegisterResponse)
 async def register(request: Request, db: Session = Depends(get_db)):
+    """
+    用户注册
+
+    创建新用户账号。需要系统开放注册功能。
+    如果系统开启了邮箱验证，需先通过 /send-verification-code 和 /verify-email 完成邮箱验证。
+
+    速率限制: 3次/分钟/IP
+    """
     adapter = AuthRegisterAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
 @router.get("/me")
 async def get_current_user_info(request: Request, db: Session = Depends(get_db)):
+    """
+    获取当前用户信息
+
+    返回当前登录用户的基本信息，包括邮箱、用户名、角色、配额等。
+    需要 Bearer Token 认证。
+    """
     adapter = AuthCurrentUserAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
 @router.patch("/password")
 async def change_password(request: Request, db: Session = Depends(get_db)):
-    """Change current user's password"""
+    """
+    修改密码
+
+    修改当前用户的登录密码，需提供旧密码验证。
+    密码长度至少 6 位。
+    """
     adapter = AuthChangePasswordAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
 @router.post("/logout", response_model=LogoutResponse)
 async def logout(request: Request, db: Session = Depends(get_db)):
+    """
+    用户登出
+
+    将当前 Token 加入黑名单，使其失效。
+    """
     adapter = AuthLogoutAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
 @router.post("/send-verification-code", response_model=SendVerificationCodeResponse)
 async def send_verification_code(request: Request, db: Session = Depends(get_db)):
-    """发送邮箱验证码"""
+    """
+    发送邮箱验证码
+
+    向指定邮箱发送验证码，用于注册前的邮箱验证。
+    验证码有效期 5 分钟，同一邮箱 60 秒内只能发送一次。
+
+    速率限制: 3次/分钟/IP
+    """
     adapter = AuthSendVerificationCodeAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
 @router.post("/verify-email", response_model=VerifyEmailResponse)
 async def verify_email(request: Request, db: Session = Depends(get_db)):
-    """验证邮箱验证码"""
+    """
+    验证邮箱验证码
+
+    验证邮箱收到的验证码是否正确。
+    验证成功后，邮箱会被标记为已验证状态，可用于注册。
+
+    速率限制: 10次/分钟/IP
+    """
     adapter = AuthVerifyEmailAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
 @router.post("/verification-status", response_model=VerificationStatusResponse)
 async def verification_status(request: Request, db: Session = Depends(get_db)):
-    """查询邮箱验证状态"""
+    """
+    查询邮箱验证状态
+
+    查询指定邮箱的验证状态，包括是否有待验证的验证码、是否已验证等。
+
+    速率限制: 20次/分钟/IP
+    """
     adapter = AuthVerificationStatusAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -458,7 +528,7 @@ class AuthCurrentUserAdapter(AuthenticatedApiAdapter):
             "used_usd": user.used_usd,
             "total_usd": user.total_usd,
             "allowed_providers": user.allowed_providers,
-            "allowed_endpoints": user.allowed_endpoints,
+            "allowed_api_formats": user.allowed_api_formats,
             "allowed_models": user.allowed_models,
             "created_at": user.created_at.isoformat(),
             "last_login_at": user.last_login_at.isoformat() if user.last_login_at else None,

src/api/base/adapter.py

@@ -15,6 +15,7 @@ class ApiMode(str, Enum):
     ADMIN = "admin"
     USER = "user"  # JWT 认证的普通用户（不要求管理员权限）
     PUBLIC = "public"
+    MANAGEMENT = "management"  # Management Token 认证
 
 
 class ApiAdapter(ABC):

src/api/base/context.py

@@ -10,7 +10,7 @@ from fastapi import HTTPException, Request
 from sqlalchemy.orm import Session
 
 from src.core.logger import logger
-from src.models.database import ApiKey, User
+from src.models.database import ApiKey, ManagementToken, User
 from src.utils.request_utils import get_client_ip
 
 
@@ -38,6 +38,9 @@ class ApiRequestContext:
     # URL 路径参数（如 Gemini API 的 /v1beta/models/{model}:generateContent）
     path_params: Dict[str, Any] = field(default_factory=dict)
 
+    # Management Token（用于管理 API 认证）
+    management_token: Optional[ManagementToken] = None
+
     # 供适配器扩展的状态存储
     extra: Dict[str, Any] = field(default_factory=dict)
     audit_metadata: Dict[str, Any] = field(default_factory=dict)

src/api/base/models_service.py

@@ -143,12 +143,13 @@ class AccessRestrictions:
                 allowed_api_formats = api_key.allowed_api_formats
 
         # 如果 API Key 没有限制，检查 User 的限制
-        # 注意: User 没有 allowed_api_formats 字段
         if user:
             if allowed_providers is None and user.allowed_providers is not None:
                 allowed_providers = user.allowed_providers
             if allowed_models is None and user.allowed_models is not None:
                 allowed_models = user.allowed_models
+            if allowed_api_formats is None and user.allowed_api_formats is not None:
+                allowed_api_formats = user.allowed_api_formats
 
         return cls(
             allowed_providers=allowed_providers,
@@ -201,20 +202,59 @@ def get_available_provider_ids(db: Session, api_formats: list[str]) -> set[str]:
     条件:
     - 端点 api_format 匹配
     - 端点是活跃的
-    - 端点下有活跃的 Key
+    - Provider 下有活跃的 Key 且支持该 api_format（Key 直属 Provider，通过 api_formats 过滤）
     """
-    rows = (
-        db.query(ProviderEndpoint.provider_id)
-        .join(ProviderAPIKey, ProviderAPIKey.endpoint_id == ProviderEndpoint.id)
+    target_formats = {f.upper() for f in api_formats}
+
+    # 1) 先找出有活跃端点的 Provider（记录每个 Provider 支持的格式集合）
+    endpoint_rows = (
+        db.query(ProviderEndpoint.provider_id, ProviderEndpoint.api_format)
         .filter(
-            ProviderEndpoint.api_format.in_(api_formats),
+            ProviderEndpoint.api_format.in_(list(target_formats)),
             ProviderEndpoint.is_active.is_(True),
-            ProviderAPIKey.is_active.is_(True),
         )
-        .distinct()
         .all()
     )
-    return {row[0] for row in rows}
+
+    if not endpoint_rows:
+        return set()
+
+    provider_to_formats: dict[str, set[str]] = {}
+    for provider_id, fmt in endpoint_rows:
+        if not provider_id or not fmt:
+            continue
+        provider_to_formats.setdefault(provider_id, set()).add(str(fmt).upper())
+
+    provider_ids_with_endpoints = set(provider_to_formats.keys())
+    if not provider_ids_with_endpoints:
+        return set()
+
+    # 2) 再检查这些 Provider 是否至少有一个活跃 Key 支持对应格式
+    key_rows = (
+        db.query(ProviderAPIKey.provider_id, ProviderAPIKey.api_formats)
+        .filter(
+            ProviderAPIKey.provider_id.in_(provider_ids_with_endpoints),
+            ProviderAPIKey.is_active.is_(True),
+        )
+        .all()
+    )
+
+    available_provider_ids: set[str] = set()
+    for provider_id, key_formats in key_rows:
+        if not provider_id:
+            continue
+        endpoint_formats = provider_to_formats.get(provider_id)
+        if not endpoint_formats:
+            continue
+
+        formats_list = key_formats if isinstance(key_formats, list) else []
+        key_formats_upper = {str(f).upper() for f in formats_list}
+
+        # 只有同时满足：请求格式 ∩ Provider 端点格式 ∩ Key 支持格式 非空，才算可用
+        if key_formats_upper & endpoint_formats & target_formats:
+            available_provider_ids.add(provider_id)
+
+    return available_provider_ids
 
 
 def _get_available_model_ids_for_format(db: Session, api_formats: list[str]) -> set[str]:
@@ -227,36 +267,64 @@ def _get_available_model_ids_for_format(db: Session, api_formats: list[str]) ->
     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)
+    target_formats = {f.upper() for f in api_formats}
+
+    # 1) 找出有活跃端点的 Provider（记录每个 Provider 支持的格式集合）
+    endpoint_rows = (
+        db.query(ProviderEndpoint.provider_id, ProviderEndpoint.api_format)
         .filter(
-            ProviderEndpoint.api_format.in_(api_formats),
+            ProviderEndpoint.api_format.in_(list(target_formats)),
             ProviderEndpoint.is_active.is_(True),
+        )
+        .all()
+    )
+
+    if not endpoint_rows:
+        return set()
+
+    provider_to_formats: dict[str, set[str]] = {}
+    for provider_id, fmt in endpoint_rows:
+        if not provider_id or not fmt:
+            continue
+        provider_to_formats.setdefault(provider_id, set()).add(str(fmt).upper())
+
+    provider_ids_with_endpoints = set(provider_to_formats.keys())
+    if not provider_ids_with_endpoints:
+        return set()
+
+    # 2) 收集每个 Provider 下「支持对应格式」的活跃 Key 的 allowed_models
+    # Key 直属 Provider，通过 key.api_formats 与 Provider 端点格式交集筛选
+    key_rows = (
+        db.query(ProviderAPIKey.provider_id, ProviderAPIKey.allowed_models, ProviderAPIKey.api_formats)
+        .filter(
+            ProviderAPIKey.provider_id.in_(provider_ids_with_endpoints),
             ProviderAPIKey.is_active.is_(True),
         )
         .all()
     )
 
-    if not endpoint_keys:
+    # provider_id -> list[(allowed_models, usable_formats)]
+    provider_key_rules: dict[str, list[tuple[object, set[str]]]] = {}
+    for provider_id, allowed_models, key_formats in key_rows:
+        if not provider_id:
+            continue
+
+        endpoint_formats = provider_to_formats.get(provider_id)
+        if not endpoint_formats:
+            continue
+
+        formats_list = key_formats if isinstance(key_formats, list) else []
+        key_formats_upper = {str(f).upper() for f in formats_list}
+        usable_formats = key_formats_upper & endpoint_formats & target_formats
+        if not usable_formats:
+            continue
+
+        provider_key_rules.setdefault(provider_id, []).append((allowed_models, usable_formats))
+
+    provider_ids_with_format = set(provider_key_rules.keys())
+    if not provider_ids_with_format:
         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)
@@ -284,22 +352,30 @@ def _get_available_model_ids_for_format(db: Session, api_formats: list[str]) ->
         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:
+        # 检查该 provider 下是否有 Key 允许这个模型（支持 list/dict 两种 allowed_models）
+        from src.core.model_permissions import check_model_allowed
+
+        rules = provider_key_rules.get(model_provider_id, [])
+        for allowed_models, usable_formats in rules:
+            # None = 不限制
             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
 
+            # 对于支持多个格式的 Key：任意一个可用格式允许即可
+            for fmt in usable_formats:
+                if check_model_allowed(
+                    model_name=model_id,
+                    allowed_models=allowed_models,  # type: ignore[arg-type]
+                    api_format=fmt,
+                    resolved_model_name=(model.provider_model_name if global_model else None),
+                ):
+                    available_model_ids.add(model_id)
+                    break
+            else:
+                continue
+            break
+
     return available_model_ids
 
 

src/api/base/pipeline.py

@@ -2,19 +2,23 @@ from __future__ import annotations
 
 import time
 from enum import Enum
-from typing import Any, Optional, Tuple
+from typing import TYPE_CHECKING, Any, Optional, Tuple
 
 from fastapi import HTTPException, Request
 from sqlalchemy.orm import Session
 
 from src.config.settings import config
+from src.core.enums import UserRole
 from src.core.exceptions import QuotaExceededException
 from src.core.logger import logger
-from src.models.database import ApiKey, AuditEventType, User, UserRole
+from src.models.database import ApiKey, AuditEventType, User
 from src.services.auth.service import AuthService
 from src.services.system.audit import AuditService
 from src.services.usage.service import UsageService
 
+if TYPE_CHECKING:
+    from src.models.database import ManagementToken
+
 from .adapter import ApiAdapter, ApiMode
 from .context import ApiRequestContext
 
@@ -47,17 +51,22 @@ class ApiRequestPipeline:
         logger.debug(f"[Pipeline] Running with mode={mode}, adapter={adapter.__class__.__name__}, "
             f"adapter.mode={adapter.mode}, path={http_request.url.path}")
         if mode == ApiMode.ADMIN:
-            user = await self._authenticate_admin(http_request, db)
+            user, management_token = await self._authenticate_admin(http_request, db)
             api_key = None
         elif mode == ApiMode.USER:
-            user = await self._authenticate_user(http_request, db)
+            user, management_token = await self._authenticate_user(http_request, db)
             api_key = None
         elif mode == ApiMode.PUBLIC:
             user = None
             api_key = None
+            management_token = None
+        elif mode == ApiMode.MANAGEMENT:
+            user, management_token = await self._authenticate_management(http_request, db)
+            api_key = None
         else:
             logger.debug("[Pipeline] 调用 _authenticate_client")
             user, api_key = self._authenticate_client(http_request, db, adapter)
+            management_token = None
             logger.debug(f"[Pipeline] 认证完成 | user={user.username if user else None}")
 
         raw_body = None
@@ -90,6 +99,9 @@ class ApiRequestPipeline:
             api_format_hint=api_format_hint,
             path_params=path_params,
         )
+        # 存储 management_token 到 context（用于权限检查）
+        if management_token:
+            context.management_token = management_token
         logger.debug(f"[Pipeline] Context构建完成 | adapter={adapter.name} | request_id={context.request_id}")
 
         if mode != ApiMode.ADMIN and user:
@@ -177,12 +189,41 @@ class ApiRequestPipeline:
 
         return user, api_key
 
-    async def _authenticate_admin(self, request: Request, db: Session) -> User:
+    async def _authenticate_admin(
+        self, request: Request, db: Session
+    ) -> Tuple[User, Optional["ManagementToken"]]:
+        """管理员认证，支持 JWT 和 Management Token 两种方式"""
+        from src.models.database import ManagementToken
+        from src.utils.request_utils import get_client_ip
+
         authorization = request.headers.get("authorization")
         if not authorization or not authorization.lower().startswith("bearer "):
             raise HTTPException(status_code=401, detail="缺少管理员凭证")
 
         token = authorization[7:].strip()
+
+        # 检查是否为 Management Token（ae_ 前缀）
+        if token.startswith(ManagementToken.TOKEN_PREFIX):
+            client_ip = get_client_ip(request)
+            result = await self.auth_service.authenticate_management_token(db, token, client_ip)
+
+            if not result:
+                raise HTTPException(status_code=401, detail="无效或过期的 Management Token")
+
+            user, management_token = result
+
+            # 检查管理员权限
+            if user.role != UserRole.ADMIN:
+                logger.warning(f"非管理员尝试通过 Management Token 访问管理端点: {user.email}")
+                raise HTTPException(status_code=403, detail="需要管理员权限")
+
+            # 存储到 request.state
+            request.state.user_id = user.id
+            request.state.management_token_id = management_token.id
+
+            return user, management_token
+
+        # JWT 认证
         try:
             payload = await self.auth_service.verify_token(token, token_type="access")
         except HTTPException:
@@ -200,16 +241,43 @@ class ApiRequestPipeline:
         if not user or not user.is_active:
             raise HTTPException(status_code=403, detail="用户不存在或已禁用")
 
-        request.state.user_id = user.id
-        return user
+        # 检查管理员权限
+        if user.role != UserRole.ADMIN:
+            logger.warning(f"非管理员尝试通过 JWT 访问管理端点: {user.email}")
+            raise HTTPException(status_code=403, detail="需要管理员权限")
+
+        request.state.user_id = user.id
+        return user, None
+
+    async def _authenticate_user(
+        self, request: Request, db: Session
+    ) -> Tuple[User, Optional["ManagementToken"]]:
+        """用户认证，支持 JWT 和 Management Token 两种方式"""
+        from src.models.database import ManagementToken
+        from src.utils.request_utils import get_client_ip
 
-    async def _authenticate_user(self, request: Request, db: Session) -> User:
-        """JWT 认证普通用户（不要求管理员权限）"""
         authorization = request.headers.get("authorization")
         if not authorization or not authorization.lower().startswith("bearer "):
             raise HTTPException(status_code=401, detail="缺少用户凭证")
 
         token = authorization[7:].strip()
+
+        # 检查是否为 Management Token（ae_ 前缀）
+        if token.startswith(ManagementToken.TOKEN_PREFIX):
+            client_ip = get_client_ip(request)
+            result = await self.auth_service.authenticate_management_token(db, token, client_ip)
+
+            if not result:
+                raise HTTPException(status_code=401, detail="无效或过期的 Management Token")
+
+            user, management_token = result
+
+            request.state.user_id = user.id
+            request.state.management_token_id = management_token.id
+
+            return user, management_token
+
+        # JWT 认证
         try:
             payload = await self.auth_service.verify_token(token, token_type="access")
         except HTTPException:
@@ -222,13 +290,47 @@ class ApiRequestPipeline:
         if not user_id:
             raise HTTPException(status_code=401, detail="无效的用户令牌")
 
-        # 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")
 
         request.state.user_id = user.id
-        return user
+        return user, None
+
+    async def _authenticate_management(
+        self, request: Request, db: Session
+    ) -> Tuple[User, "ManagementToken"]:
+        """Management Token 认证"""
+        from src.models.database import ManagementToken
+        from src.utils.request_utils import get_client_ip
+
+        authorization = request.headers.get("authorization")
+        if not authorization or not authorization.lower().startswith("bearer "):
+            raise HTTPException(status_code=401, detail="缺少 Management Token")
+
+        token = authorization[7:].strip()
+
+        # 检查是否为 Management Token 格式
+        if not token.startswith(ManagementToken.TOKEN_PREFIX):
+            raise HTTPException(
+                status_code=401,
+                detail=f"无效的 Token 格式，需要 Management Token ({ManagementToken.TOKEN_PREFIX}xxx)",
+            )
+
+        client_ip = get_client_ip(request)
+
+        result = await self.auth_service.authenticate_management_token(db, token, client_ip)
+
+        if not result:
+            raise HTTPException(status_code=401, detail="无效或过期的 Management Token")
+
+        user, management_token = result
+
+        # 存储到 request.state
+        request.state.user_id = user.id
+        request.state.management_token_id = management_token.id
+
+        return user, management_token
 
     def _calculate_quota_remaining(self, user: Optional[User]) -> Optional[float]:
         if not user:

src/api/dashboard/routes.py

@@ -45,6 +45,29 @@ def format_tokens(num: int) -> str:
 
 @router.get("/stats")
 async def get_dashboard_stats(request: Request, db: Session = Depends(get_db)):
+    """
+    获取仪表盘统计数据
+
+    根据用户角色返回不同的统计数据。管理员可以看到全局数据，普通用户只能看到自己的数据。
+
+    **返回字段（管理员）**:
+    - `stats`: 统计卡片数组，包含总请求、总费用、总Token、总缓存等信息
+    - `today`: 今日统计（requests, cost, actual_cost, tokens, cache_creation_tokens, cache_read_tokens）
+    - `api_keys`: API Key 统计（total, active）
+    - `tokens`: 本月 Token 统计
+    - `token_breakdown`: Token 详细分类（input, output, cache_creation, cache_read）
+    - `system_health`: 系统健康指标（avg_response_time, error_rate, error_requests, fallback_count, total_requests）
+    - `cost_stats`: 成本统计（total_cost, total_actual_cost, cost_savings）
+    - `cache_stats`: 缓存统计信息
+    - `users`: 用户统计（total, active）
+
+    **返回字段（普通用户）**:
+    - `stats`: 统计卡片数组，包含 API 密钥、本月请求、配额使用、总Token 等信息
+    - `today`: 今日统计
+    - `token_breakdown`: Token 详细分类
+    - `cache_stats`: 缓存统计信息
+    - `monthly_cost`: 本月费用
+    """
     adapter = DashboardStatsAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -55,6 +78,23 @@ async def get_recent_requests(
     limit: int = Query(10, ge=1, le=100),
     db: Session = Depends(get_db),
 ):
+    """
+    获取最近请求列表
+
+    获取最近的 API 请求记录。管理员可以看到所有用户的请求，普通用户只能看到自己的请求。
+
+    **查询参数**:
+    - `limit`: 返回记录数，默认 10，最大 100
+
+    **返回字段**:
+    - `requests`: 请求列表，每条记录包含：
+      - `id`: 请求 ID
+      - `user`: 用户名
+      - `model`: 使用的模型
+      - `tokens`: Token 数量
+      - `time`: 请求时间（HH:MM 格式）
+      - `is_stream`: 是否为流式请求
+    """
     adapter = DashboardRecentRequestsAdapter(limit=limit)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -65,6 +105,17 @@ async def get_recent_requests(
 
 @router.get("/provider-status")
 async def get_provider_status(request: Request, db: Session = Depends(get_db)):
+    """
+    获取提供商状态
+
+    获取所有活跃提供商的状态和最近 24 小时的请求统计。
+
+    **返回字段**:
+    - `providers`: 提供商列表，每个提供商包含：
+      - `name`: 提供商名称
+      - `status`: 状态（active/inactive）
+      - `requests`: 最近 24 小时的请求数
+    """
     adapter = DashboardProviderStatusAdapter()
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -75,6 +126,28 @@ async def get_daily_stats(
     days: int = Query(7, ge=1, le=30),
     db: Session = Depends(get_db),
 ):
+    """
+    获取每日统计数据
+
+    获取指定天数的每日使用统计数据，用于生成图表。
+
+    **查询参数**:
+    - `days`: 统计天数，默认 7 天，最大 30 天
+
+    **返回字段**:
+    - `daily_stats`: 每日统计数组，每天包含：
+      - `date`: 日期（ISO 格式）
+      - `requests`: 请求数
+      - `tokens`: Token 数量
+      - `cost`: 费用（USD）
+      - `avg_response_time`: 平均响应时间（秒）
+      - `unique_models`: 使用的模型数量（仅管理员）
+      - `unique_providers`: 使用的提供商数量（仅管理员）
+      - `fallback_count`: 故障转移次数（仅管理员）
+      - `model_breakdown`: 按模型分解的统计（仅管理员）
+    - `model_summary`: 模型使用汇总，按费用排序
+    - `period`: 统计周期信息（start_date, end_date, days）
+    """
     adapter = DashboardDailyStatsAdapter(days=days)
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
@@ -82,7 +155,7 @@ async def get_daily_stats(
 class DashboardAdapter(ApiAdapter):
     """需要登录的仪表盘适配器基类。"""
 
-    mode = ApiMode.ADMIN
+    mode = ApiMode.USER  # 普通用户也可访问仪表盘
 
     def authorize(self, context):  # type: ignore[override]
         if not context.user:
@@ -693,7 +766,7 @@ class DashboardProviderStatusAdapter(DashboardAdapter):
         for provider in providers:
             count = (
                 db.query(func.count(Usage.id))
-                .filter(and_(Usage.provider == provider.name, Usage.created_at >= since))
+                .filter(and_(Usage.provider_name == provider.name, Usage.created_at >= since))
                 .scalar()
             )
             entries.append(
@@ -781,7 +854,7 @@ class DashboardDailyStatsAdapter(DashboardAdapter):
                     .scalar() or 0
                 )
                 today_unique_providers = (
-                    db.query(func.count(func.distinct(Usage.provider)))
+                    db.query(func.count(func.distinct(Usage.provider_name)))
                     .filter(Usage.created_at >= today)
                     .scalar() or 0
                 )

src/api/handlers/base/base_handler.py

@@ -98,6 +98,7 @@ class MessageTelemetry:
         request_headers: Dict[str, Any],
         response_body: Any,
         response_headers: Dict[str, Any],
+        client_response_headers: Optional[Dict[str, Any]] = None,
         cache_creation_tokens: int = 0,
         cache_read_tokens: int = 0,
         is_stream: bool = False,
@@ -143,6 +144,7 @@ class MessageTelemetry:
             request_body=request_body,
             provider_request_headers=provider_request_headers or {},
             response_headers=response_headers,
+            client_response_headers=client_response_headers,
             response_body=response_body,
             request_id=self.request_id,
             # Provider 侧追踪信息（用于记录真实成本）
@@ -192,6 +194,8 @@ class MessageTelemetry:
         cache_creation_tokens: int = 0,
         cache_read_tokens: int = 0,
         response_body: Optional[Dict[str, Any]] = None,
+        response_headers: Optional[Dict[str, Any]] = None,
+        client_response_headers: Optional[Dict[str, Any]] = None,
         # 模型映射信息
         target_model: Optional[str] = None,
     ) -> None:
@@ -207,6 +211,8 @@ class MessageTelemetry:
             cache_creation_tokens: 缓存创建 tokens
             cache_read_tokens: 缓存读取 tokens
             response_body: 响应体（如果有部分响应）
+            response_headers: 响应头（Provider 返回的原始响应头）
+            client_response_headers: 返回给客户端的响应头
             target_model: 映射后的目标模型名（如果发生了映射）
         """
         provider_name = provider or "unknown"
@@ -232,7 +238,8 @@ class MessageTelemetry:
             request_headers=request_headers,
             request_body=request_body,
             provider_request_headers=provider_request_headers or {},
-            response_headers={},
+            response_headers=response_headers or {},
+            client_response_headers=client_response_headers,
             response_body=response_body or {"error": error_message},
             request_id=self.request_id,
             # 模型映射信息

src/api/handlers/base/chat_adapter_base.py

@@ -351,9 +351,9 @@ class ChatAdapterBase(ApiAdapter):
         # 确定错误消息
         if isinstance(e, ProviderAuthException):
             error_message = (
-                f"提供商认证失败: {str(e)}"
+                "上游服务认证失败"
                 if result.metadata.provider != "unknown"
-                else "服务端错误: 无可用提供商"
+                else "服务暂时不可用"
             )
             result.error_message = error_message