refactor: 移除独立 hub/proxy/executor/gateway crate，统一为 gateway tunnel 架构

- 删除 aether-hub、aether-proxy 独立项目及其 Dockerfile/配置 - 删除 crates/aether-executor 和 crates/aether-gateway 全部模块 - 新增 apps/ 目录作为应用入口 - 将 hub 概念重构为 gateway tunnel transport - 将 executor 重构为 execution runtime - 新增 tunnel.rs 合约定义和 testkit tunnel/execution_runtime 模块 - 更新 Python 服务层和测试适配新架构命名
2026-09-03 09:50:21 +08:00 · 2026-04-03 14:59:58 +08:00
parent ddf18fed9a
commit 8f26e1a31f
983 changed files with 103098 additions and 105837 deletions
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -36,6 +36,8 @@ version = "0.1.0"
 name = "aether-contracts"
 version = "0.1.0"
 dependencies = [
+ "bytes",
+ "flate2",
 "serde",
 "serde_json",
 "thiserror 2.0.18",
@@ -74,31 +76,6 @@ dependencies = [
 "uuid",
 ]

-[[package]]
-name = "aether-executor"
-version = "0.1.0"
-dependencies = [
- "aether-contracts",
- "aether-http",
- "aether-runtime",
- "async-stream",
- "axum",
- "base64",
- "bytes",
- "clap",
- "flate2",
- "futures-util",
- "http",
- "reqwest",
- "rustls",
- "serde",
- "serde_json",
- "thiserror 2.0.18",
- "tokio",
- "tracing",
- "webpki-roots 0.26.11",
-]
-
 [[package]]
 name = "aether-gateway"
 version = "0.1.0"
@@ -120,11 +97,13 @@ dependencies = [
 "chrono",
 "chrono-tz",
 "clap",
+ "dashmap",
 "flate2",
 "futures-util",
 "hmac",
 "http",
 "ldap3",
+ "parking_lot",
 "redis",
 "regex",
 "reqwest",
@@ -150,37 +129,17 @@ dependencies = [
 "serde",
 ]

-[[package]]
-name = "aether-hub"
-version = "0.2.0"
-dependencies = [
- "aether-http",
- "aether-runtime",
- "async-stream",
- "axum",
- "bytes",
- "clap",
- "dashmap",
- "flate2",
- "futures-util",
- "http-body-util",
- "parking_lot",
- "reqwest",
- "serde",
- "serde_json",
- "tokio",
- "tokio-tungstenite 0.28.0",
- "tracing",
-]
-
 [[package]]
 name = "aether-proxy"
 version = "0.2.5"
 dependencies = [
+ "aether-contracts",
+ "aether-gateway",
 "aether-http",
 "aether-runtime",
 "anyhow",
 "arc-swap",
+ "axum",
 "base64",
 "bytes",
 "clap",
@@ -234,10 +193,8 @@ version = "0.1.0"
 dependencies = [
 "aether-contracts",
 "aether-data",
- "aether-executor",
 "aether-gateway",
 "aether-http",
- "aether-hub",
 "aether-runtime",
 "async-stream",
 "axum",
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -1,15 +1,13 @@
 [workspace]
 members = [
-    "aether-hub",
-    "aether-proxy",
+    "apps/aether-proxy",
    "crates/aether-cache",
    "crates/aether-billing",
    "crates/aether-wallet",
    "crates/aether-crypto",
    "crates/aether-contracts",
    "crates/aether-data",
-    "crates/aether-executor",
-    "crates/aether-gateway",
+    "apps/aether-gateway",
    "crates/aether-http",
    "crates/aether-runtime",
    "crates/aether-testkit",
@@ -22,15 +20,13 @@ license = "LicenseRef-Aether-NonCommercial"
 repository = "https://github.com/fawney19/Aether.git"

 [workspace.dependencies]
-aether-hub = { path = "aether-hub" }
 aether-cache = { path = "crates/aether-cache" }
 aether-billing = { path = "crates/aether-billing" }
 aether-wallet = { path = "crates/aether-wallet" }
 aether-crypto = { path = "crates/aether-crypto" }
 aether-contracts = { path = "crates/aether-contracts" }
 aether-data = { path = "crates/aether-data" }
-aether-executor = { path = "crates/aether-executor" }
-aether-gateway = { path = "crates/aether-gateway" }
+aether-gateway = { path = "apps/aether-gateway" }
 aether-http = { path = "crates/aether-http" }
 aether-runtime = { path = "crates/aether-runtime" }
 aether-testkit = { path = "crates/aether-testkit" }
--- a/Dockerfile.app
+++ b/Dockerfile.app
@@ -9,15 +9,26 @@ WORKDIR /app
 COPY frontend/ ./frontend/
 RUN cd frontend && npm run build

+FROM rust:1.86-slim AS gateway-builder
+WORKDIR /build
+RUN --mount=type=cache,target=/var/cache/apt,sharing=locked \
+    --mount=type=cache,target=/var/lib/apt,sharing=locked \
+    apt-get update && apt-get install -y --no-install-recommends \
+    build-essential \
+    pkg-config \
+    perl
+COPY Cargo.toml Cargo.lock ./
+COPY apps/ ./apps/
+COPY crates/ ./crates/
+RUN --mount=type=cache,target=/usr/local/cargo/registry \
+    --mount=type=cache,target=/build/target \
+    cargo build --release --locked -p aether-gateway && \
+    cp target/release/aether-gateway /tmp/aether-gateway
+
 # ==================== 运行时镜像 ====================
 FROM python:3.13-slim
 WORKDIR /app

-ARG HUB_RELEASE_REPO=fawney19/Aether
-ARG HUB_TAG
-ARG TARGETARCH
-ARG GITHUB_TOKEN
-
 # 运行时依赖（无 gcc/nodejs/npm，使用 BuildKit 缓存加速）
 RUN --mount=type=cache,target=/var/cache/apt,sharing=locked \
    --mount=type=cache,target=/var/lib/apt,sharing=locked \
@@ -37,37 +48,7 @@ COPY --from=builder /usr/local/lib/python3.13/site-packages /usr/local/lib/pytho
 COPY --from=builder /usr/local/bin/gunicorn /usr/local/bin/
 COPY --from=builder /usr/local/bin/uvicorn /usr/local/bin/
 COPY --from=builder /usr/local/bin/alembic /usr/local/bin/
-# Hub 预编译二进制（构建时从 GitHub Release 下载）
-# GITHUB_TOKEN 可选：未认证 API 限流 60 次/小时，认证后 5000 次/小时
-RUN set -eux; \
-    auth_header=""; \
-    if [ -n "${GITHUB_TOKEN:-}" ]; then \
-        auth_header="Authorization: token ${GITHUB_TOKEN}"; \
-    fi; \
-    tag="${HUB_TAG:-}"; \
-    if [ -z "$tag" ]; then \
-        tag="$(curl -sL ${auth_header:+-H "$auth_header"} "https://api.github.com/repos/${HUB_RELEASE_REPO}/releases" | python3 -c "import json,sys;print(next((r['tag_name'] for r in json.load(sys.stdin) if r.get('tag_name','').startswith('hub-v') and not r.get('draft') and not r.get('prerelease')),''))")"; \
-    fi; \
-    if [ -z "$tag" ]; then \
-        echo "Failed to resolve hub release tag"; \
-        exit 1; \
-    fi; \
-    arch="${TARGETARCH:-}"; \
-    if [ -z "$arch" ]; then \
-        arch="$(dpkg --print-architecture)"; \
-    fi; \
-    case "$arch" in \
-        amd64|arm64) ;; \
-        x86_64) arch="amd64" ;; \
-        aarch64) arch="arm64" ;; \
-        *) echo "Unsupported architecture: $arch"; exit 1 ;; \
-    esac; \
-    echo "Using Hub release tag: $tag"; \
-    url="https://github.com/${HUB_RELEASE_REPO}/releases/download/${tag}/aether-hub-linux-${arch}.tar.gz"; \
-    curl -L --fail -o /tmp/aether-hub.tar.gz "$url"; \
-    tar xzf /tmp/aether-hub.tar.gz -C /usr/local/bin; \
-    chmod +x /usr/local/bin/aether-hub; \
-    rm -f /tmp/aether-hub.tar.gz
+COPY --from=gateway-builder /tmp/aether-gateway /usr/local/bin/aether-gateway
 # 从 builder 阶段复制前端构建产物
 COPY --from=builder /app/frontend/dist /usr/share/nginx/html
 RUN chmod -R 755 /usr/share/nginx/html
@@ -119,9 +100,9 @@ RUN printf '%s\n' \
    '        return 404;' \
    '    }' \
    '' \
-    '    # WebSocket 隧道端点（aether-proxy tunnel 模式）' \
+    '    # WebSocket 隧道端点（gateway-owned tunnel 模式）' \
    '    location = /api/internal/proxy-tunnel {' \
-    '        proxy_pass http://127.0.0.1:8085/proxy;' \
+    '        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 $real_ip;' \
@@ -237,14 +218,14 @@ RUN printf '%s\n' \
    'pidfile=/var/run/supervisord.pid' \
    '' \
    '[program:nginx]' \
-    'command=/bin/bash -c "sed \"s/PORT_PLACEHOLDER/8084/g\" /etc/nginx/sites-available/default.template > /etc/nginx/sites-available/default && /usr/sbin/nginx -g \"daemon off;\""' \
+    'command=/bin/bash -c "sed \"s/PORT_PLACEHOLDER/%(ENV_PORT)s/g\" /etc/nginx/sites-available/default.template > /etc/nginx/sites-available/default && /usr/sbin/nginx -g \"daemon off;\""' \
    'autostart=true' \
    'autorestart=true' \
    'stdout_logfile=/var/log/nginx/access.log' \
    'stderr_logfile=/var/log/nginx/error.log' \
    '' \
-    '[program:app]' \
-    'command=/bin/bash -c "MAX_REQUESTS_JITTER=$((${MAX_REQUESTS:-50000}/20)); exec gunicorn src.main:app -c gunicorn_conf.py --preload -w %(ENV_GUNICORN_WORKERS)s -k uvicorn.workers.UvicornWorker --bind 127.0.0.1:8084 --max-requests ${MAX_REQUESTS:-50000} --max-requests-jitter $MAX_REQUESTS_JITTER --access-logfile - --error-logfile - --log-level info"' \
+    '[program:gateway]' \
+    'command=/usr/local/bin/aether-gateway --bind 127.0.0.1:%(ENV_PORT)s --upstream http://127.0.0.1:%(ENV_LEGACY_APP_PORT)s' \
    'directory=/app' \
    'autostart=true' \
    'autorestart=true' \
@@ -252,16 +233,18 @@ RUN printf '%s\n' \
    'stdout_logfile_maxbytes=0' \
    'stderr_logfile=/dev/stderr' \
    'stderr_logfile_maxbytes=0' \
-    'environment=PYTHONUNBUFFERED=1,PYTHONIOENCODING=utf-8,LANG=C.UTF-8,LC_ALL=C.UTF-8,DOCKER_CONTAINER=true,LD_PRELOAD=/usr/local/lib/libjemalloc.so.2,MALLOC_CONF="background_thread:true,dirty_decay_ms:5000,muzzy_decay_ms:5000"' \
+    'environment=RUST_LOG=aether_gateway=info' \
    '' \
-    '[program:tunnel-hub]' \
-    'command=/usr/local/bin/aether-hub --bind 0.0.0.0:8085' \
+    '[program:app]' \
+    'command=/bin/bash -c "MAX_REQUESTS_JITTER=$((${MAX_REQUESTS:-50000}/20)); exec gunicorn src.main:app -c gunicorn_conf.py --preload -w %(ENV_GUNICORN_WORKERS)s -k uvicorn.workers.UvicornWorker --bind 127.0.0.1:%(ENV_LEGACY_APP_PORT)s --max-requests ${MAX_REQUESTS:-50000} --max-requests-jitter $MAX_REQUESTS_JITTER --access-logfile - --error-logfile - --log-level info"' \
+    'directory=/app' \
    'autostart=true' \
    'autorestart=true' \
    'stdout_logfile=/dev/stdout' \
    'stdout_logfile_maxbytes=0' \
    'stderr_logfile=/dev/stderr' \
-    'stderr_logfile_maxbytes=0' > /etc/supervisor/conf.d/supervisord.conf
+    'stderr_logfile_maxbytes=0' \
+    'environment=PYTHONUNBUFFERED=1,PYTHONIOENCODING=utf-8,LANG=C.UTF-8,LC_ALL=C.UTF-8,DOCKER_CONTAINER=true,LD_PRELOAD=/usr/local/lib/libjemalloc.so.2,MALLOC_CONF="background_thread:true,dirty_decay_ms:5000,muzzy_decay_ms:5000"' > /etc/supervisor/conf.d/supervisord.conf
 # 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）
@@ -276,6 +259,7 @@ ENV PYTHONUNBUFFERED=1 \
    LD_PRELOAD=/usr/local/lib/libjemalloc.so.2 \
    MALLOC_CONF=background_thread:true,dirty_decay_ms:5000,muzzy_decay_ms:5000 \
    PORT=8084 \
+    LEGACY_APP_PORT=18084 \
    GUNICORN_WORKERS=2 \
    MAX_REQUESTS=4000
 EXPOSE 80
--- a/Dockerfile.app.local
+++ b/Dockerfile.app.local
@@ -11,21 +11,28 @@ WORKDIR /app
 COPY frontend/ ./frontend/
 RUN cd frontend && npm run build

+FROM rust:1.86-slim AS gateway-builder
+WORKDIR /build
+RUN --mount=type=cache,target=/var/cache/apt,sharing=locked \
+    --mount=type=cache,target=/var/lib/apt,sharing=locked \
+    sed -i 's/deb.debian.org/mirrors.tuna.tsinghua.edu.cn/g' /etc/apt/sources.list.d/debian.sources && \
+    apt-get update && apt-get install -y --no-install-recommends \
+    build-essential \
+    pkg-config \
+    perl
+COPY Cargo.toml Cargo.lock ./
+COPY apps/ ./apps/
+COPY crates/ ./crates/
+RUN --mount=type=cache,target=/usr/local/cargo/registry \
+    --mount=type=cache,target=/build/target \
+    cargo build --release --locked -p aether-gateway && \
+    cp target/release/aether-gateway /tmp/aether-gateway
+
 # ==================== 运行时镜像 ====================
 FROM python:3.13-slim

 WORKDIR /app

-ARG HUB_RELEASE_REPO=fawney19/Aether
-ARG HUB_TAG
-ARG TARGETARCH
-ARG GITHUB_TOKEN
-# GitHub 下载镜像前缀，国内构建时传入可用的镜像加速地址
-# 用法: --build-arg GITHUB_MIRROR=https://ghfast.top
-#   或: --build-arg GITHUB_MIRROR=https://gh-proxy.com
-#   或: --build-arg GITHUB_MIRROR=https://mirror.ghproxy.com
-ARG GITHUB_MIRROR
-
 # 运行时依赖（使用清华镜像源 + BuildKit 缓存加速）
 RUN --mount=type=cache,target=/var/cache/apt,sharing=locked \
    --mount=type=cache,target=/var/lib/apt,sharing=locked \
@@ -48,45 +55,7 @@ COPY --from=builder /usr/local/lib/python3.13/site-packages /usr/local/lib/pytho
 COPY --from=builder /usr/local/bin/gunicorn /usr/local/bin/
 COPY --from=builder /usr/local/bin/uvicorn /usr/local/bin/
 COPY --from=builder /usr/local/bin/alembic /usr/local/bin/
-
-# Hub 预编译二进制
-# 国内构建: --build-arg GITHUB_MIRROR=https://ghfast.top 即可走镜像下载
-# GITHUB_TOKEN 可选：未认证 API 限流 60 次/小时，认证后 5000 次/小时
-RUN set -eux; \
-    arch="${TARGETARCH:-}"; \
-    if [ -z "$arch" ]; then \
-        arch="$(dpkg --print-architecture)"; \
-    fi; \
-    case "$arch" in \
-        amd64|arm64) ;; \
-        x86_64) arch="amd64" ;; \
-        aarch64) arch="arm64" ;; \
-        *) echo "Unsupported architecture: $arch"; exit 1 ;; \
-    esac; \
-    auth_header=""; \
-    if [ -n "${GITHUB_TOKEN:-}" ]; then \
-        auth_header="Authorization: token ${GITHUB_TOKEN}"; \
-    fi; \
-    tag="${HUB_TAG:-}"; \
-    if [ -z "$tag" ]; then \
-        tag="$(curl -sL ${auth_header:+-H "$auth_header"} "https://api.github.com/repos/${HUB_RELEASE_REPO}/releases" | python3 -c "import json,sys;print(next((r['tag_name'] for r in json.load(sys.stdin) if r.get('tag_name','').startswith('hub-v') and not r.get('draft') and not r.get('prerelease')),''))")"; \
-    fi; \
-    if [ -z "$tag" ]; then \
-        echo "Failed to resolve hub release tag"; \
-        exit 1; \
-    fi; \
-    echo "Using Hub release tag: $tag"; \
-    origin_url="https://github.com/${HUB_RELEASE_REPO}/releases/download/${tag}/aether-hub-linux-${arch}.tar.gz"; \
-    if [ -n "${GITHUB_MIRROR:-}" ]; then \
-        url="${GITHUB_MIRROR}/https://github.com/${HUB_RELEASE_REPO}/releases/download/${tag}/aether-hub-linux-${arch}.tar.gz"; \
-        echo "Using mirror: ${GITHUB_MIRROR}"; \
-    else \
-        url="$origin_url"; \
-    fi; \
-    curl -L --fail -o /tmp/aether-hub.tar.gz "$url"; \
-    tar xzf /tmp/aether-hub.tar.gz -C /usr/local/bin; \
-    chmod +x /usr/local/bin/aether-hub; \
-    rm -f /tmp/aether-hub.tar.gz
+COPY --from=gateway-builder /tmp/aether-gateway /usr/local/bin/aether-gateway

 # 从 builder 阶段复制前端构建产物
 COPY --from=builder /app/frontend/dist /usr/share/nginx/html
@@ -141,9 +110,9 @@ RUN printf '%s\n' \
    '        return 404;' \
    '    }' \
    '' \
-    '    # WebSocket 隧道端点（aether-proxy tunnel 模式）' \
+    '    # WebSocket 隧道端点（gateway-owned tunnel 模式）' \
    '    location = /api/internal/proxy-tunnel {' \
-    '        proxy_pass http://127.0.0.1:8085/proxy;' \
+    '        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 $real_ip;' \
@@ -260,14 +229,14 @@ RUN printf '%s\n' \
    'pidfile=/var/run/supervisord.pid' \
    '' \
    '[program:nginx]' \
-    'command=/bin/bash -c "sed \"s/PORT_PLACEHOLDER/${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/%(ENV_PORT)s/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=/bin/bash -c "MAX_REQUESTS_JITTER=$((${MAX_REQUESTS:-50000}/20)); exec gunicorn src.main:app -c gunicorn_conf.py --preload -w %(ENV_GUNICORN_WORKERS)s -k uvicorn.workers.UvicornWorker --bind 0.0.0.0:%(ENV_PORT)s --max-requests ${MAX_REQUESTS:-50000} --max-requests-jitter $MAX_REQUESTS_JITTER --access-logfile - --error-logfile - --log-level info"' \
+    '[program:gateway]' \
+    'command=/usr/local/bin/aether-gateway --bind 127.0.0.1:%(ENV_PORT)s --upstream http://127.0.0.1:%(ENV_LEGACY_APP_PORT)s' \
    'directory=/app' \
    'autostart=true' \
    'autorestart=true' \
@@ -275,16 +244,18 @@ RUN printf '%s\n' \
    'stdout_logfile_maxbytes=0' \
    'stderr_logfile=/dev/stderr' \
    'stderr_logfile_maxbytes=0' \
-    'environment=PYTHONUNBUFFERED=1,PYTHONIOENCODING=utf-8,LANG=C.UTF-8,LC_ALL=C.UTF-8,DOCKER_CONTAINER=true,LD_PRELOAD=/usr/local/lib/libjemalloc.so.2,MALLOC_CONF="background_thread:true,dirty_decay_ms:5000,muzzy_decay_ms:5000"' \
+    'environment=RUST_LOG=aether_gateway=info' \
    '' \
-    '[program:tunnel-hub]' \
-    'command=/usr/local/bin/aether-hub --bind 0.0.0.0:8085' \
+    '[program:app]' \
+    'command=/bin/bash -c "MAX_REQUESTS_JITTER=$((${MAX_REQUESTS:-50000}/20)); exec gunicorn src.main:app -c gunicorn_conf.py --preload -w %(ENV_GUNICORN_WORKERS)s -k uvicorn.workers.UvicornWorker --bind 127.0.0.1:%(ENV_LEGACY_APP_PORT)s --max-requests ${MAX_REQUESTS:-50000} --max-requests-jitter $MAX_REQUESTS_JITTER --access-logfile - --error-logfile - --log-level info"' \
+    'directory=/app' \
    'autostart=true' \
    'autorestart=true' \
    'stdout_logfile=/dev/stdout' \
    'stdout_logfile_maxbytes=0' \
    'stderr_logfile=/dev/stderr' \
-    'stderr_logfile_maxbytes=0' > /etc/supervisor/conf.d/supervisord.conf
+    'stderr_logfile_maxbytes=0' \
+    'environment=PYTHONUNBUFFERED=1,PYTHONIOENCODING=utf-8,LANG=C.UTF-8,LC_ALL=C.UTF-8,DOCKER_CONTAINER=true,LD_PRELOAD=/usr/local/lib/libjemalloc.so.2,MALLOC_CONF="background_thread:true,dirty_decay_ms:5000,muzzy_decay_ms:5000"' > /etc/supervisor/conf.d/supervisord.conf

 # 创建目录
 RUN mkdir -p /var/log/supervisor /app/logs /app/data
@@ -302,6 +273,7 @@ ENV PYTHONUNBUFFERED=1 \
    LD_PRELOAD=/usr/local/lib/libjemalloc.so.2 \
    MALLOC_CONF=background_thread:true,dirty_decay_ms:5000,muzzy_decay_ms:5000 \
    PORT=8084 \
+    LEGACY_APP_PORT=18084 \
    GUNICORN_WORKERS=2 \
    MAX_REQUESTS=4000

--- a/README.md
+++ b/README.md
@@ -87,48 +87,35 @@ cd frontend && npm install && npm run dev
 | 角色 | 本地地址 | 说明 |
 |------|----------|------|
 | Rust frontdoor | `http://localhost:8084` | `aether-gateway`，本地唯一公开入口；浏览器、CLI、SDK 都应优先连这里 |
-| Rust executor sidecar | `http://127.0.0.1:5219` | `aether-executor`，供 gateway 直连 |

 本地默认链路是：

 ```text
-client -> rust frontdoor (aether-gateway) -> aether-executor
+client -> rust frontdoor (aether-gateway) -> execution_runtime/provider transport
 ```

 其中：

- `aether-gateway` 负责公开入口、健康检查，以及当前已迁到 Rust 的 frontdoor/control/background 路径。
- `./dev.sh` 不再启动 Python 宿主；未下沉到 Rust 的 legacy 路由会直接失败，除非你手动提供 `AETHER_GATEWAY_UPSTREAM` / `AETHER_GATEWAY_CONTROL_URL`。
+- `aether-gateway` 负责公开入口、健康检查、格式转换、本地执行 runtime，以及当前已迁到 Rust 的 frontdoor/control/background 路径。
+- `./dev.sh` 不再启动 Python 宿主；未下沉到 Rust 的 legacy 路由会直接失败，除非你手动提供 `AETHER_GATEWAY_UPSTREAM`。
 - `./dev.sh` 默认把 `AETHER_GATEWAY_VIDEO_TASK_TRUTH_SOURCE_MODE` 设为 `rust-authoritative`，避免本地还依赖 Python sync report 语义。

 如果你需要把 gateway 指到一个外部 legacy 宿主，可显式设置：

 ```bash
 AETHER_GATEWAY_UPSTREAM=http://127.0.0.1:18084 \
-AETHER_GATEWAY_CONTROL_URL=http://127.0.0.1:18084 \
 ./dev.sh
 ```

 如果不提供上述变量，`./dev.sh` 会把 legacy upstream 指到一个不可用的本地地址，让未迁移路径快速失败，而不是悄悄回退到 Python。

-`./dev.sh` 会固定把 `aether-executor` 切到 TCP 模式，并自动把
-`AETHER_GATEWAY_EXECUTOR_URL` 指向 `EXECUTOR_BASE_URL`，
-这样 `aether-gateway` 可以直接命中 `aether-executor` 的窄路径直连逻辑。
-
-如果只想单独验证 Rust executor 本地链路，可以运行：
-
-```bash
-cargo build -p aether-executor
-uv run python tests/e2e_rust_executor.py
-```
-
 ## Aether Proxy (可选)

 Aether Proxy 是配套的正向代理节点，部署在海外 VPS 上，为墙内的 Aether 实例中转 API 流量。或者部署在其他服务器为指定的提供商、账号、Key使用不同的节点访问。支持 TUI 向导一键配置、systemd 服务管理、TLS 加密、DNS 缓存及连接池调优。

 - Docker Compose 部署或下载预编译二进制直接运行
 - 通过 `aether-proxy setup` 完成交互式配置，自动注册为系统服务
- 详细文档见 [aether-proxy/README.md](aether-proxy/README.md)
+- 详细文档见 [apps/aether-proxy/README.md](apps/aether-proxy/README.md)

 ## 环境变量

--- a/aether-hub/.dockerignore
+++ b/aether-hub/.dockerignore
@@ -1,3 +0,0 @@
-target/
-.git/
-.DS_Store
--- a/aether-hub/Cargo.lock
+++ b/aether-hub/Cargo.lock
--- a/aether-hub/Cargo.toml
+++ b/aether-hub/Cargo.toml
@@ -1,26 +0,0 @@
-[package]
-name = "aether-hub"
-version = "0.2.0"
-edition = "2021"
-description = "Tunnel Hub for Aether - frame router between workers and proxies"
-
-[dependencies]
-aether-http.workspace = true
-aether-runtime.workspace = true
-tokio = { version = "1", features = ["full"] }
-axum = { version = "0.8", features = ["ws"] }
-serde = { version = "1", features = ["derive"] }
-serde_json = "1"
-tracing = "0.1"
-clap = { version = "4", features = ["derive", "env"] }
-dashmap = "6"
-parking_lot = "0.12"
-flate2 = "1"
-futures-util = "0.3"
-bytes = "1"
-async-stream = "0.3"
-http-body-util = "0.1"
-reqwest.workspace = true
-
-[dev-dependencies]
-tokio-tungstenite = { version = "0.28", features = ["rustls-tls-webpki-roots"] }
--- a/aether-hub/Dockerfile
+++ b/aether-hub/Dockerfile
@@ -1,34 +0,0 @@
-# syntax=docker/dockerfile:1
-
-FROM rust:1.85-slim AS builder
-WORKDIR /build/aether-hub
-
-# 可选：配置国内 Cargo 镜像源（本地构建时传 --build-arg CARGO_MIRROR=1）
-ARG CARGO_MIRROR
-RUN if [ -n "$CARGO_MIRROR" ]; then \
-    printf '[source.crates-io]\nreplace-with = "tuna"\n\n[source.tuna]\nregistry = "sparse+https://mirrors.tuna.tsinghua.edu.cn/crates.io-index/"\n' \
-    > /usr/local/cargo/config.toml; \
-    fi
-
-# 先构建依赖层，最大化后续代码变更时的缓存命中
-COPY Cargo.toml Cargo.lock ./
-RUN mkdir src && printf 'fn main() {}\n' > src/main.rs
-RUN --mount=type=cache,target=/usr/local/cargo/registry,sharing=locked \
-    --mount=type=cache,target=/build/aether-hub/target,sharing=locked \
-    cargo build --release --locked
-RUN rm -rf src
-
-COPY src ./src
-RUN --mount=type=cache,target=/usr/local/cargo/registry,sharing=locked \
-    --mount=type=cache,target=/build/aether-hub/target,sharing=locked \
-    cargo build --release --locked && \
-    cp target/release/aether-hub /tmp/aether-hub
-
-FROM debian:bookworm-slim
-RUN apt-get update && apt-get install -y --no-install-recommends ca-certificates && \
-    rm -rf /var/lib/apt/lists/*
-COPY --from=builder /tmp/aether-hub /usr/local/bin/aether-hub
-
-EXPOSE 8085
-ENTRYPOINT ["/usr/local/bin/aether-hub"]
-CMD ["--bind", "0.0.0.0:8085"]
--- a/aether-hub/README.md
+++ b/aether-hub/README.md
@@ -1,38 +0,0 @@
-# aether-hub
-
-`aether-hub` 是 Tunnel Hub 服务，负责在 proxy 与 worker 之间路由帧。
-
-已集成在Docker镜像中, 无需单独部署。
-
-## 部署端指定 Hub 版本并构建
-
-```bash
-cd /path/to/Aether
-./deploy.sh --hub-tag hub-v0.1.0
-```
-
-不指定 `--hub-tag` 时，`./deploy.sh` 会自动解析最新 `hub-v*` release，并在构建 app 镜像时从 GitHub Release 下载对应架构的 Hub 二进制。
-
-## build.sh 模式说明
-
- 默认是 `binary` 模式（`cross` 构建二进制）。
- `--upload <hub-vX.Y.Z>` 会把构建产物上传到 GitHub Release。
- 加 `--image` 后进入镜像模式（`docker buildx`，可选）。
-
-常用参数：
-
- `--tag <tag>`: 镜像 tag
- `--image-name <name>`: 镜像名（默认 `ghcr.io/fawney19/aether-hub`）
- `--platforms <list>`: 例如 `linux/amd64,linux/arm64`
- `--push`: 推送镜像
- `--load`: 加载到本地 Docker（单平台）
- `--latest`: 额外打 `latest` tag
-
-## 运行时参数
-
- `TUNNEL_HUB_WORKER_IDLE_TIMEOUT`：worker 心跳空闲超时，默认 `60` 秒
- `TUNNEL_HUB_OUTBOUND_QUEUE_CAPACITY`：单连接出站队列容量，默认 `128`；队列打满时会把连接视为拥塞并主动关闭，避免 Hub 内存无限增长
-
-## 与部署脚本关系
-
- `./deploy.sh`: 本地构建部署（会本地构建 app/base，并在构建 app 时从 GitHub Release 下载 Hub，可用 `--hub-tag` 固定版本）。
--- a/aether-hub/build.sh
+++ b/aether-hub/build.sh
@@ -1,272 +0,0 @@
-#!/bin/bash
-# aether-hub 构建脚本
-#
-# 支持两种模式:
-# 1) binary 模式（默认）: 构建多架构二进制并可上传 GitHub Release
-# 2) image 模式: 构建并推送/加载 Docker 镜像（推荐生产发布用）
-#
-# 示例:
-#   # binary 模式（兼容旧行为）
-#   ./build.sh
-#   ./build.sh amd64
-#   ./build.sh --upload hub-v0.1.0
-#
-#   # image 模式（多架构推送）
-#   ./build.sh --image --tag v0.2.5 --push --latest
-#   ./build.sh --image --tag sha-abc123 --image-name ghcr.io/fawney19/aether-hub --push
-#   ./build.sh --image --tag local-test --platforms linux/amd64 --load
-
-set -euo pipefail
-
-SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
-PROJECT_DIR="$(cd "$SCRIPT_DIR/.." && pwd)"
-DIST_DIR="$SCRIPT_DIR/dist"
-
-# -------------------------------
-# Defaults
-# -------------------------------
-MODE="binary"                  # binary | image
-
-# binary mode options
-UPLOAD=false
-UPLOAD_TAG=""
-BINARY_TARGETS=""
-
-# image mode options
-IMAGE_NAME="${IMAGE_NAME:-ghcr.io/fawney19/aether-hub}"
-IMAGE_TAG=""
-IMAGE_PLATFORMS="linux/amd64,linux/arm64"
-IMAGE_PUSH=false
-IMAGE_LOAD=false
-IMAGE_LATEST=false
-
-usage() {
-    cat <<'EOF'
-用法:
-  ./build.sh [binary-args]
-  ./build.sh --image [image-args]
-
-binary 模式（默认）:
-  amd64|arm64              仅构建指定架构（可重复）
-  --upload <hub-vX.Y.Z>    上传到 GitHub Release（需要 gh CLI）
-
-image 模式:
-  --image                  启用镜像模式
-  --tag <tag>              镜像 tag（默认自动从 git describe 推导）
-  --image-name <name>      镜像名（默认 ghcr.io/fawney19/aether-hub）
-  --platforms <list>       平台列表，逗号分隔（默认 linux/amd64,linux/arm64）
-  --push                   推送镜像到仓库
-  --load                   加载到本地 Docker（仅单平台）
-  --latest                 额外打 latest tag
-
-通用:
-  -h, --help               显示帮助
-EOF
-}
-
-while [ $# -gt 0 ]; do
-    case "$1" in
-        --image)
-            MODE="image"
-            shift
-            ;;
-        --tag)
-            IMAGE_TAG="${2:-}"
-            shift 2
-            ;;
-        --image-name)
-            IMAGE_NAME="${2:-}"
-            shift 2
-            ;;
-        --platforms)
-            IMAGE_PLATFORMS="${2:-}"
-            shift 2
-            ;;
-        --push)
-            IMAGE_PUSH=true
-            shift
-            ;;
-        --load)
-            IMAGE_LOAD=true
-            shift
-            ;;
-        --latest)
-            IMAGE_LATEST=true
-            shift
-            ;;
-        --upload)
-            UPLOAD=true
-            UPLOAD_TAG="${2:-}"
-            shift 2
-            ;;
-        amd64|arm64)
-            BINARY_TARGETS="$BINARY_TARGETS $1"
-            shift
-            ;;
-        -h|--help)
-            usage
-            exit 0
-            ;;
-        *)
-            echo "❌ 未知参数: $1"
-            usage
-            exit 1
-            ;;
-    esac
-done
-
-build_binary() {
-    if [ -z "$BINARY_TARGETS" ]; then
-        BINARY_TARGETS="amd64 arm64"
-    fi
-
-    if ! command -v cross >/dev/null 2>&1; then
-        echo "❌ 需要安装 cross: cargo install cross --git https://github.com/cross-rs/cross"
-        exit 1
-    fi
-
-    mkdir -p "$DIST_DIR"
-
-    echo "🔨 开始构建 aether-hub 二进制..."
-    echo "   目标平台: $BINARY_TARGETS"
-    echo ""
-
-    ARTIFACTS=""
-    for arch in $BINARY_TARGETS; do
-        case "$arch" in
-            amd64) target="x86_64-unknown-linux-gnu" ;;
-            arm64) target="aarch64-unknown-linux-gnu" ;;
-            *) echo "❌ 未知架构: $arch"; exit 1 ;;
-        esac
-
-        echo ">>> 构建 $arch ($target)..."
-        cd "$SCRIPT_DIR"
-        cross build --release --target "$target" --locked
-
-        BIN="target/$target/release/aether-hub"
-        if [ ! -f "$BIN" ]; then
-            echo "❌ 未找到二进制文件: $BIN"
-            exit 1
-        fi
-
-        ARCHIVE="$DIST_DIR/aether-hub-linux-$arch.tar.gz"
-        tar czf "$ARCHIVE" -C "target/$target/release" aether-hub
-        ARTIFACTS="$ARTIFACTS $ARCHIVE"
-
-        SIZE=$(du -h "$ARCHIVE" | cut -f1)
-        echo "✅ $arch 构建完成: $ARCHIVE ($SIZE)"
-        echo ""
-    done
-
-    cd "$DIST_DIR"
-    shasum -a 256 aether-hub-*.tar.gz > SHA256SUMS.txt
-    echo "📋 SHA256 校验和:"
-    cat SHA256SUMS.txt
-    echo ""
-
-    if [ "$UPLOAD" = true ]; then
-        if [ -z "$UPLOAD_TAG" ]; then
-            echo "❌ --upload 需要指定 tag，例如: ./build.sh --upload hub-v0.1.0"
-            exit 1
-        fi
-        if ! command -v gh >/dev/null 2>&1; then
-            echo "❌ 需要安装 GitHub CLI: brew install gh"
-            exit 1
-        fi
-
-        echo "📦 上传到 GitHub Release: $UPLOAD_TAG"
-        cd "$PROJECT_DIR"
-
-        if ! git rev-parse "$UPLOAD_TAG" >/dev/null 2>&1; then
-            git tag "$UPLOAD_TAG"
-            git push origin "$UPLOAD_TAG"
-        fi
-
-        gh release create "$UPLOAD_TAG" \
-            --title "aether-hub ${UPLOAD_TAG#hub-}" \
-            --generate-notes \
-            $ARTIFACTS \
-            "$DIST_DIR/SHA256SUMS.txt"
-
-        echo "✅ 上传完成!"
-    fi
-
-    echo "🎉 binary 模式完成!"
-}
-
-build_image() {
-    if ! command -v docker >/dev/null 2>&1; then
-        echo "❌ 未找到 docker，请先安装 Docker"
-        exit 1
-    fi
-    if ! docker buildx version >/dev/null 2>&1; then
-        echo "❌ 未找到 docker buildx，请先启用 buildx"
-        exit 1
-    fi
-
-    if [ "$IMAGE_PUSH" = true ] && [ "$IMAGE_LOAD" = true ]; then
-        echo "❌ --push 与 --load 不能同时使用"
-        exit 1
-    fi
-
-    if [ "$IMAGE_PUSH" = false ] && [ "$IMAGE_LOAD" = false ]; then
-        # image 模式默认走 push，符合发布场景
-        IMAGE_PUSH=true
-    fi
-
-    if [ -z "$IMAGE_TAG" ]; then
-        IMAGE_TAG=$(git -C "$PROJECT_DIR" describe --tags --always 2>/dev/null | sed 's/^v//')
-        if [ -z "$IMAGE_TAG" ]; then
-            IMAGE_TAG=$(date +%Y%m%d%H%M%S)
-        fi
-    fi
-
-    if [ "$IMAGE_LOAD" = true ] && [[ "$IMAGE_PLATFORMS" == *,* ]]; then
-        echo "❌ --load 仅支持单平台，请用 --platforms linux/amd64（或 arm64）"
-        exit 1
-    fi
-
-    local ref="${IMAGE_NAME}:${IMAGE_TAG}"
-    local cmd=(docker buildx build
-        --platform "$IMAGE_PLATFORMS"
-        -f "$SCRIPT_DIR/Dockerfile"
-        -t "$ref"
-    )
-
-    if [ "$IMAGE_LATEST" = true ]; then
-        cmd+=(-t "${IMAGE_NAME}:latest")
-    fi
-
-    if [ "$IMAGE_PUSH" = true ]; then
-        cmd+=(--push)
-    else
-        cmd+=(--load)
-    fi
-
-    cmd+=("$SCRIPT_DIR")
-
-    echo "🔨 开始构建 aether-hub 镜像..."
-    echo "   image: $ref"
-    echo "   platforms: $IMAGE_PLATFORMS"
-    echo "   mode: $([ "$IMAGE_PUSH" = true ] && echo push || echo load)"
-    echo ""
-
-    "${cmd[@]}"
-
-    if [ "$IMAGE_PUSH" = true ]; then
-        echo "✅ 镜像已推送: $ref"
-        if [ "$IMAGE_LATEST" = true ]; then
-            echo "✅ 镜像已推送: ${IMAGE_NAME}:latest"
-        fi
-    else
-        echo "✅ 镜像已加载到本地: $ref"
-    fi
-
-    echo "🎉 image 模式完成!"
-}
-
-if [ "$MODE" = "image" ]; then
-    build_image
-else
-    build_binary
-fi
--- a/aether-hub/src/control_plane.rs
+++ b/aether-hub/src/control_plane.rs
@@ -1,88 +0,0 @@
-use aether_http::{build_http_client, HttpClientConfig};
-use reqwest::Client;
-
-#[derive(Clone)]
-pub struct ControlPlaneClient {
-    client: Option<Client>,
-    base_url: String,
-}
-
-impl ControlPlaneClient {
-    pub fn new(base_url: String) -> Self {
-        let client = build_http_client(&HttpClientConfig {
-            request_timeout_ms: Some(10_000),
-            user_agent: Some("aether-hub/control-plane".to_string()),
-            ..HttpClientConfig::default()
-        })
-        .ok();
-        Self { client, base_url }
-    }
-
-    pub fn disabled() -> Self {
-        Self {
-            client: None,
-            base_url: String::new(),
-        }
-    }
-
-    pub async fn heartbeat_ack(&self, payload: &[u8]) -> Result<Vec<u8>, String> {
-        let Some(client) = &self.client else {
-            return Ok(b"{}".to_vec());
-        };
-        let url = format!(
-            "{}/api/internal/hub/heartbeat",
-            self.base_url.trim_end_matches('/')
-        );
-        let response = client
-            .post(&url)
-            .header("content-type", "application/json")
-            .body(payload.to_vec())
-            .send()
-            .await
-            .map_err(|e| format!("heartbeat callback request failed: {e}"))?;
-        if !response.status().is_success() {
-            return Err(format!(
-                "heartbeat callback failed with status {}",
-                response.status()
-            ));
-        }
-        response
-            .bytes()
-            .await
-            .map(|bytes| bytes.to_vec())
-            .map_err(|e| format!("heartbeat callback body read failed: {e}"))
-    }
-
-    pub async fn push_node_status(
-        &self,
-        node_id: &str,
-        connected: bool,
-        conn_count: usize,
-    ) -> Result<(), String> {
-        let Some(client) = &self.client else {
-            return Ok(());
-        };
-        let url = format!(
-            "{}/api/internal/hub/node-status",
-            self.base_url.trim_end_matches('/')
-        );
-        let response = client
-            .post(&url)
-            .json(&serde_json::json!({
-                "node_id": node_id,
-                "connected": connected,
-                "conn_count": conn_count,
-            }))
-            .send()
-            .await
-            .map_err(|e| format!("node-status callback request failed: {e}"))?;
-        if response.status().is_success() {
-            Ok(())
-        } else {
-            Err(format!(
-                "node-status callback failed with status {}",
-                response.status()
-            ))
-        }
-    }
-}
--- a/aether-hub/src/main.rs
+++ b/aether-hub/src/main.rs
@@ -1,250 +0,0 @@
-use std::net::SocketAddr;
-use std::time::Duration;
-
-use aether_hub::{build_router_with_state, AppState, ConnConfig, ControlPlaneClient};
-use aether_runtime::{
-    init_service_runtime, DistributedConcurrencyGate, RedisDistributedConcurrencyConfig,
-    ServiceRuntimeConfig,
-};
-use clap::Parser;
-use tracing::info;
-
-#[derive(Parser, Debug)]
-#[command(name = "aether-hub", about = "Tunnel Hub for Aether")]
-struct Args {
-    #[arg(long, default_value = "0.0.0.0:8085", env = "TUNNEL_HUB_BIND")]
-    bind: String,
-
-    #[arg(long, default_value_t = 0, env = "TUNNEL_HUB_PROXY_IDLE_TIMEOUT")]
-    proxy_idle_timeout: u64,
-
-    #[arg(long, default_value_t = 15, env = "TUNNEL_HUB_PING_INTERVAL")]
-    ping_interval: u64,
-
-    #[arg(long, default_value_t = 2048, env = "TUNNEL_HUB_MAX_STREAMS")]
-    max_streams: usize,
-
-    #[arg(
-        long,
-        default_value_t = 128,
-        env = "TUNNEL_HUB_OUTBOUND_QUEUE_CAPACITY"
-    )]
-    outbound_queue_capacity: usize,
-
-    #[arg(
-        long,
-        default_value = "http://127.0.0.1:8084",
-        env = "TUNNEL_HUB_APP_BASE_URL"
-    )]
-    app_base_url: String,
-
-    #[arg(long, env = "TUNNEL_HUB_MAX_IN_FLIGHT_REQUESTS")]
-    max_in_flight_requests: Option<usize>,
-
-    #[arg(long, env = "TUNNEL_HUB_DISTRIBUTED_REQUEST_LIMIT")]
-    distributed_request_limit: Option<usize>,
-
-    #[arg(long, env = "TUNNEL_HUB_DISTRIBUTED_REQUEST_REDIS_URL")]
-    distributed_request_redis_url: Option<String>,
-
-    #[arg(long, env = "TUNNEL_HUB_DISTRIBUTED_REQUEST_REDIS_KEY_PREFIX")]
-    distributed_request_redis_key_prefix: Option<String>,
-
-    #[arg(
-        long,
-        env = "TUNNEL_HUB_DISTRIBUTED_REQUEST_LEASE_TTL_MS",
-        default_value_t = 30_000
-    )]
-    distributed_request_lease_ttl_ms: u64,
-
-    #[arg(
-        long,
-        env = "TUNNEL_HUB_DISTRIBUTED_REQUEST_RENEW_INTERVAL_MS",
-        default_value_t = 10_000
-    )]
-    distributed_request_renew_interval_ms: u64,
-
-    #[arg(
-        long,
-        env = "TUNNEL_HUB_DISTRIBUTED_REQUEST_COMMAND_TIMEOUT_MS",
-        default_value_t = 1_000
-    )]
-    distributed_request_command_timeout_ms: u64,
-}
-
-#[tokio::main]
-async fn main() -> Result<(), Box<dyn std::error::Error>> {
-    init_service_runtime(ServiceRuntimeConfig::new("aether-hub", "aether_hub=info"))?;
-
-    let args = Args::parse();
-    let outbound_queue_capacity = args.outbound_queue_capacity.clamp(8, 4096);
-    let ping_interval = Duration::from_secs(args.ping_interval);
-    let mut state = AppState::new(
-        ControlPlaneClient::new(args.app_base_url),
-        ConnConfig {
-            ping_interval,
-            idle_timeout: Duration::from_secs(args.proxy_idle_timeout),
-            outbound_queue_capacity,
-        },
-        args.max_streams,
-    )
-    .with_request_concurrency_limit(args.max_in_flight_requests);
-
-    if let Some(limit) = args.distributed_request_limit.filter(|limit| *limit > 0) {
-        let redis_url = args
-            .distributed_request_redis_url
-            .as_deref()
-            .map(str::trim)
-            .filter(|value| !value.is_empty())
-            .ok_or_else(|| {
-                std::io::Error::new(
-                    std::io::ErrorKind::InvalidInput,
-                    "TUNNEL_HUB_DISTRIBUTED_REQUEST_REDIS_URL is required when distributed request limit is enabled",
-                )
-            })?;
-        state = state.with_distributed_request_gate(DistributedConcurrencyGate::new_redis(
-            "hub_requests_distributed",
-            limit,
-            RedisDistributedConcurrencyConfig {
-                url: redis_url.to_string(),
-                key_prefix: args
-                    .distributed_request_redis_key_prefix
-                    .as_deref()
-                    .map(str::trim)
-                    .filter(|value| !value.is_empty())
-                    .map(ToOwned::to_owned),
-                lease_ttl_ms: args.distributed_request_lease_ttl_ms.max(1),
-                renew_interval_ms: args.distributed_request_renew_interval_ms.max(1),
-                command_timeout_ms: Some(args.distributed_request_command_timeout_ms.max(1)),
-            },
-        )?);
-    }
-
-    let app = build_router_with_state(state);
-    let listener = tokio::net::TcpListener::bind(&args.bind).await?;
-    info!(bind = %args.bind, "aether-hub started");
-
-    axum::serve(
-        listener,
-        app.into_make_service_with_connect_info::<SocketAddr>(),
-    )
-    .await?;
-    Ok(())
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-    use axum::http::{HeaderValue, StatusCode};
-    use tokio_tungstenite::tungstenite::client::IntoClientRequest;
-
-    async fn start_server(app: axum::Router) -> (String, tokio::task::JoinHandle<()>) {
-        let listener = tokio::net::TcpListener::bind("127.0.0.1:0")
-            .await
-            .expect("listener should bind");
-        let addr = listener.local_addr().expect("local addr should resolve");
-        let handle = tokio::spawn(async move {
-            axum::serve(listener, app).await.expect("server should run");
-        });
-        (format!("http://{addr}"), handle)
-    }
-
-    #[tokio::test]
-    async fn hub_exposes_metrics_endpoint() {
-        let app = build_router_with_state(AppState::new(
-            ControlPlaneClient::disabled(),
-            ConnConfig {
-                ping_interval: Duration::from_secs(15),
-                idle_timeout: Duration::from_secs(0),
-                outbound_queue_capacity: 128,
-            },
-            128,
-        ));
-        let (base_url, handle) = start_server(app).await;
-
-        let response = reqwest::Client::new()
-            .get(format!("{base_url}/metrics"))
-            .send()
-            .await
-            .expect("request should succeed");
-
-        assert_eq!(response.status(), axum::http::StatusCode::OK);
-        assert_eq!(
-            response
-                .headers()
-                .get(axum::http::header::CONTENT_TYPE)
-                .and_then(|value| value.to_str().ok()),
-            Some("text/plain; version=0.0.4; charset=utf-8")
-        );
-        let body = response.text().await.expect("body should read");
-        assert!(body.contains("service_up{service=\"aether-hub\"} 1"));
-        assert!(body.contains("hub_proxy_connections 0"));
-        assert!(body.contains("hub_active_streams 0"));
-
-        handle.abort();
-    }
-
-    #[tokio::test]
-    async fn hub_rejects_second_proxy_connection_with_distributed_overload() {
-        let distributed_gate =
-            DistributedConcurrencyGate::new_in_memory("hub_requests_distributed", 1);
-        let app_a = build_router_with_state(
-            AppState::new(
-                ControlPlaneClient::disabled(),
-                ConnConfig {
-                    ping_interval: Duration::from_secs(15),
-                    idle_timeout: Duration::from_secs(0),
-                    outbound_queue_capacity: 128,
-                },
-                128,
-            )
-            .with_distributed_request_gate(distributed_gate.clone()),
-        );
-        let app_b = build_router_with_state(
-            AppState::new(
-                ControlPlaneClient::disabled(),
-                ConnConfig {
-                    ping_interval: Duration::from_secs(15),
-                    idle_timeout: Duration::from_secs(0),
-                    outbound_queue_capacity: 128,
-                },
-                128,
-            )
-            .with_distributed_request_gate(distributed_gate),
-        );
-        let (base_a, handle_a) = start_server(app_a).await;
-        let (base_b, handle_b) = start_server(app_b).await;
-
-        let request_a = format!("{}/proxy", base_a.replace("http://", "ws://"))
-            .into_client_request()
-            .expect("request should build");
-        let mut request_a = request_a;
-        request_a
-            .headers_mut()
-            .insert("x-node-id", HeaderValue::from_static("node-a"));
-        let (socket, _) = tokio_tungstenite::connect_async(request_a)
-            .await
-            .expect("first websocket should connect");
-
-        let request_b = format!("{}/proxy", base_b.replace("http://", "ws://"))
-            .into_client_request()
-            .expect("request should build");
-        let mut request_b = request_b;
-        request_b
-            .headers_mut()
-            .insert("x-node-id", HeaderValue::from_static("node-b"));
-        let error = tokio_tungstenite::connect_async(request_b)
-            .await
-            .expect_err("second websocket should be rejected");
-        match error {
-            tokio_tungstenite::tungstenite::Error::Http(response) => {
-                assert_eq!(response.status(), StatusCode::SERVICE_UNAVAILABLE);
-            }
-            other => panic!("unexpected websocket error: {other}"),
-        }
-
-        drop(socket);
-        handle_a.abort();
-        handle_b.abort();
-    }
-}
--- a/aether-hub/src/protocol.rs
+++ b/aether-hub/src/protocol.rs
@@ -1,176 +0,0 @@
-/// Tunnel binary frame protocol
-///
-/// Frame format (10-byte header + payload):
-/// | stream_id (4B) | msg_type (1B) | flags (1B) | payload_len (4B) | payload (NB) |
-use std::io::Read;
-
-use flate2::read::GzDecoder;
-use flate2::write::GzEncoder;
-use flate2::Compression;
-
-pub const HEADER_SIZE: usize = 10;
-
-// Message types
-pub const REQUEST_HEADERS: u8 = 0x01;
-pub const REQUEST_BODY: u8 = 0x02;
-pub const RESPONSE_HEADERS: u8 = 0x03;
-pub const RESPONSE_BODY: u8 = 0x04;
-pub const STREAM_END: u8 = 0x05;
-pub const STREAM_ERROR: u8 = 0x06;
-pub const PING: u8 = 0x10;
-pub const PONG: u8 = 0x11;
-pub const GOAWAY: u8 = 0x12;
-pub const HEARTBEAT_DATA: u8 = 0x13;
-pub const HEARTBEAT_ACK: u8 = 0x14;
-// Flags
-pub const FLAG_END_STREAM: u8 = 0x01;
-pub const FLAG_GZIP_COMPRESSED: u8 = 0x02;
-
-#[derive(Debug, Clone, Copy)]
-pub struct FrameHeader {
-    pub stream_id: u32,
-    pub msg_type: u8,
-    pub flags: u8,
-    pub payload_len: u32,
-}
-
-impl FrameHeader {
-    /// Parse frame header from raw bytes (must be >= HEADER_SIZE)
-    #[inline]
-    pub fn parse(data: &[u8]) -> Option<Self> {
-        if data.len() < HEADER_SIZE {
-            return None;
-        }
-        Some(Self {
-            stream_id: u32::from_be_bytes([data[0], data[1], data[2], data[3]]),
-            msg_type: data[4],
-            flags: data[5],
-            payload_len: u32::from_be_bytes([data[6], data[7], data[8], data[9]]),
-        })
-    }
-}
-
-#[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
-pub struct RequestMeta {
-    pub method: String,
-    pub url: String,
-    pub headers: std::collections::HashMap<String, String>,
-    #[serde(default = "default_timeout", deserialize_with = "deserialize_timeout")]
-    pub timeout: u64,
-}
-
-fn default_timeout() -> u64 {
-    60
-}
-
-fn deserialize_timeout<'de, D>(deserializer: D) -> Result<u64, D::Error>
-where
-    D: serde::Deserializer<'de>,
-{
-    #[derive(serde::Deserialize)]
-    #[serde(untagged)]
-    enum TimeoutValue {
-        Int(u64),
-        Float(f64),
-    }
-
-    match <TimeoutValue as serde::Deserialize>::deserialize(deserializer)? {
-        TimeoutValue::Int(v) => Ok(v),
-        TimeoutValue::Float(v) => {
-            if !v.is_finite() || v < 0.0 {
-                return Err(serde::de::Error::custom(
-                    "timeout must be a non-negative finite number",
-                ));
-            }
-            if v.fract() != 0.0 {
-                return Err(serde::de::Error::custom("timeout must be integer seconds"));
-            }
-            if v > (u64::MAX as f64) {
-                return Err(serde::de::Error::custom("timeout is too large"));
-            }
-            Ok(v as u64)
-        }
-    }
-}
-
-#[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
-pub struct ResponseMeta {
-    pub status: u16,
-    pub headers: Vec<(String, String)>,
-}
-
-pub fn encode_frame(stream_id: u32, msg_type: u8, flags: u8, payload: &[u8]) -> Vec<u8> {
-    let mut buf = Vec::with_capacity(HEADER_SIZE + payload.len());
-    buf.extend_from_slice(&stream_id.to_be_bytes());
-    buf.push(msg_type);
-    buf.push(flags);
-    buf.extend_from_slice(&(payload.len() as u32).to_be_bytes());
-    buf.extend_from_slice(payload);
-    buf
-}
-
-/// Encode a STREAM_ERROR frame for a given stream_id with an error message
-pub fn encode_stream_error(stream_id: u32, msg: &str) -> Vec<u8> {
-    encode_frame(stream_id, STREAM_ERROR, 0, msg.as_bytes())
-}
-
-/// Encode a PING frame (stream_id=0)
-pub fn encode_ping() -> Vec<u8> {
-    encode_frame(0, PING, 0, &[])
-}
-
-/// Encode a PONG frame (stream_id=0, echo payload)
-pub fn encode_pong(payload: &[u8]) -> Vec<u8> {
-    encode_frame(0, PONG, 0, payload)
-}
-
-/// Encode a GOAWAY frame (stream_id=0)
-pub fn encode_goaway() -> Vec<u8> {
-    encode_frame(0, GOAWAY, 0, &[])
-}
-
-#[inline]
-pub fn frame_payload_by_header<'a>(data: &'a [u8], header: &FrameHeader) -> Option<&'a [u8]> {
-    let payload_len = header.payload_len as usize;
-    let end = HEADER_SIZE.checked_add(payload_len)?;
-    if data.len() < end {
-        return None;
-    }
-    Some(&data[HEADER_SIZE..end])
-}
-
-pub fn decode_payload(data: &[u8], header: &FrameHeader) -> Result<Vec<u8>, String> {
-    let payload = frame_payload_by_header(data, header)
-        .ok_or_else(|| "incomplete frame payload".to_string())?;
-    if header.flags & FLAG_GZIP_COMPRESSED != 0 {
-        let mut decoder = GzDecoder::new(payload);
-        let mut decoded = Vec::new();
-        decoder
-            .read_to_end(&mut decoded)
-            .map_err(|e| format!("failed to decompress payload: {e}"))?;
-        Ok(decoded)
-    } else {
-        Ok(payload.to_vec())
-    }
-}
-
-pub fn compress_payload(payload: &[u8]) -> Result<(Vec<u8>, u8), std::io::Error> {
-    maybe_recompress_payload(payload, true)
-}
-
-fn maybe_recompress_payload(
-    payload: &[u8],
-    prefer_gzip: bool,
-) -> Result<(Vec<u8>, u8), std::io::Error> {
-    if !prefer_gzip {
-        return Ok((payload.to_vec(), 0));
-    }
-    let mut encoder = GzEncoder::new(Vec::new(), Compression::default());
-    std::io::Write::write_all(&mut encoder, payload)?;
-    let compressed = encoder.finish()?;
-    if compressed.len() < payload.len() {
-        Ok((compressed, FLAG_GZIP_COMPRESSED))
-    } else {
-        Ok((payload.to_vec(), 0))
-    }
-}
--- a/aether-proxy/src/tunnel/mod.rs
+++ b/aether-proxy/src/tunnel/mod.rs
@@ -1,236 +0,0 @@
-pub mod client;
-pub mod dispatcher;
-pub mod heartbeat;
-pub mod protocol;
-pub mod stream_handler;
-pub mod writer;
-
-use std::sync::Arc;
-use std::time::{Duration, Instant, SystemTime, UNIX_EPOCH};
-
-use tokio::sync::watch;
-use tracing::{error, info};
-
-use crate::state::{AppState, ServerContext};
-
-/// If a tunnel stays connected at least this long, treat the next disconnect
-/// as a non-failure and reset reconnect backoff.
-const STABLE_SESSION_RESET_AFTER: Duration = Duration::from_secs(30);
-/// Startup staggering step per secondary connection, used to avoid
-/// simultaneous bursts when a pool of tunnels starts together.
-const STARTUP_STAGGER_STEP_MS: u64 = 150;
-/// Upper bound for startup staggering.
-const MAX_STARTUP_STAGGER_MS: u64 = 1_500;
-/// Keep a tiny floor for repeated reconnects; first retry is still immediate.
-const MIN_RECONNECT_DELAY_MS: u64 = 50;
-/// Even under sustained failures, keep probing frequently so recovery is fast
-/// once cross-border network quality improves.
-const RECONNECT_PROBE_MAX_DELAY_MS: u64 = 3_000;
-
-/// Run the tunnel mode main loop (connect, dispatch, reconnect).
-///
-/// `conn_idx` identifies which connection in the pool this is (0-based).
-/// Only connection 0 sends heartbeats to avoid resetting shared metrics.
-pub async fn run(
-    state: &Arc<AppState>,
-    server: &Arc<ServerContext>,
-    conn_idx: usize,
-    mut shutdown: watch::Receiver<bool>,
-) {
-    info!(server = %server.server_label, conn = conn_idx, "starting tunnel");
-    let reconnect_salt = compute_connection_salt(server, conn_idx);
-
-    let startup_delay = compute_startup_stagger(conn_idx, reconnect_salt);
-    if !startup_delay.is_zero() {
-        info!(
-            server = %server.server_label,
-            conn = conn_idx,
-            delay_ms = startup_delay.as_millis(),
-            "startup stagger before first connect"
-        );
-        tokio::select! {
-            _ = tokio::time::sleep(startup_delay) => {}
-            _ = shutdown.changed() => {
-                info!(server = %server.server_label, conn = conn_idx, "shutdown requested during startup stagger");
-                return;
-            }
-        }
-    }
-
-    let mut consecutive_failures: u32 = 0;
-
-    loop {
-        let started_at = Instant::now();
-        match client::connect_and_run(state, server, conn_idx, &mut shutdown).await {
-            Ok(client::TunnelOutcome::Shutdown) => {
-                info!(server = %server.server_label, conn = conn_idx, "tunnel shut down gracefully");
-                return;
-            }
-            Ok(client::TunnelOutcome::Disconnected) => {
-                info!(server = %server.server_label, conn = conn_idx, "tunnel disconnected, reconnecting");
-            }
-            Err(e) => {
-                error!(server = %server.server_label, conn = conn_idx, error = %e, "tunnel connection error, reconnecting");
-            }
-        }
-
-        if *shutdown.borrow() {
-            info!(server = %server.server_label, conn = conn_idx, "shutdown requested, not reconnecting");
-            return;
-        }
-
-        // Reset backoff after a stable session to keep recovery snappy when
-        // failures are only occasional.
-        let connected_for = started_at.elapsed();
-        if connected_for >= STABLE_SESSION_RESET_AFTER {
-            consecutive_failures = 0;
-        } else {
-            consecutive_failures = consecutive_failures.saturating_add(1);
-        }
-
-        let reconnect_delay = compute_reconnect_delay(
-            state.config.tunnel_reconnect_base_ms,
-            state.config.tunnel_reconnect_max_ms,
-            consecutive_failures,
-            reconnect_salt,
-        );
-        info!(
-            server = %server.server_label,
-            conn = conn_idx,
-            failures = consecutive_failures,
-            delay_ms = reconnect_delay.as_millis(),
-            "waiting before reconnect"
-        );
-
-        tokio::select! {
-            _ = tokio::time::sleep(reconnect_delay) => {}
-            _ = shutdown.changed() => {
-                info!(server = %server.server_label, conn = conn_idx, "shutdown requested during reconnect wait");
-                return;
-            }
-        }
-    }
-}
-
-fn compute_connection_salt(server: &ServerContext, conn_idx: usize) -> u64 {
-    // FNV-1a style hash over server label + connection index.
-    let mut h: u64 = 0xcbf29ce484222325;
-    for &b in server.server_label.as_bytes() {
-        h ^= b as u64;
-        h = h.wrapping_mul(0x100000001b3);
-    }
-    h ^= conn_idx as u64;
-    mix_u64(h)
-}
-
-fn compute_startup_stagger(conn_idx: usize, salt: u64) -> Duration {
-    if conn_idx == 0 {
-        return Duration::ZERO;
-    }
-    let base = (conn_idx as u64).saturating_mul(STARTUP_STAGGER_STEP_MS);
-    let jitter = mix_u64(salt) % 301; // 0..=300ms
-    Duration::from_millis((base + jitter).min(MAX_STARTUP_STAGGER_MS))
-}
-
-fn compute_reconnect_delay(
-    base_ms: u64,
-    max_ms: u64,
-    consecutive_failures: u32,
-    salt: u64,
-) -> Duration {
-    // First retry should be immediate to maximize recovery speed on transient
-    // blips (the user's primary expectation in poor networks).
-    if consecutive_failures <= 1 {
-        return Duration::ZERO;
-    }
-
-    // Keep a sane minimum for repeated failures.
-    let base_ms = base_ms.max(MIN_RECONNECT_DELAY_MS);
-    let max_ms = max_ms.max(base_ms);
-    let cap_ms = compute_reconnect_cap_ms(base_ms, max_ms, consecutive_failures)
-        .min(RECONNECT_PROBE_MAX_DELAY_MS.max(base_ms));
-
-    // Equal-jitter: randomize in [cap/2, cap], preventing synchronized reconnect
-    // storms while keeping reconnect latency bounded.
-    if cap_ms <= 1 {
-        return Duration::from_millis(cap_ms);
-    }
-
-    let half = cap_ms / 2;
-    let span = cap_ms - half;
-    let now_nanos = SystemTime::now()
-        .duration_since(UNIX_EPOCH)
-        .map(|d| d.subsec_nanos() as u64)
-        .unwrap_or(0);
-    let mixed = mix_u64(now_nanos ^ salt);
-    let jitter = if span == 0 { 0 } else { mixed % (span + 1) };
-    Duration::from_millis(half + jitter)
-}
-
-fn compute_reconnect_cap_ms(base_ms: u64, max_ms: u64, consecutive_failures: u32) -> u64 {
-    if consecutive_failures <= 1 {
-        return base_ms.min(max_ms);
-    }
-
-    let shift = (consecutive_failures - 1).min(31);
-    let factor = 1u64 << shift;
-    base_ms.saturating_mul(factor).min(max_ms)
-}
-
-fn mix_u64(mut x: u64) -> u64 {
-    // SplitMix64 finalizer - cheap bit mixing for pseudo-random jitter.
-    x ^= x >> 30;
-    x = x.wrapping_mul(0xbf58476d1ce4e5b9);
-    x ^= x >> 27;
-    x = x.wrapping_mul(0x94d049bb133111eb);
-    x ^ (x >> 31)
-}
-
-#[cfg(test)]
-mod tests {
-    use std::time::Duration;
-
-    use super::{
-        compute_reconnect_cap_ms, compute_reconnect_delay, compute_startup_stagger,
-        MAX_STARTUP_STAGGER_MS, RECONNECT_PROBE_MAX_DELAY_MS, STARTUP_STAGGER_STEP_MS,
-    };
-
-    #[test]
-    fn reconnect_cap_grows_exponentially_and_caps() {
-        let base = 500;
-        let max = 30_000;
-        assert_eq!(compute_reconnect_cap_ms(base, max, 0), 500);
-        assert_eq!(compute_reconnect_cap_ms(base, max, 1), 500);
-        assert_eq!(compute_reconnect_cap_ms(base, max, 2), 1_000);
-        assert_eq!(compute_reconnect_cap_ms(base, max, 3), 2_000);
-        assert_eq!(compute_reconnect_cap_ms(base, max, 4), 4_000);
-        assert_eq!(compute_reconnect_cap_ms(base, max, 5), 8_000);
-        assert_eq!(compute_reconnect_cap_ms(base, max, 6), 16_000);
-        assert_eq!(compute_reconnect_cap_ms(base, max, 7), 30_000);
-        assert_eq!(compute_reconnect_cap_ms(base, max, 20), 30_000);
-    }
-
-    #[test]
-    fn startup_stagger_is_zero_for_primary_and_bounded_for_secondary() {
-        assert_eq!(compute_startup_stagger(0, 42), Duration::ZERO);
-
-        let d1 = compute_startup_stagger(1, 42);
-        let d2 = compute_startup_stagger(2, 42);
-
-        assert!(d1 >= Duration::from_millis(STARTUP_STAGGER_STEP_MS));
-        assert!(d1 <= Duration::from_millis(MAX_STARTUP_STAGGER_MS));
-        assert!(d2 >= Duration::from_millis(STARTUP_STAGGER_STEP_MS * 2));
-        assert!(d2 <= Duration::from_millis(MAX_STARTUP_STAGGER_MS));
-    }
-
-    #[test]
-    fn reconnect_delay_is_immediate_on_first_failure() {
-        assert_eq!(compute_reconnect_delay(700, 45_000, 1, 123), Duration::ZERO);
-    }
-
-    #[test]
-    fn reconnect_delay_stays_within_probe_ceiling_after_many_failures() {
-        let d = compute_reconnect_delay(500, 45_000, 100, 12345);
-        assert!(d <= Duration::from_millis(RECONNECT_PROBE_MAX_DELAY_MS));
-    }
-}
--- a/crates/aether-gateway/Cargo.toml
+++ b/crates/aether-gateway/Cargo.toml
@@ -17,18 +17,20 @@ aether-runtime.workspace = true
 aether-wallet.workspace = true
 async-stream.workspace = true
 async-trait.workspace = true
-axum = { version = "0.8" }
+axum = { version = "0.8", features = ["ws"] }
 base64.workspace = true
 bcrypt.workspace = true
 bytes.workspace = true
 chrono.workspace = true
 chrono-tz.workspace = true
 clap = { version = "4", features = ["derive", "env"] }
+dashmap = "6"
 flate2.workspace = true
 futures-util.workspace = true
 hmac.workspace = true
 http.workspace = true
 ldap3 = "0.11"
+parking_lot = "0.12"
 regex.workspace = true
 redis.workspace = true
 reqwest.workspace = true
--- a/apps/aether-gateway/examples/execution_runtime_harness.rs
+++ b/apps/aether-gateway/examples/execution_runtime_harness.rs
@@ -3,57 +3,71 @@ use std::path::PathBuf;
 use clap::Parser;
 use tracing::info;

-use aether_executor::server;
+use aether_gateway::{serve_execution_runtime_tcp, serve_execution_runtime_unix};
 use aether_runtime::{
    init_service_runtime, DistributedConcurrencyGate, RedisDistributedConcurrencyConfig,
    ServiceRuntimeConfig,
 };

 #[derive(Parser, Debug)]
-#[command(name = "aether-executor", about = "Internal Rust executor for Aether")]
+#[command(
+    name = "execution-runtime-harness",
+    about = "Internal execution runtime harness for Aether tests"
+)]
 struct Args {
-    #[arg(long, env = "AETHER_EXECUTOR_TRANSPORT", default_value = "unix_socket")]
+    #[arg(
+        long,
+        env = "AETHER_EXECUTION_RUNTIME_TRANSPORT",
+        default_value = "unix_socket"
+    )]
    transport: String,

-    #[arg(long, env = "AETHER_EXECUTOR_BIND", default_value = "127.0.0.1:5219")]
+    #[arg(
+        long,
+        env = "AETHER_EXECUTION_RUNTIME_BIND",
+        default_value = "127.0.0.1:5219"
+    )]
    bind: String,

    #[arg(
        long,
-        env = "AETHER_EXECUTOR_UNIX_SOCKET",
-        default_value = "/tmp/aether-executor.sock"
+        env = "AETHER_EXECUTION_RUNTIME_UNIX_SOCKET",
+        default_value = "/tmp/aether-execution-runtime.sock"
    )]
    unix_socket: PathBuf,

-    #[arg(long, env = "AETHER_EXECUTOR_MAX_IN_FLIGHT_REQUESTS")]
+    #[arg(long, env = "AETHER_EXECUTION_RUNTIME_MAX_IN_FLIGHT_REQUESTS")]
    max_in_flight_requests: Option<usize>,

-    #[arg(long, env = "AETHER_EXECUTOR_DISTRIBUTED_REQUEST_LIMIT")]
+    #[arg(long, env = "AETHER_EXECUTION_RUNTIME_DISTRIBUTED_REQUEST_LIMIT")]
    distributed_request_limit: Option<usize>,

-    #[arg(long, env = "AETHER_EXECUTOR_DISTRIBUTED_REQUEST_REDIS_URL")]
+    #[arg(long, env = "AETHER_EXECUTION_RUNTIME_DISTRIBUTED_REQUEST_REDIS_URL")]
    distributed_request_redis_url: Option<String>,

-    #[arg(long, env = "AETHER_EXECUTOR_DISTRIBUTED_REQUEST_REDIS_KEY_PREFIX")]
+    #[arg(
+        long,
+        env = "AETHER_EXECUTION_RUNTIME_DISTRIBUTED_REQUEST_REDIS_KEY_PREFIX"
+    )]
    distributed_request_redis_key_prefix: Option<String>,

    #[arg(
        long,
-        env = "AETHER_EXECUTOR_DISTRIBUTED_REQUEST_LEASE_TTL_MS",
+        env = "AETHER_EXECUTION_RUNTIME_DISTRIBUTED_REQUEST_LEASE_TTL_MS",
        default_value_t = 30_000
    )]
    distributed_request_lease_ttl_ms: u64,

    #[arg(
        long,
-        env = "AETHER_EXECUTOR_DISTRIBUTED_REQUEST_RENEW_INTERVAL_MS",
+        env = "AETHER_EXECUTION_RUNTIME_DISTRIBUTED_REQUEST_RENEW_INTERVAL_MS",
        default_value_t = 10_000
    )]
    distributed_request_renew_interval_ms: u64,

    #[arg(
        long,
-        env = "AETHER_EXECUTOR_DISTRIBUTED_REQUEST_COMMAND_TIMEOUT_MS",
+        env = "AETHER_EXECUTION_RUNTIME_DISTRIBUTED_REQUEST_COMMAND_TIMEOUT_MS",
        default_value_t = 1_000
    )]
    distributed_request_command_timeout_ms: u64,
@@ -64,8 +78,8 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
    let _ = rustls::crypto::ring::default_provider().install_default();

    init_service_runtime(ServiceRuntimeConfig::new(
-        "aether-executor",
-        "aether_executor=info",
+        "aether-execution-runtime-harness",
+        "aether_gateway=info",
    ))?;

    let args = Args::parse();
@@ -77,13 +91,13 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
                .map(str::trim)
                .filter(|value| !value.is_empty())
                .ok_or_else(|| {
-                    std::io::Error::new(
-                        std::io::ErrorKind::InvalidInput,
-                        "AETHER_EXECUTOR_DISTRIBUTED_REQUEST_REDIS_URL is required when distributed request limit is enabled",
-                    )
+                        std::io::Error::new(
+                            std::io::ErrorKind::InvalidInput,
+                            "AETHER_EXECUTION_RUNTIME_DISTRIBUTED_REQUEST_REDIS_URL is required when distributed request limit is enabled",
+                        )
                })?;
            Some(DistributedConcurrencyGate::new_redis(
-                "executor_requests_distributed",
+                "execution_runtime_requests_distributed",
                limit,
                RedisDistributedConcurrencyConfig {
                    url: redis_url.to_string(),
@@ -101,10 +115,14 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
        }
        None => None,
    };
+
    match args.transport.trim().to_ascii_lowercase().as_str() {
        "unix_socket" | "unix" | "uds" => {
-            info!(socket = %args.unix_socket.display(), "aether-executor started");
-            server::serve_unix(
+            info!(
+                socket = %args.unix_socket.display(),
+                "aether execution-runtime harness started"
+            );
+            serve_execution_runtime_unix(
                &args.unix_socket,
                args.max_in_flight_requests,
                distributed_request_gate.clone(),
@@ -116,9 +134,9 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
                bind = %args.bind,
                max_in_flight_requests = args.max_in_flight_requests.unwrap_or_default(),
                distributed_request_limit = args.distributed_request_limit.unwrap_or_default(),
-                "aether-executor started"
+                "aether execution-runtime harness started"
            );
-            server::serve_tcp(
+            serve_execution_runtime_tcp(
                &args.bind,
                args.max_in_flight_requests,
                distributed_request_gate,
@@ -126,7 +144,7 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
            .await?;
        }
        other => {
-            return Err(format!("unsupported executor transport: {other}").into());
+            return Err(format!("unsupported execution runtime transport: {other}").into());
        }
    }

--- a/apps/aether-gateway/examples/tunnel_runtime_harness.rs
+++ b/apps/aether-gateway/examples/tunnel_runtime_harness.rs
@@ -0,0 +1,161 @@
+use std::net::SocketAddr;
+use std::time::Duration;
+
+use aether_gateway::{
+    build_tunnel_runtime_router_with_state, TunnelConnConfig, TunnelControlPlaneClient,
+    TunnelRuntimeState,
+};
+use aether_runtime::{
+    init_service_runtime, DistributedConcurrencyGate, RedisDistributedConcurrencyConfig,
+    ServiceRuntimeConfig,
+};
+use clap::Parser;
+use tracing::info;
+
+#[derive(Parser, Debug)]
+#[command(
+    name = "aether-tunnel-runtime-harness",
+    about = "Standalone tunnel relay harness backed by aether-gateway tunnel runtime"
+)]
+struct Args {
+    #[arg(
+        long,
+        default_value = "0.0.0.0:8085",
+        env = "AETHER_TUNNEL_STANDALONE_BIND"
+    )]
+    bind: String,
+
+    #[arg(
+        long,
+        default_value_t = 0,
+        env = "AETHER_TUNNEL_STANDALONE_PROXY_IDLE_TIMEOUT"
+    )]
+    proxy_idle_timeout: u64,
+
+    #[arg(
+        long,
+        default_value_t = 15,
+        env = "AETHER_TUNNEL_STANDALONE_PING_INTERVAL"
+    )]
+    ping_interval: u64,
+
+    #[arg(
+        long,
+        default_value_t = 2048,
+        env = "AETHER_TUNNEL_STANDALONE_MAX_STREAMS"
+    )]
+    max_streams: usize,
+
+    #[arg(
+        long,
+        default_value_t = 128,
+        env = "AETHER_TUNNEL_STANDALONE_OUTBOUND_QUEUE_CAPACITY"
+    )]
+    outbound_queue_capacity: usize,
+
+    #[arg(
+        long,
+        default_value = "http://127.0.0.1:8084",
+        env = "AETHER_TUNNEL_STANDALONE_APP_BASE_URL"
+    )]
+    app_base_url: String,
+
+    #[arg(long, env = "AETHER_TUNNEL_STANDALONE_MAX_IN_FLIGHT_REQUESTS")]
+    max_in_flight_requests: Option<usize>,
+
+    #[arg(long, env = "AETHER_TUNNEL_STANDALONE_DISTRIBUTED_REQUEST_LIMIT")]
+    distributed_request_limit: Option<usize>,
+
+    #[arg(long, env = "AETHER_TUNNEL_STANDALONE_DISTRIBUTED_REQUEST_REDIS_URL")]
+    distributed_request_redis_url: Option<String>,
+
+    #[arg(
+        long,
+        env = "AETHER_TUNNEL_STANDALONE_DISTRIBUTED_REQUEST_REDIS_KEY_PREFIX"
+    )]
+    distributed_request_redis_key_prefix: Option<String>,
+
+    #[arg(
+        long,
+        env = "AETHER_TUNNEL_STANDALONE_DISTRIBUTED_REQUEST_LEASE_TTL_MS",
+        default_value_t = 30_000
+    )]
+    distributed_request_lease_ttl_ms: u64,
+
+    #[arg(
+        long,
+        env = "AETHER_TUNNEL_STANDALONE_DISTRIBUTED_REQUEST_RENEW_INTERVAL_MS",
+        default_value_t = 10_000
+    )]
+    distributed_request_renew_interval_ms: u64,
+
+    #[arg(
+        long,
+        env = "AETHER_TUNNEL_STANDALONE_DISTRIBUTED_REQUEST_COMMAND_TIMEOUT_MS",
+        default_value_t = 1_000
+    )]
+    distributed_request_command_timeout_ms: u64,
+}
+
+#[tokio::main]
+async fn main() -> Result<(), Box<dyn std::error::Error>> {
+    init_service_runtime(ServiceRuntimeConfig::new(
+        "aether-tunnel-standalone",
+        "aether_gateway=info",
+    ))?;
+
+    let args = Args::parse();
+    let outbound_queue_capacity = args.outbound_queue_capacity.clamp(8, 4096);
+    let ping_interval = Duration::from_secs(args.ping_interval);
+    let mut state = TunnelRuntimeState::new(
+        TunnelControlPlaneClient::new(args.app_base_url),
+        TunnelConnConfig {
+            ping_interval,
+            idle_timeout: Duration::from_secs(args.proxy_idle_timeout),
+            outbound_queue_capacity,
+        },
+        args.max_streams,
+    )
+    .with_request_concurrency_limit(args.max_in_flight_requests);
+
+    if let Some(limit) = args.distributed_request_limit.filter(|limit| *limit > 0) {
+        let redis_url = args
+            .distributed_request_redis_url
+            .as_deref()
+            .map(str::trim)
+            .filter(|value| !value.is_empty())
+            .ok_or_else(|| {
+                std::io::Error::new(
+                    std::io::ErrorKind::InvalidInput,
+                    "AETHER_TUNNEL_STANDALONE_DISTRIBUTED_REQUEST_REDIS_URL is required when distributed request limit is enabled",
+                )
+            })?;
+        state = state.with_distributed_request_gate(DistributedConcurrencyGate::new_redis(
+            "tunnel_requests_distributed",
+            limit,
+            RedisDistributedConcurrencyConfig {
+                url: redis_url.to_string(),
+                key_prefix: args
+                    .distributed_request_redis_key_prefix
+                    .as_deref()
+                    .map(str::trim)
+                    .filter(|value| !value.is_empty())
+                    .map(ToOwned::to_owned),
+                lease_ttl_ms: args.distributed_request_lease_ttl_ms.max(1),
+                renew_interval_ms: args.distributed_request_renew_interval_ms.max(1),
+                command_timeout_ms: Some(args.distributed_request_command_timeout_ms.max(1)),
+            },
+        )?);
+    }
+
+    let app = build_tunnel_runtime_router_with_state(state);
+    let listener = tokio::net::TcpListener::bind(&args.bind).await?;
+    info!(bind = %args.bind, "tunnel runtime harness started");
+
+    axum::serve(
+        listener,
+        app.into_make_service_with_connect_info::<SocketAddr>(),
+    )
+    .await?;
+    Ok(())
+}
--- a/apps/aether-gateway/src/ai_pipeline/conversion/error.rs
+++ b/apps/aether-gateway/src/ai_pipeline/conversion/error.rs
@@ -0,0 +1,165 @@
+use serde_json::{Map, Value};
+
+#[derive(Clone, Copy, Debug, Eq, PartialEq)]
+pub(crate) enum LocalCoreSyncErrorKind {
+    InvalidRequest,
+    Authentication,
+    PermissionDenied,
+    NotFound,
+    RateLimit,
+    ContextLengthExceeded,
+    Overloaded,
+    ServerError,
+}
+
+pub(crate) fn is_core_error_finalize_kind(report_kind: &str) -> bool {
+    core_error_default_client_api_format(report_kind).is_some()
+}
+
+pub(crate) fn core_error_default_client_api_format(report_kind: &str) -> Option<&'static str> {
+    match report_kind {
+        "openai_chat_sync_finalize" => Some("openai:chat"),
+        "claude_chat_sync_finalize" => Some("claude:chat"),
+        "gemini_chat_sync_finalize" => Some("gemini:chat"),
+        "openai_cli_sync_finalize" => Some("openai:cli"),
+        "openai_compact_sync_finalize" => Some("openai:compact"),
+        "claude_cli_sync_finalize" => Some("claude:cli"),
+        "gemini_cli_sync_finalize" => Some("gemini:cli"),
+        _ => None,
+    }
+}
+
+pub(crate) fn core_error_background_report_kind(report_kind: &str) -> Option<&'static str> {
+    match report_kind {
+        "openai_chat_sync_finalize" => Some("openai_chat_sync_error"),
+        "claude_chat_sync_finalize" => Some("claude_chat_sync_error"),
+        "gemini_chat_sync_finalize" => Some("gemini_chat_sync_error"),
+        "openai_cli_sync_finalize" => Some("openai_cli_sync_error"),
+        "openai_compact_sync_finalize" => Some("openai_compact_sync_error"),
+        "claude_cli_sync_finalize" => Some("claude_cli_sync_error"),
+        "gemini_cli_sync_finalize" => Some("gemini_cli_sync_error"),
+        _ => None,
+    }
+}
+
+#[cfg(test)]
+pub(crate) fn core_success_background_report_kind(report_kind: &str) -> Option<&'static str> {
+    match report_kind {
+        "openai_chat_sync_finalize" => Some("openai_chat_sync_success"),
+        "claude_chat_sync_finalize" => Some("claude_chat_sync_success"),
+        "gemini_chat_sync_finalize" => Some("gemini_chat_sync_success"),
+        "openai_cli_sync_finalize" | "openai_compact_sync_finalize" => {
+            Some("openai_cli_sync_success")
+        }
+        "claude_cli_sync_finalize" => Some("claude_cli_sync_success"),
+        "gemini_cli_sync_finalize" => Some("gemini_cli_sync_success"),
+        _ => None,
+    }
+}
+
+pub(crate) fn build_core_error_body_for_client_format(
+    client_api_format: &str,
+    message: &str,
+    code: Option<&str>,
+    kind: LocalCoreSyncErrorKind,
+) -> Option<Value> {
+    let mut error_object = Map::new();
+    error_object.insert("message".to_string(), Value::String(message.to_string()));
+
+    match client_api_format.trim().to_ascii_lowercase().as_str() {
+        "openai:chat" | "openai:cli" | "openai:compact" => {
+            error_object.insert(
+                "type".to_string(),
+                Value::String(map_local_sync_error_kind_to_openai_type(kind).to_string()),
+            );
+            if let Some(code) = code.filter(|value| !value.is_empty()) {
+                error_object.insert("code".to_string(), Value::String(code.to_string()));
+            }
+            Some(Value::Object(Map::from_iter([(
+                "error".to_string(),
+                Value::Object(error_object),
+            )])))
+        }
+        "claude:chat" | "claude:cli" => {
+            error_object.insert(
+                "type".to_string(),
+                Value::String(map_local_sync_error_kind_to_claude_type(kind).to_string()),
+            );
+            if let Some(code) = code.filter(|value| !value.is_empty()) {
+                error_object.insert("code".to_string(), Value::String(code.to_string()));
+            }
+            Some(Value::Object(Map::from_iter([
+                ("type".to_string(), Value::String("error".to_string())),
+                ("error".to_string(), Value::Object(error_object)),
+            ])))
+        }
+        "gemini:chat" | "gemini:cli" => Some(Value::Object(Map::from_iter([(
+            "error".to_string(),
+            Value::Object(Map::from_iter([
+                (
+                    "code".to_string(),
+                    Value::from(map_local_sync_error_kind_to_gemini_code(kind)),
+                ),
+                ("message".to_string(), Value::String(message.to_string())),
+                (
+                    "status".to_string(),
+                    Value::String(map_local_sync_error_kind_to_gemini_status(kind).to_string()),
+                ),
+            ])),
+        )]))),
+        _ => None,
+    }
+}
+
+fn map_local_sync_error_kind_to_openai_type(kind: LocalCoreSyncErrorKind) -> &'static str {
+    match kind {
+        LocalCoreSyncErrorKind::InvalidRequest => "invalid_request_error",
+        LocalCoreSyncErrorKind::Authentication => "authentication_error",
+        LocalCoreSyncErrorKind::PermissionDenied => "permission_error",
+        LocalCoreSyncErrorKind::NotFound => "not_found_error",
+        LocalCoreSyncErrorKind::RateLimit => "rate_limit_error",
+        LocalCoreSyncErrorKind::ContextLengthExceeded => "context_length_exceeded",
+        LocalCoreSyncErrorKind::Overloaded | LocalCoreSyncErrorKind::ServerError => "server_error",
+    }
+}
+
+fn map_local_sync_error_kind_to_claude_type(kind: LocalCoreSyncErrorKind) -> &'static str {
+    match kind {
+        LocalCoreSyncErrorKind::InvalidRequest | LocalCoreSyncErrorKind::ContextLengthExceeded => {
+            "invalid_request_error"
+        }
+        LocalCoreSyncErrorKind::Authentication => "authentication_error",
+        LocalCoreSyncErrorKind::PermissionDenied => "permission_error",
+        LocalCoreSyncErrorKind::NotFound => "not_found_error",
+        LocalCoreSyncErrorKind::RateLimit => "rate_limit_error",
+        LocalCoreSyncErrorKind::Overloaded | LocalCoreSyncErrorKind::ServerError => "api_error",
+    }
+}
+
+fn map_local_sync_error_kind_to_gemini_code(kind: LocalCoreSyncErrorKind) -> u16 {
+    match kind {
+        LocalCoreSyncErrorKind::InvalidRequest | LocalCoreSyncErrorKind::ContextLengthExceeded => {
+            400
+        }
+        LocalCoreSyncErrorKind::Authentication => 401,
+        LocalCoreSyncErrorKind::PermissionDenied => 403,
+        LocalCoreSyncErrorKind::NotFound => 404,
+        LocalCoreSyncErrorKind::RateLimit => 429,
+        LocalCoreSyncErrorKind::Overloaded => 503,
+        LocalCoreSyncErrorKind::ServerError => 500,
+    }
+}
+
+fn map_local_sync_error_kind_to_gemini_status(kind: LocalCoreSyncErrorKind) -> &'static str {
+    match kind {
+        LocalCoreSyncErrorKind::InvalidRequest | LocalCoreSyncErrorKind::ContextLengthExceeded => {
+            "INVALID_ARGUMENT"
+        }
+        LocalCoreSyncErrorKind::Authentication => "UNAUTHENTICATED",
+        LocalCoreSyncErrorKind::PermissionDenied => "PERMISSION_DENIED",
+        LocalCoreSyncErrorKind::NotFound => "NOT_FOUND",
+        LocalCoreSyncErrorKind::RateLimit => "RESOURCE_EXHAUSTED",
+        LocalCoreSyncErrorKind::Overloaded => "UNAVAILABLE",
+        LocalCoreSyncErrorKind::ServerError => "INTERNAL",
+    }
+}
--- a/apps/aether-gateway/src/ai_pipeline/conversion/mod.rs
+++ b/apps/aether-gateway/src/ai_pipeline/conversion/mod.rs
@@ -0,0 +1,17 @@
+pub(crate) mod error;
+pub(crate) mod registry;
+pub(crate) mod request;
+pub(crate) mod response;
+
+#[cfg(test)]
+pub(crate) use error::core_success_background_report_kind;
+pub(crate) use error::{
+    build_core_error_body_for_client_format, core_error_background_report_kind,
+    core_error_default_client_api_format, is_core_error_finalize_kind, LocalCoreSyncErrorKind,
+};
+pub(crate) use registry::{
+    request_conversion_direct_auth, request_conversion_kind,
+    request_conversion_transport_supported, sync_chat_response_conversion_kind,
+    sync_cli_response_conversion_kind, RequestConversionKind, SyncChatResponseConversionKind,
+    SyncCliResponseConversionKind,
+};
--- a/apps/aether-gateway/src/ai_pipeline/conversion/registry.rs
+++ b/apps/aether-gateway/src/ai_pipeline/conversion/registry.rs
@@ -0,0 +1,204 @@
+use crate::gateway::provider_transport::{
+    resolve_local_gemini_auth, resolve_local_openai_chat_auth, resolve_local_standard_auth,
+    supports_local_gemini_transport_with_network, supports_local_openai_chat_transport,
+    supports_local_standard_transport_with_network, GatewayProviderTransportSnapshot,
+};
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub(crate) enum RequestConversionKind {
+    ToOpenAIChat,
+    ToOpenAIFamilyCli,
+    ToOpenAICompact,
+    ToClaudeStandard,
+    ToGeminiStandard,
+}
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub(crate) enum SyncChatResponseConversionKind {
+    ToOpenAIChat,
+    ToClaudeChat,
+    ToGeminiChat,
+}
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub(crate) enum SyncCliResponseConversionKind {
+    ToOpenAIFamilyCli,
+    ToClaudeCli,
+    ToGeminiCli,
+}
+
+pub(crate) fn request_conversion_kind(
+    client_api_format: &str,
+    provider_api_format: &str,
+) -> Option<RequestConversionKind> {
+    let client_api_format = client_api_format.trim().to_ascii_lowercase();
+    let provider_api_format = provider_api_format.trim().to_ascii_lowercase();
+    if client_api_format == provider_api_format {
+        return None;
+    }
+    if !is_standard_api_format(client_api_format.as_str())
+        || !is_standard_api_format(provider_api_format.as_str())
+    {
+        return None;
+    }
+
+    match provider_api_format.as_str() {
+        "openai:chat" => Some(RequestConversionKind::ToOpenAIChat),
+        "openai:cli" => Some(RequestConversionKind::ToOpenAIFamilyCli),
+        "openai:compact" => Some(RequestConversionKind::ToOpenAICompact),
+        "claude:chat" | "claude:cli" => Some(RequestConversionKind::ToClaudeStandard),
+        "gemini:chat" | "gemini:cli" => Some(RequestConversionKind::ToGeminiStandard),
+        _ => None,
+    }
+}
+
+pub(crate) fn request_conversion_transport_supported(
+    transport: &GatewayProviderTransportSnapshot,
+    _kind: RequestConversionKind,
+) -> bool {
+    match transport
+        .endpoint
+        .api_format
+        .trim()
+        .to_ascii_lowercase()
+        .as_str()
+    {
+        "openai:chat" => supports_local_openai_chat_transport(transport),
+        "openai:cli" => supports_local_standard_transport_with_network(transport, "openai:cli"),
+        "openai:compact" => {
+            supports_local_standard_transport_with_network(transport, "openai:compact")
+        }
+        "claude:chat" => supports_local_standard_transport_with_network(transport, "claude:chat"),
+        "claude:cli" => supports_local_standard_transport_with_network(transport, "claude:cli"),
+        "gemini:chat" => supports_local_gemini_transport_with_network(transport, "gemini:chat"),
+        "gemini:cli" => supports_local_gemini_transport_with_network(transport, "gemini:cli"),
+        _ => false,
+    }
+}
+
+pub(crate) fn request_conversion_direct_auth(
+    transport: &GatewayProviderTransportSnapshot,
+    _kind: RequestConversionKind,
+) -> Option<(String, String)> {
+    match transport
+        .endpoint
+        .api_format
+        .trim()
+        .to_ascii_lowercase()
+        .as_str()
+    {
+        "openai:chat" => resolve_local_openai_chat_auth(transport),
+        "gemini:chat" | "gemini:cli" => resolve_local_gemini_auth(transport),
+        "openai:cli" | "openai:compact" | "claude:chat" | "claude:cli" => {
+            resolve_local_standard_auth(transport)
+        }
+        _ => None,
+    }
+}
+
+pub(crate) fn sync_chat_response_conversion_kind(
+    provider_api_format: &str,
+    client_api_format: &str,
+) -> Option<SyncChatResponseConversionKind> {
+    let provider_api_format = provider_api_format.trim().to_ascii_lowercase();
+    let client_api_format = client_api_format.trim().to_ascii_lowercase();
+    if provider_api_format == client_api_format {
+        return None;
+    }
+    if !is_standard_api_format(provider_api_format.as_str()) {
+        return None;
+    }
+    match client_api_format.as_str() {
+        "openai:chat" => Some(SyncChatResponseConversionKind::ToOpenAIChat),
+        "claude:chat" => Some(SyncChatResponseConversionKind::ToClaudeChat),
+        "gemini:chat" => Some(SyncChatResponseConversionKind::ToGeminiChat),
+        _ => None,
+    }
+}
+
+pub(crate) fn sync_cli_response_conversion_kind(
+    provider_api_format: &str,
+    client_api_format: &str,
+) -> Option<SyncCliResponseConversionKind> {
+    let provider_api_format = provider_api_format.trim().to_ascii_lowercase();
+    let client_api_format = client_api_format.trim().to_ascii_lowercase();
+    if provider_api_format == client_api_format {
+        return None;
+    }
+    if !is_standard_api_format(provider_api_format.as_str()) {
+        return None;
+    }
+    match client_api_format.as_str() {
+        "openai:cli" | "openai:compact" => Some(SyncCliResponseConversionKind::ToOpenAIFamilyCli),
+        "claude:cli" => Some(SyncCliResponseConversionKind::ToClaudeCli),
+        "gemini:cli" => Some(SyncCliResponseConversionKind::ToGeminiCli),
+        _ => None,
+    }
+}
+
+fn is_standard_api_format(api_format: &str) -> bool {
+    matches!(
+        api_format,
+        "openai:chat"
+            | "openai:cli"
+            | "openai:compact"
+            | "claude:chat"
+            | "claude:cli"
+            | "gemini:chat"
+            | "gemini:cli"
+    )
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn request_conversion_registry_supports_bidirectional_standard_matrix() {
+        assert_eq!(
+            request_conversion_kind("claude:chat", "openai:chat"),
+            Some(RequestConversionKind::ToOpenAIChat)
+        );
+        assert_eq!(
+            request_conversion_kind("gemini:chat", "claude:chat"),
+            Some(RequestConversionKind::ToClaudeStandard)
+        );
+        assert_eq!(
+            request_conversion_kind("gemini:cli", "openai:compact"),
+            Some(RequestConversionKind::ToOpenAICompact)
+        );
+        assert_eq!(
+            request_conversion_kind("openai:compact", "gemini:cli"),
+            Some(RequestConversionKind::ToGeminiStandard)
+        );
+        assert_eq!(request_conversion_kind("claude:chat", "claude:chat"), None);
+    }
+
+    #[test]
+    fn sync_response_conversion_registry_supports_bidirectional_standard_matrix() {
+        assert_eq!(
+            sync_chat_response_conversion_kind("openai:chat", "claude:chat"),
+            Some(SyncChatResponseConversionKind::ToClaudeChat)
+        );
+        assert_eq!(
+            sync_chat_response_conversion_kind("claude:chat", "gemini:chat"),
+            Some(SyncChatResponseConversionKind::ToGeminiChat)
+        );
+        assert_eq!(
+            sync_chat_response_conversion_kind("gemini:chat", "openai:chat"),
+            Some(SyncChatResponseConversionKind::ToOpenAIChat)
+        );
+        assert_eq!(
+            sync_cli_response_conversion_kind("openai:cli", "gemini:cli"),
+            Some(SyncCliResponseConversionKind::ToGeminiCli)
+        );
+        assert_eq!(
+            sync_cli_response_conversion_kind("claude:cli", "openai:compact"),
+            Some(SyncCliResponseConversionKind::ToOpenAIFamilyCli)
+        );
+        assert_eq!(
+            sync_cli_response_conversion_kind("gemini:cli", "claude:cli"),
+            Some(SyncCliResponseConversionKind::ToClaudeCli)
+        );
+    }
+}
--- a/apps/aether-gateway/src/ai_pipeline/conversion/request/from_openai_chat/claude.rs
+++ b/apps/aether-gateway/src/ai_pipeline/conversion/request/from_openai_chat/claude.rs
@@ -0,0 +1,363 @@
+use serde_json::{json, Map, Value};
+use uuid::Uuid;
+
+use super::shared::parse_openai_tool_arguments;
+use super::super::to_openai_chat::{extract_openai_text_content, parse_openai_tool_result_content};
+use crate::gateway::ai_pipeline::planner::standard::{
+    copy_request_number_field, map_openai_reasoning_effort_to_claude_output,
+    parse_openai_stop_sequences, resolve_openai_chat_max_tokens,
+};
+
+pub(crate) fn convert_openai_chat_request_to_claude_request(
+    body_json: &Value,
+    mapped_model: &str,
+    upstream_is_stream: bool,
+) -> Option<Value> {
+    let request = body_json.as_object()?;
+    let mut system_segments = Vec::new();
+    let mut messages = Vec::new();
+
+    if let Some(message_values) = request.get("messages").and_then(Value::as_array) {
+        for message in message_values {
+            let message_object = message.as_object()?;
+            let role = message_object
+                .get("role")
+                .and_then(Value::as_str)
+                .unwrap_or_default()
+                .trim()
+                .to_ascii_lowercase();
+            match role.as_str() {
+                "system" | "developer" => {
+                    let text = extract_openai_text_content(message_object.get("content"))?;
+                    if !text.trim().is_empty() {
+                        system_segments.push(text);
+                    }
+                }
+                "user" => {
+                    let blocks =
+                        convert_openai_content_to_claude_blocks(message_object.get("content"), true)?;
+                    if !blocks.is_empty() {
+                        messages.push(build_claude_message("user", blocks));
+                    }
+                }
+                "assistant" => {
+                    let mut blocks = convert_openai_content_to_claude_blocks(
+                        message_object.get("content"),
+                        false,
+                    )?;
+                    if let Some(tool_calls) =
+                        message_object.get("tool_calls").and_then(Value::as_array)
+                    {
+                        for tool_call in tool_calls {
+                            let tool_call_object = tool_call.as_object()?;
+                            let function = tool_call_object.get("function")?.as_object()?;
+                            let tool_name = function
+                                .get("name")
+                                .and_then(Value::as_str)
+                                .map(str::trim)
+                                .filter(|value| !value.is_empty())?
+                                .to_string();
+                            let tool_call_id = tool_call_object
+                                .get("id")
+                                .and_then(Value::as_str)
+                                .map(str::trim)
+                                .filter(|value| !value.is_empty())
+                                .map(ToOwned::to_owned)
+                                .unwrap_or_else(|| format!("toolu_{}", Uuid::new_v4().simple()));
+                            let tool_input =
+                                parse_openai_tool_arguments(function.get("arguments"))?;
+                            blocks.push(json!({
+                                "type": "tool_use",
+                                "id": tool_call_id,
+                                "name": tool_name,
+                                "input": tool_input,
+                            }));
+                        }
+                    }
+                    if !blocks.is_empty() {
+                        messages.push(build_claude_message("assistant", blocks));
+                    }
+                }
+                "tool" => {
+                    let tool_use_id = message_object
+                        .get("tool_call_id")
+                        .and_then(Value::as_str)
+                        .map(str::trim)
+                        .filter(|value| !value.is_empty())?
+                        .to_string();
+                    let tool_result =
+                        parse_openai_tool_result_content(message_object.get("content"));
+                    messages.push(json!({
+                        "role": "user",
+                        "content": [{
+                            "type": "tool_result",
+                            "tool_use_id": tool_use_id,
+                            "content": tool_result,
+                            "is_error": false,
+                        }],
+                    }));
+                }
+                _ => {}
+            }
+        }
+    }
+
+    let mut output = Map::new();
+    output.insert("model".to_string(), Value::String(mapped_model.to_string()));
+    output.insert(
+        "messages".to_string(),
+        Value::Array(compact_claude_messages(messages)),
+    );
+    output.insert(
+        "max_tokens".to_string(),
+        Value::from(resolve_openai_chat_max_tokens(request)),
+    );
+
+    let system_text = system_segments
+        .into_iter()
+        .filter(|value| !value.trim().is_empty())
+        .collect::<Vec<_>>()
+        .join("\n\n");
+    if !system_text.is_empty() {
+        output.insert("system".to_string(), Value::String(system_text));
+    }
+    if upstream_is_stream {
+        output.insert("stream".to_string(), Value::Bool(true));
+    }
+    copy_request_number_field(request, &mut output, "temperature");
+    copy_request_number_field(request, &mut output, "top_p");
+    copy_request_number_field(request, &mut output, "top_k");
+    if let Some(stop_sequences) = parse_openai_stop_sequences(request.get("stop")) {
+        output.insert("stop_sequences".to_string(), Value::Array(stop_sequences));
+    }
+    if let Some(tools) = convert_openai_tools_to_claude(request.get("tools")) {
+        output.insert("tools".to_string(), Value::Array(tools));
+    }
+    if let Some(tool_choice) = convert_openai_tool_choice_to_claude(request.get("tool_choice")) {
+        output.insert("tool_choice".to_string(), tool_choice);
+    }
+    if let Some(metadata) = request.get("metadata").cloned() {
+        output.insert("metadata".to_string(), metadata);
+    }
+    if let Some(reasoning_effort) = request.get("reasoning_effort").and_then(Value::as_str) {
+        if let Some(output_effort) = map_openai_reasoning_effort_to_claude_output(reasoning_effort)
+        {
+            output.insert(
+                "output_config".to_string(),
+                json!({ "effort": output_effort }),
+            );
+        }
+    }
+
+    Some(Value::Object(output))
+}
+
+fn convert_openai_content_to_claude_blocks(
+    content: Option<&Value>,
+    allow_images: bool,
+) -> Option<Vec<Value>> {
+    match content {
+        None | Some(Value::Null) => Some(Vec::new()),
+        Some(Value::String(text)) => {
+            let trimmed = text.trim();
+            if trimmed.is_empty() {
+                Some(Vec::new())
+            } else {
+                Some(vec![json!({ "type": "text", "text": text })])
+            }
+        }
+        Some(Value::Array(parts)) => {
+            let mut blocks = Vec::new();
+            for part in parts {
+                let part_object = part.as_object()?;
+                let part_type = part_object
+                    .get("type")
+                    .and_then(Value::as_str)
+                    .unwrap_or_default();
+                match part_type {
+                    "text" | "input_text" => {
+                        if let Some(text) = part_object.get("text").and_then(Value::as_str) {
+                            if !text.trim().is_empty() {
+                                blocks.push(json!({ "type": "text", "text": text }));
+                            }
+                        }
+                    }
+                    "image_url" | "input_image" if allow_images => {
+                        let url = part_object
+                            .get("image_url")
+                            .and_then(|value| {
+                                value.as_str().map(ToOwned::to_owned).or_else(|| {
+                                    value
+                                        .as_object()
+                                        .and_then(|object| object.get("url"))
+                                        .and_then(Value::as_str)
+                                        .map(ToOwned::to_owned)
+                                })
+                            })
+                            .filter(|value| !value.trim().is_empty())?;
+                        blocks.push(json!({
+                            "type": "image",
+                            "source": {
+                                "type": "url",
+                                "url": url,
+                            }
+                        }));
+                    }
+                    _ => {}
+                }
+            }
+            Some(blocks)
+        }
+        _ => None,
+    }
+}
+
+fn convert_openai_tools_to_claude(tools: Option<&Value>) -> Option<Vec<Value>> {
+    let tool_values = tools?.as_array()?;
+    let mut converted = Vec::new();
+    for tool in tool_values {
+        let tool_object = tool.as_object()?;
+        if tool_object
+            .get("type")
+            .and_then(Value::as_str)
+            .is_some_and(|value| value != "function")
+        {
+            continue;
+        }
+        let function = tool_object.get("function")?.as_object()?;
+        let name = function
+            .get("name")
+            .and_then(Value::as_str)
+            .map(str::trim)
+            .filter(|value| !value.is_empty())?;
+        let mut converted_tool = Map::new();
+        converted_tool.insert("name".to_string(), Value::String(name.to_string()));
+        if let Some(description) = function.get("description").cloned() {
+            converted_tool.insert("description".to_string(), description);
+        }
+        converted_tool.insert(
+            "input_schema".to_string(),
+            function
+                .get("parameters")
+                .cloned()
+                .unwrap_or_else(|| json!({})),
+        );
+        converted.push(Value::Object(converted_tool));
+    }
+    (!converted.is_empty()).then_some(converted)
+}
+
+fn convert_openai_tool_choice_to_claude(tool_choice: Option<&Value>) -> Option<Value> {
+    let tool_choice = tool_choice?;
+    match tool_choice {
+        Value::String(value) => match value.trim().to_ascii_lowercase().as_str() {
+            "none" => Some(json!({ "type": "none" })),
+            "required" => Some(json!({ "type": "any" })),
+            "auto" => Some(json!({ "type": "auto" })),
+            _ => None,
+        },
+        Value::Object(object) => {
+            let function_name = object
+                .get("function")
+                .and_then(Value::as_object)
+                .and_then(|function| function.get("name"))
+                .and_then(Value::as_str)
+                .map(str::trim)
+                .filter(|value| !value.is_empty())?;
+            Some(json!({
+                "type": "tool",
+                "name": function_name,
+            }))
+        }
+        _ => None,
+    }
+}
+
+fn compact_claude_messages(messages: Vec<Value>) -> Vec<Value> {
+    let mut compact: Vec<Value> = Vec::new();
+    for message in messages {
+        let role = message
+            .get("role")
+            .and_then(Value::as_str)
+            .unwrap_or_default()
+            .to_string();
+        if let Some(last) = compact.last_mut() {
+            let last_role = last
+                .get("role")
+                .and_then(Value::as_str)
+                .unwrap_or_default()
+                .to_string();
+            if last_role == role {
+                merge_claude_message_content(last, message);
+                continue;
+            }
+        }
+        compact.push(message);
+    }
+    if compact
+        .first()
+        .and_then(|value| value.get("role"))
+        .and_then(Value::as_str)
+        .is_some_and(|value| value == "assistant")
+    {
+        compact.insert(0, json!({ "role": "user", "content": "" }));
+    }
+    compact
+}
+
+fn merge_claude_message_content(target: &mut Value, message: Value) {
+    let Some(target_object) = target.as_object_mut() else {
+        return;
+    };
+    let incoming_content = message.get("content").cloned().unwrap_or(Value::Null);
+    let merged_blocks = extract_claude_content_blocks(target_object.get("content"))
+        .into_iter()
+        .chain(extract_claude_content_blocks(Some(&incoming_content)))
+        .collect::<Vec<_>>();
+    target_object.insert(
+        "content".to_string(),
+        simplify_claude_content(merged_blocks),
+    );
+}
+
+fn build_claude_message(role: &str, blocks: Vec<Value>) -> Value {
+    json!({
+        "role": role,
+        "content": simplify_claude_content(blocks),
+    })
+}
+
+fn simplify_claude_content(blocks: Vec<Value>) -> Value {
+    if blocks.is_empty() {
+        return Value::String(String::new());
+    }
+    let mut text_values = Vec::new();
+    for block in &blocks {
+        let Some(block_object) = block.as_object() else {
+            return Value::Array(blocks);
+        };
+        if block_object
+            .get("type")
+            .and_then(Value::as_str)
+            .is_some_and(|value| value == "text")
+        {
+            if let Some(text) = block_object.get("text").and_then(Value::as_str) {
+                text_values.push(text.to_string());
+                continue;
+            }
+        }
+        return Value::Array(blocks);
+    }
+    Value::String(text_values.join("\n"))
+}
+
+fn extract_claude_content_blocks(content: Option<&Value>) -> Vec<Value> {
+    match content {
+        Some(Value::String(text)) if !text.is_empty() => vec![json!({
+            "type": "text",
+            "text": text,
+        })],
+        Some(Value::Array(blocks)) => blocks.clone(),
+        _ => Vec::new(),
+    }
+}
--- a/apps/aether-gateway/src/ai_pipeline/conversion/request/from_openai_chat/gemini.rs
+++ b/apps/aether-gateway/src/ai_pipeline/conversion/request/from_openai_chat/gemini.rs
@@ -0,0 +1,351 @@
+use std::collections::BTreeMap;
+
+use serde_json::{json, Map, Value};
+use uuid::Uuid;
+
+use super::shared::parse_openai_tool_arguments;
+use super::super::to_openai_chat::{extract_openai_text_content, parse_openai_tool_result_content};
+use crate::gateway::ai_pipeline::planner::standard::{
+    copy_request_number_field_as, map_openai_reasoning_effort_to_gemini_budget,
+    parse_openai_stop_sequences, value_as_u64,
+};
+
+pub(crate) fn convert_openai_chat_request_to_gemini_request(
+    body_json: &Value,
+    mapped_model: &str,
+    upstream_is_stream: bool,
+) -> Option<Value> {
+    let request = body_json.as_object()?;
+    let mut system_segments = Vec::new();
+    let mut tool_name_by_id = BTreeMap::new();
+    let mut contents = Vec::new();
+
+    if let Some(message_values) = request.get("messages").and_then(Value::as_array) {
+        for message in message_values {
+            let message_object = message.as_object()?;
+            let role = message_object
+                .get("role")
+                .and_then(Value::as_str)
+                .unwrap_or_default()
+                .trim()
+                .to_ascii_lowercase();
+            match role.as_str() {
+                "system" | "developer" => {
+                    let text = extract_openai_text_content(message_object.get("content"))?;
+                    if !text.trim().is_empty() {
+                        system_segments.push(text);
+                    }
+                }
+                "user" => {
+                    let parts =
+                        convert_openai_content_to_gemini_parts(message_object.get("content"), true)?;
+                    if !parts.is_empty() {
+                        contents.push(json!({
+                            "role": "user",
+                            "parts": parts,
+                        }));
+                    }
+                }
+                "assistant" => {
+                    let mut parts = convert_openai_content_to_gemini_parts(
+                        message_object.get("content"),
+                        false,
+                    )?;
+                    if let Some(tool_calls) =
+                        message_object.get("tool_calls").and_then(Value::as_array)
+                    {
+                        for tool_call in tool_calls {
+                            let tool_call_object = tool_call.as_object()?;
+                            let function = tool_call_object.get("function")?.as_object()?;
+                            let tool_name = function
+                                .get("name")
+                                .and_then(Value::as_str)
+                                .map(str::trim)
+                                .filter(|value| !value.is_empty())?
+                                .to_string();
+                            let tool_call_id = tool_call_object
+                                .get("id")
+                                .and_then(Value::as_str)
+                                .map(str::trim)
+                                .filter(|value| !value.is_empty())
+                                .map(ToOwned::to_owned)
+                                .unwrap_or_else(|| format!("toolu_{}", Uuid::new_v4().simple()));
+                            let tool_input =
+                                parse_openai_tool_arguments(function.get("arguments"))?;
+                            tool_name_by_id.insert(tool_call_id.clone(), tool_name.clone());
+                            parts.push(json!({
+                                "functionCall": {
+                                    "name": tool_name,
+                                    "args": tool_input,
+                                    "id": tool_call_id,
+                                }
+                            }));
+                        }
+                    }
+                    if !parts.is_empty() {
+                        contents.push(json!({
+                            "role": "model",
+                            "parts": parts,
+                        }));
+                    }
+                }
+                "tool" => {
+                    let tool_use_id = message_object
+                        .get("tool_call_id")
+                        .and_then(Value::as_str)
+                        .map(str::trim)
+                        .filter(|value| !value.is_empty())?
+                        .to_string();
+                    let tool_name = tool_name_by_id
+                        .get(&tool_use_id)
+                        .cloned()
+                        .unwrap_or_else(|| tool_use_id.clone());
+                    let tool_result =
+                        parse_openai_tool_result_content(message_object.get("content"));
+                    contents.push(json!({
+                        "role": "user",
+                        "parts": [{
+                            "functionResponse": {
+                                "name": tool_name,
+                                "id": tool_use_id,
+                                "response": {
+                                    "result": tool_result,
+                                },
+                            }
+                        }],
+                    }));
+                }
+                _ => {}
+            }
+        }
+    }
+
+    let mut output = Map::new();
+    output.insert("model".to_string(), Value::String(mapped_model.to_string()));
+    output.insert(
+        "contents".to_string(),
+        Value::Array(compact_gemini_contents(contents)),
+    );
+    if upstream_is_stream {
+        output.insert("stream".to_string(), Value::Bool(true));
+    }
+    let system_text = system_segments
+        .into_iter()
+        .filter(|value| !value.trim().is_empty())
+        .collect::<Vec<_>>()
+        .join("\n\n");
+    if !system_text.is_empty() {
+        output.insert(
+            "systemInstruction".to_string(),
+            json!({ "parts": [{ "text": system_text }] }),
+        );
+    }
+
+    let mut generation_config = Map::new();
+    if let Some(max_tokens) = request
+        .get("max_completion_tokens")
+        .and_then(value_as_u64)
+        .or_else(|| request.get("max_tokens").and_then(value_as_u64))
+    {
+        generation_config.insert("maxOutputTokens".to_string(), Value::from(max_tokens));
+    }
+    copy_request_number_field_as(
+        request,
+        &mut generation_config,
+        "temperature",
+        "temperature",
+    );
+    copy_request_number_field_as(request, &mut generation_config, "top_p", "topP");
+    copy_request_number_field_as(request, &mut generation_config, "top_k", "topK");
+    if let Some(stop_sequences) = parse_openai_stop_sequences(request.get("stop")) {
+        generation_config.insert("stopSequences".to_string(), Value::Array(stop_sequences));
+    }
+    if let Some(reasoning_effort) = request.get("reasoning_effort").and_then(Value::as_str) {
+        if let Some(thinking_budget) =
+            map_openai_reasoning_effort_to_gemini_budget(reasoning_effort)
+        {
+            generation_config.insert(
+                "thinkingConfig".to_string(),
+                json!({
+                    "includeThoughts": true,
+                    "thinkingBudget": thinking_budget,
+                }),
+            );
+        }
+    }
+    if !generation_config.is_empty() {
+        output.insert(
+            "generationConfig".to_string(),
+            Value::Object(generation_config),
+        );
+    }
+    if let Some(tools) = convert_openai_tools_to_gemini(request.get("tools")) {
+        output.insert("tools".to_string(), tools);
+    }
+    if let Some(tool_config) = convert_openai_tool_choice_to_gemini(request.get("tool_choice")) {
+        output.insert("toolConfig".to_string(), tool_config);
+    }
+    if let Some(extra_body) = request.get("extra_body").and_then(Value::as_object) {
+        if let Some(google) = extra_body.get("google").and_then(Value::as_object) {
+            if let Some(existing) = output
+                .get_mut("generationConfig")
+                .and_then(Value::as_object_mut)
+            {
+                if let Some(response_modalities) = google.get("response_modalities").cloned() {
+                    existing.insert("responseModalities".to_string(), response_modalities);
+                }
+                if let Some(thinking_config) = google.get("thinking_config").cloned() {
+                    existing
+                        .entry("thinkingConfig".to_string())
+                        .or_insert(thinking_config);
+                }
+            }
+        }
+    }
+
+    Some(Value::Object(output))
+}
+
+fn convert_openai_content_to_gemini_parts(
+    content: Option<&Value>,
+    allow_images: bool,
+) -> Option<Vec<Value>> {
+    match content {
+        None | Some(Value::Null) => Some(Vec::new()),
+        Some(Value::String(text)) => {
+            let trimmed = text.trim();
+            if trimmed.is_empty() {
+                Some(Vec::new())
+            } else {
+                Some(vec![json!({ "text": text })])
+            }
+        }
+        Some(Value::Array(parts)) => {
+            let mut converted = Vec::new();
+            for part in parts {
+                let part_object = part.as_object()?;
+                let part_type = part_object
+                    .get("type")
+                    .and_then(Value::as_str)
+                    .unwrap_or_default();
+                match part_type {
+                    "text" | "input_text" => {
+                        if let Some(text) = part_object.get("text").and_then(Value::as_str) {
+                            if !text.trim().is_empty() {
+                                converted.push(json!({ "text": text }));
+                            }
+                        }
+                    }
+                    "image_url" | "input_image" if allow_images => return None,
+                    _ => {}
+                }
+            }
+            Some(converted)
+        }
+        _ => None,
+    }
+}
+
+fn convert_openai_tools_to_gemini(tools: Option<&Value>) -> Option<Value> {
+    let tool_values = tools?.as_array()?;
+    let mut declarations = Vec::new();
+    for tool in tool_values {
+        let tool_object = tool.as_object()?;
+        if tool_object
+            .get("type")
+            .and_then(Value::as_str)
+            .is_some_and(|value| value != "function")
+        {
+            continue;
+        }
+        let function = tool_object.get("function")?.as_object()?;
+        let name = function
+            .get("name")
+            .and_then(Value::as_str)
+            .map(str::trim)
+            .filter(|value| !value.is_empty())?;
+        let mut declaration = Map::new();
+        declaration.insert("name".to_string(), Value::String(name.to_string()));
+        if let Some(description) = function.get("description").cloned() {
+            declaration.insert("description".to_string(), description);
+        }
+        declaration.insert(
+            "parameters".to_string(),
+            function
+                .get("parameters")
+                .cloned()
+                .unwrap_or_else(|| json!({})),
+        );
+        declarations.push(Value::Object(declaration));
+    }
+    (!declarations.is_empty()).then(|| json!([{ "functionDeclarations": declarations }]))
+}
+
+fn convert_openai_tool_choice_to_gemini(tool_choice: Option<&Value>) -> Option<Value> {
+    let tool_choice = tool_choice?;
+    match tool_choice {
+        Value::String(value) => {
+            let mode = match value.trim().to_ascii_lowercase().as_str() {
+                "none" => "NONE",
+                "required" => "ANY",
+                "auto" => "AUTO",
+                _ => return None,
+            };
+            Some(json!({
+                "functionCallingConfig": {
+                    "mode": mode,
+                }
+            }))
+        }
+        Value::Object(object) => {
+            let function_name = object
+                .get("function")
+                .and_then(Value::as_object)
+                .and_then(|function| function.get("name"))
+                .and_then(Value::as_str)
+                .map(str::trim)
+                .filter(|value| !value.is_empty())?;
+            Some(json!({
+                "functionCallingConfig": {
+                    "mode": "ANY",
+                    "allowedFunctionNames": [function_name],
+                }
+            }))
+        }
+        _ => None,
+    }
+}
+
+fn compact_gemini_contents(contents: Vec<Value>) -> Vec<Value> {
+    let mut compact: Vec<Value> = Vec::new();
+    for content in contents {
+        let role = content
+            .get("role")
+            .and_then(Value::as_str)
+            .unwrap_or_default()
+            .to_string();
+        let parts = content
+            .get("parts")
+            .and_then(Value::as_array)
+            .cloned()
+            .unwrap_or_default();
+        if parts.is_empty() {
+            continue;
+        }
+        if let Some(last) = compact.last_mut() {
+            let last_role = last
+                .get("role")
+                .and_then(Value::as_str)
+                .unwrap_or_default()
+                .to_string();
+            if last_role == role {
+                if let Some(last_parts) = last.get_mut("parts").and_then(Value::as_array_mut) {
+                    last_parts.extend(parts);
+                }
+                continue;
+            }
+        }
+        compact.push(content);
+    }
+    compact
+}
--- a/apps/aether-gateway/src/ai_pipeline/conversion/request/from_openai_chat/mod.rs
+++ b/apps/aether-gateway/src/ai_pipeline/conversion/request/from_openai_chat/mod.rs
@@ -0,0 +1,8 @@
+mod claude;
+mod gemini;
+mod openai_cli;
+mod shared;
+
+pub(crate) use claude::convert_openai_chat_request_to_claude_request;
+pub(crate) use gemini::convert_openai_chat_request_to_gemini_request;
+pub(crate) use openai_cli::convert_openai_chat_request_to_openai_cli_request;
--- a/apps/aether-gateway/src/ai_pipeline/conversion/request/from_openai_chat/openai_cli.rs
+++ b/apps/aether-gateway/src/ai_pipeline/conversion/request/from_openai_chat/openai_cli.rs
@@ -0,0 +1,413 @@
+use std::collections::BTreeMap;
+
+use serde_json::{json, Map, Value};
+
+use super::super::to_openai_chat::extract_openai_text_content;
+use crate::gateway::ai_pipeline::planner::standard::copy_request_number_field;
+
+pub(crate) fn convert_openai_chat_request_to_openai_cli_request(
+    body_json: &Value,
+    mapped_model: &str,
+    upstream_is_stream: bool,
+    compact: bool,
+) -> Option<Value> {
+    let request = body_json.as_object()?;
+    let mut instructions = Vec::new();
+    let mut input_items = Vec::new();
+    let mut next_generated_tool_call_index = 0usize;
+    let mut tool_call_id_aliases = BTreeMap::new();
+
+    if let Some(message_values) = request.get("messages").and_then(Value::as_array) {
+        for message in message_values {
+            let message_object = message.as_object()?;
+            let role = message_object
+                .get("role")
+                .and_then(Value::as_str)
+                .unwrap_or_default()
+                .trim()
+                .to_ascii_lowercase();
+            match role.as_str() {
+                "system" | "developer" => {
+                    let text = extract_openai_text_content(message_object.get("content"))?;
+                    if !text.trim().is_empty() {
+                        instructions.push(text);
+                    }
+                }
+                "user" | "assistant" => {
+                    let content_items = convert_openai_content_to_openai_cli_items(
+                        message_object.get("content"),
+                        role.as_str(),
+                    )?;
+                    if !content_items.is_empty() {
+                        input_items.push(json!({
+                            "type": "message",
+                            "role": role,
+                            "content": content_items,
+                        }));
+                    }
+
+                    if role == "assistant" {
+                        if let Some(tool_calls) =
+                            message_object.get("tool_calls").and_then(Value::as_array)
+                        {
+                            for tool_call in tool_calls {
+                                let tool_call_object = tool_call.as_object()?;
+                                let function = tool_call_object.get("function")?.as_object()?;
+                                let tool_name = function
+                                    .get("name")
+                                    .and_then(Value::as_str)
+                                    .map(str::trim)
+                                    .filter(|value| !value.is_empty())?;
+                                let raw_call_id = tool_call_object
+                                    .get("id")
+                                    .and_then(Value::as_str)
+                                    .map(str::trim)
+                                    .unwrap_or_default();
+                                let call_id = if raw_call_id.is_empty() {
+                                    let generated =
+                                        format!("call_auto_{next_generated_tool_call_index}");
+                                    next_generated_tool_call_index += 1;
+                                    generated
+                                } else {
+                                    raw_call_id.to_string()
+                                };
+                                if !raw_call_id.is_empty() && raw_call_id != call_id {
+                                    tool_call_id_aliases
+                                        .insert(raw_call_id.to_string(), call_id.clone());
+                                }
+                                let arguments = function
+                                    .get("arguments")
+                                    .and_then(Value::as_str)
+                                    .map(ToOwned::to_owned)
+                                    .unwrap_or_else(|| "{}".to_string());
+                                input_items.push(json!({
+                                    "type": "function_call",
+                                    "call_id": call_id,
+                                    "name": tool_name,
+                                    "arguments": arguments,
+                                }));
+                            }
+                        }
+                    }
+                }
+                "tool" => {
+                    let raw_tool_call_id = message_object
+                        .get("tool_call_id")
+                        .and_then(Value::as_str)
+                        .map(str::trim)
+                        .unwrap_or_default();
+                    let tool_call_id = if raw_tool_call_id.is_empty() {
+                        let generated = format!("call_auto_{next_generated_tool_call_index}");
+                        next_generated_tool_call_index += 1;
+                        generated
+                    } else {
+                        tool_call_id_aliases
+                            .get(raw_tool_call_id)
+                            .cloned()
+                            .unwrap_or_else(|| raw_tool_call_id.to_string())
+                    };
+                    let output = match message_object.get("content") {
+                        Some(Value::String(text)) => text.clone(),
+                        Some(other) => serde_json::to_string(other).ok()?,
+                        None => String::new(),
+                    };
+                    input_items.push(json!({
+                        "type": "function_call_output",
+                        "call_id": tool_call_id,
+                        "output": output,
+                    }));
+                }
+                _ => {}
+            }
+        }
+    }
+
+    let mut output = Map::new();
+    output.insert("model".to_string(), Value::String(mapped_model.to_string()));
+    if !instructions.is_empty() {
+        output.insert(
+            "instructions".to_string(),
+            Value::String(
+                instructions
+                    .into_iter()
+                    .filter(|value: &String| !value.trim().is_empty())
+                    .collect::<Vec<_>>()
+                    .join("\n\n"),
+            ),
+        );
+    }
+    output.insert("input".to_string(), Value::Array(input_items));
+
+    if upstream_is_stream && !compact {
+        output.insert("stream".to_string(), Value::Bool(true));
+    }
+    if let Some(max_tokens) = request.get("max_tokens").and_then(Value::as_u64) {
+        output.insert("max_output_tokens".to_string(), Value::from(max_tokens));
+    }
+    copy_request_number_field(request, &mut output, "temperature");
+    copy_request_number_field(request, &mut output, "top_p");
+    copy_request_integer_field(request, &mut output, "top_logprobs");
+    copy_request_bool_field(request, &mut output, "parallel_tool_calls");
+
+    for passthrough_key in [
+        "prompt_cache_key",
+        "service_tier",
+        "metadata",
+        "store",
+        "previous_response_id",
+        "truncation",
+        "reasoning",
+        "stop",
+    ] {
+        if let Some(value) = request.get(passthrough_key) {
+            output.insert(passthrough_key.to_string(), value.clone());
+        }
+    }
+
+    if let Some(text) = build_openai_cli_text_config_from_openai_chat_request(request) {
+        output.insert("text".to_string(), Value::Object(text));
+    }
+    if let Some(tools) = build_openai_cli_tools_from_openai_chat_request(request) {
+        output.insert("tools".to_string(), Value::Array(tools));
+    }
+    if let Some(tool_choice) = build_openai_cli_tool_choice_from_openai_chat_request(request) {
+        output.insert("tool_choice".to_string(), tool_choice);
+    }
+
+    Some(Value::Object(output))
+}
+
+fn convert_openai_content_to_openai_cli_items(
+    content: Option<&Value>,
+    role: &str,
+) -> Option<Vec<Value>> {
+    let Some(content) = content else {
+        return Some(Vec::new());
+    };
+    match content {
+        Value::String(text) => {
+            if text.is_empty() {
+                Some(Vec::new())
+            } else {
+                Some(vec![json!({
+                    "type": if role == "assistant" { "output_text" } else { "input_text" },
+                    "text": text,
+                })])
+            }
+        }
+        Value::Array(parts) => {
+            let mut items = Vec::new();
+            for part in parts {
+                let part_object = part.as_object()?;
+                let part_type = part_object
+                    .get("type")
+                    .and_then(Value::as_str)
+                    .unwrap_or("text")
+                    .trim()
+                    .to_ascii_lowercase();
+                match part_type.as_str() {
+                    "text" | "input_text" | "output_text" => {
+                        if let Some(text) = part_object.get("text").and_then(Value::as_str) {
+                            if !text.is_empty() {
+                                items.push(json!({
+                                    "type": if role == "assistant" { "output_text" } else { "input_text" },
+                                    "text": text,
+                                }));
+                            }
+                        }
+                    }
+                    "image_url" => {
+                        let image_url = part_object
+                            .get("image_url")
+                            .and_then(Value::as_object)
+                            .and_then(|value| value.get("url"))
+                            .and_then(Value::as_str)
+                            .or_else(|| part_object.get("image_url").and_then(Value::as_str))?;
+                        items.push(json!({
+                            "type": if role == "assistant" { "output_image" } else { "input_image" },
+                            "image_url": image_url,
+                        }));
+                    }
+                    "input_image" | "output_image" => {
+                        let image_url = part_object
+                            .get("image_url")
+                            .and_then(Value::as_str)
+                            .or_else(|| part_object.get("url").and_then(Value::as_str))?;
+                        items.push(json!({
+                            "type": if role == "assistant" { "output_image" } else { "input_image" },
+                            "image_url": image_url,
+                        }));
+                    }
+                    _ => {}
+                }
+            }
+            Some(items)
+        }
+        _ => None,
+    }
+}
+
+fn build_openai_cli_text_config_from_openai_chat_request(
+    request: &Map<String, Value>,
+) -> Option<Map<String, Value>> {
+    let mut text = Map::new();
+    if let Some(response_format) = request.get("response_format") {
+        text.insert("format".to_string(), response_format.clone());
+    }
+    if let Some(verbosity) = request.get("verbosity") {
+        text.insert("verbosity".to_string(), verbosity.clone());
+    }
+    (!text.is_empty()).then_some(text)
+}
+
+fn build_openai_cli_tools_from_openai_chat_request(
+    request: &Map<String, Value>,
+) -> Option<Vec<Value>> {
+    let mut tools = Vec::new();
+    if let Some(tool_values) = request.get("tools").and_then(Value::as_array) {
+        for tool in tool_values {
+            let tool_object = tool.as_object()?;
+            let tool_type = tool_object
+                .get("type")
+                .and_then(Value::as_str)
+                .unwrap_or("function")
+                .trim()
+                .to_ascii_lowercase();
+            match tool_type.as_str() {
+                "function" => {
+                    let function = tool_object.get("function")?.as_object()?;
+                    let name = function
+                        .get("name")
+                        .and_then(Value::as_str)
+                        .map(str::trim)
+                        .filter(|value| !value.is_empty())?;
+                    let mut rebuilt = Map::new();
+                    rebuilt.insert("type".to_string(), Value::String("function".to_string()));
+                    rebuilt.insert("name".to_string(), Value::String(name.to_string()));
+                    if let Some(description) = function.get("description") {
+                        rebuilt.insert("description".to_string(), description.clone());
+                    }
+                    if let Some(parameters) = function.get("parameters") {
+                        rebuilt.insert("parameters".to_string(), parameters.clone());
+                    }
+                    tools.push(Value::Object(rebuilt));
+                }
+                "custom" => {
+                    let custom = tool_object.get("custom").and_then(Value::as_object)?;
+                    let mut rebuilt = Map::new();
+                    rebuilt.insert("type".to_string(), Value::String("custom".to_string()));
+                    if let Some(name) = custom.get("name") {
+                        rebuilt.insert("name".to_string(), name.clone());
+                    }
+                    if let Some(description) = custom.get("description") {
+                        rebuilt.insert("description".to_string(), description.clone());
+                    }
+                    if let Some(format) = custom.get("format") {
+                        rebuilt.insert("format".to_string(), format.clone());
+                    }
+                    tools.push(Value::Object(rebuilt));
+                }
+                _ => tools.push(tool.clone()),
+            }
+        }
+    }
+
+    if let Some(web_search_options) = request.get("web_search_options").and_then(Value::as_object) {
+        let mut tool = Map::new();
+        tool.insert("type".to_string(), Value::String("web_search".to_string()));
+        if let Some(user_location) = web_search_options
+            .get("user_location")
+            .and_then(Value::as_object)
+        {
+            if user_location.get("type").and_then(Value::as_str) == Some("approximate") {
+                if let Some(approximate) =
+                    user_location.get("approximate").and_then(Value::as_object)
+                {
+                    let mut flattened = Map::new();
+                    flattened.insert("type".to_string(), Value::String("approximate".to_string()));
+                    if let Some(country) = approximate.get("country") {
+                        flattened.insert("country".to_string(), country.clone());
+                    }
+                    if let Some(city) = approximate.get("city") {
+                        flattened.insert("city".to_string(), city.clone());
+                    }
+                    tool.insert("user_location".to_string(), Value::Object(flattened));
+                }
+            }
+        }
+        if let Some(search_context_size) = web_search_options.get("search_context_size") {
+            tool.insert(
+                "search_context_size".to_string(),
+                search_context_size.clone(),
+            );
+        }
+        tools.push(Value::Object(tool));
+    }
+
+    (!tools.is_empty()).then_some(tools)
+}
+
+fn build_openai_cli_tool_choice_from_openai_chat_request(
+    request: &Map<String, Value>,
+) -> Option<Value> {
+    let tool_choice = request.get("tool_choice")?;
+    match tool_choice {
+        Value::String(value) => Some(Value::String(value.clone())),
+        Value::Object(object) => {
+            let choice_type = object
+                .get("type")
+                .and_then(Value::as_str)
+                .unwrap_or_default()
+                .trim()
+                .to_ascii_lowercase();
+            match choice_type.as_str() {
+                "function" => {
+                    let function = object.get("function").and_then(Value::as_object)?;
+                    let name = function.get("name")?.as_str()?;
+                    Some(json!({
+                        "type": "function",
+                        "name": name,
+                    }))
+                }
+                "custom" => {
+                    let custom = object.get("custom").and_then(Value::as_object)?;
+                    let name = custom.get("name")?.as_str()?;
+                    Some(json!({
+                        "type": "custom",
+                        "name": name,
+                    }))
+                }
+                "allowed_tools" => {
+                    let allowed_tools = object.get("allowed_tools").and_then(Value::as_object)?;
+                    Some(json!({
+                        "type": "allowed_tools",
+                        "mode": allowed_tools.get("mode").cloned().unwrap_or_else(|| Value::String("auto".to_string())),
+                        "tools": allowed_tools.get("tools").cloned().unwrap_or_else(|| Value::Array(Vec::new())),
+                    }))
+                }
+                _ => Some(tool_choice.clone()),
+            }
+        }
+        _ => Some(tool_choice.clone()),
+    }
+}
+
+fn copy_request_integer_field(
+    request: &Map<String, Value>,
+    output: &mut Map<String, Value>,
+    field: &str,
+) {
+    if let Some(value) = request.get(field).and_then(Value::as_i64) {
+        output.insert(field.to_string(), Value::from(value));
+    }
+}
+
+fn copy_request_bool_field(
+    request: &Map<String, Value>,
+    output: &mut Map<String, Value>,
+    field: &str,
+) {
+    if let Some(value) = request.get(field).and_then(Value::as_bool) {
+        output.insert(field.to_string(), Value::Bool(value));
+    }
+}
--- a/apps/aether-gateway/src/ai_pipeline/conversion/request/from_openai_chat/shared.rs
+++ b/apps/aether-gateway/src/ai_pipeline/conversion/request/from_openai_chat/shared.rs
@@ -0,0 +1,21 @@
+use serde_json::{json, Value};
+
+pub(super) fn parse_openai_tool_arguments(arguments: Option<&Value>) -> Option<Value> {
+    match arguments {
+        Some(Value::Object(object)) => Some(Value::Object(object.clone())),
+        Some(Value::String(raw)) => {
+            let trimmed = raw.trim();
+            if trimmed.is_empty() {
+                Some(json!({}))
+            } else {
+                match serde_json::from_str::<Value>(trimmed) {
+                    Ok(Value::Object(object)) => Some(Value::Object(object)),
+                    Ok(other) => Some(json!({ "input": other })),
+                    Err(_) => Some(json!({ "input": trimmed })),
+                }
+            }
+        }
+        Some(other) => Some(json!({ "input": other })),
+        None => Some(json!({})),
+    }
+}
--- a/apps/aether-gateway/src/ai_pipeline/conversion/request/mod.rs
+++ b/apps/aether-gateway/src/ai_pipeline/conversion/request/mod.rs
@@ -0,0 +1,12 @@
+mod from_openai_chat;
+mod to_openai_chat;
+
+pub(crate) use from_openai_chat::{
+    convert_openai_chat_request_to_claude_request, convert_openai_chat_request_to_gemini_request,
+    convert_openai_chat_request_to_openai_cli_request,
+};
+pub(crate) use to_openai_chat::{
+    extract_openai_text_content, normalize_claude_request_to_openai_chat_request,
+    normalize_gemini_request_to_openai_chat_request,
+    normalize_openai_cli_request_to_openai_chat_request, parse_openai_tool_result_content,
+};
--- a/apps/aether-gateway/src/ai_pipeline/conversion/request/to_openai_chat/claude.rs
+++ b/apps/aether-gateway/src/ai_pipeline/conversion/request/to_openai_chat/claude.rs
@@ -0,0 +1,313 @@
+use serde_json::{json, Map, Value};
+use uuid::Uuid;
+
+use super::shared::canonical_json_string;
+
+pub(crate) fn normalize_claude_request_to_openai_chat_request(body_json: &Value) -> Option<Value> {
+    let request = body_json.as_object()?;
+    let mut output = Map::new();
+    if let Some(model) = request.get("model") {
+        output.insert("model".to_string(), model.clone());
+    }
+
+    let mut messages = Vec::new();
+    if let Some(system_text) = extract_claude_system_text(request.get("system")) {
+        if !system_text.trim().is_empty() {
+            messages.push(json!({
+                "role": "system",
+                "content": system_text,
+            }));
+        }
+    }
+
+    if let Some(message_values) = request.get("messages").and_then(Value::as_array) {
+        for message in message_values {
+            let message_object = message.as_object()?;
+            let role = message_object
+                .get("role")
+                .and_then(Value::as_str)
+                .unwrap_or_default()
+                .trim()
+                .to_ascii_lowercase();
+            match role.as_str() {
+                "user" => {
+                    let mut text_segments = Vec::new();
+                    if let Some(content) = message_object.get("content") {
+                        for block in normalize_claude_content_blocks(content)? {
+                            match block {
+                                ClaudeNormalizedBlock::Text(text) => {
+                                    if !text.trim().is_empty() {
+                                        text_segments.push(text);
+                                    }
+                                }
+                                ClaudeNormalizedBlock::ToolResult {
+                                    tool_use_id,
+                                    content,
+                                } => {
+                                    messages.push(json!({
+                                        "role": "tool",
+                                        "tool_call_id": tool_use_id,
+                                        "content": content,
+                                    }));
+                                }
+                                ClaudeNormalizedBlock::ToolUse { .. } => {}
+                            }
+                        }
+                    }
+                    let text = text_segments.join("\n\n");
+                    if !text.trim().is_empty() {
+                        messages.push(json!({
+                            "role": "user",
+                            "content": text,
+                        }));
+                    }
+                }
+                "assistant" => {
+                    let mut text_segments = Vec::new();
+                    let mut tool_calls = Vec::new();
+                    if let Some(content) = message_object.get("content") {
+                        for block in normalize_claude_content_blocks(content)? {
+                            match block {
+                                ClaudeNormalizedBlock::Text(text) => {
+                                    if !text.trim().is_empty() {
+                                        text_segments.push(text);
+                                    }
+                                }
+                                ClaudeNormalizedBlock::ToolUse { id, name, input } => {
+                                    tool_calls.push(json!({
+                                        "id": id.unwrap_or_else(|| format!("toolu_{}", Uuid::new_v4().simple())),
+                                        "type": "function",
+                                        "function": {
+                                            "name": name,
+                                            "arguments": canonical_json_string(input.unwrap_or(Value::Object(Map::new()))),
+                                        }
+                                    }));
+                                }
+                                ClaudeNormalizedBlock::ToolResult { .. } => {}
+                            }
+                        }
+                    }
+                    let mut assistant = Map::new();
+                    assistant.insert("role".to_string(), Value::String("assistant".to_string()));
+                    assistant.insert(
+                        "content".to_string(),
+                        if text_segments.is_empty() && !tool_calls.is_empty() {
+                            Value::Null
+                        } else {
+                            Value::String(text_segments.join("\n\n"))
+                        },
+                    );
+                    if !tool_calls.is_empty() {
+                        assistant.insert("tool_calls".to_string(), Value::Array(tool_calls));
+                    }
+                    messages.push(Value::Object(assistant));
+                }
+                _ => {}
+            }
+        }
+    }
+    output.insert("messages".to_string(), Value::Array(messages));
+
+    if let Some(max_tokens) = request.get("max_tokens").cloned() {
+        output.insert("max_completion_tokens".to_string(), max_tokens);
+    }
+    for passthrough_key in ["temperature", "top_p", "metadata", "stop", "stream"] {
+        if let Some(value) = request.get(passthrough_key) {
+            output.insert(passthrough_key.to_string(), value.clone());
+        }
+    }
+    if let Some(tools) = normalize_claude_tools_to_openai(request.get("tools"))? {
+        output.insert("tools".to_string(), Value::Array(tools));
+    }
+    if let Some(tool_choice) = normalize_claude_tool_choice_to_openai(request.get("tool_choice"))? {
+        output.insert("tool_choice".to_string(), tool_choice);
+    }
+
+    Some(Value::Object(output))
+}
+
+#[derive(Debug)]
+enum ClaudeNormalizedBlock {
+    Text(String),
+    ToolUse {
+        id: Option<String>,
+        name: String,
+        input: Option<Value>,
+    },
+    ToolResult {
+        tool_use_id: String,
+        content: Value,
+    },
+}
+
+fn normalize_claude_content_blocks(content: &Value) -> Option<Vec<ClaudeNormalizedBlock>> {
+    match content {
+        Value::String(text) => Some(vec![ClaudeNormalizedBlock::Text(text.clone())]),
+        Value::Array(blocks) => {
+            let mut normalized = Vec::new();
+            for block in blocks {
+                let block = block.as_object()?;
+                match block.get("type")?.as_str()? {
+                    "text" | "thinking" => {
+                        let text = block
+                            .get("text")
+                            .and_then(Value::as_str)
+                            .unwrap_or_default();
+                        normalized.push(ClaudeNormalizedBlock::Text(text.to_string()));
+                    }
+                    "tool_use" => {
+                        let name = block
+                            .get("name")
+                            .and_then(Value::as_str)
+                            .map(str::trim)
+                            .filter(|value| !value.is_empty())?
+                            .to_string();
+                        normalized.push(ClaudeNormalizedBlock::ToolUse {
+                            id: block
+                                .get("id")
+                                .and_then(Value::as_str)
+                                .filter(|value| !value.is_empty())
+                                .map(ToOwned::to_owned),
+                            name,
+                            input: block.get("input").cloned(),
+                        });
+                    }
+                    "tool_result" => {
+                        let tool_use_id = block
+                            .get("tool_use_id")
+                            .and_then(Value::as_str)
+                            .map(str::trim)
+                            .filter(|value| !value.is_empty())?
+                            .to_string();
+                        let content = block.get("content").cloned().unwrap_or(Value::Null);
+                        normalized.push(ClaudeNormalizedBlock::ToolResult {
+                            tool_use_id,
+                            content,
+                        });
+                    }
+                    _ => {}
+                }
+            }
+            Some(normalized)
+        }
+        _ => None,
+    }
+}
+
+fn extract_claude_system_text(system: Option<&Value>) -> Option<String> {
+    let system = system?;
+    let text = match system {
+        Value::String(text) => text.clone(),
+        Value::Array(blocks) => {
+            let mut segments = Vec::new();
+            for block in blocks {
+                let block = block.as_object()?;
+                if block.get("type").and_then(Value::as_str).unwrap_or("text") == "text" {
+                    let text = block
+                        .get("text")
+                        .and_then(Value::as_str)
+                        .unwrap_or_default();
+                    if !text.trim().is_empty() {
+                        segments.push(text.to_string());
+                    }
+                }
+            }
+            segments.join("\n\n")
+        }
+        _ => return None,
+    };
+    Some(strip_claude_billing_header(&text))
+}
+
+fn strip_claude_billing_header(text: &str) -> String {
+    let trimmed = text.trim();
+    let prefix = "x-anthropic-billing-header:";
+    if !trimmed.to_ascii_lowercase().starts_with(prefix) {
+        return trimmed.to_string();
+    }
+    let remainder = trimmed
+        .split_once('\n')
+        .map(|(_, rest)| rest.trim_start())
+        .unwrap_or_default();
+    remainder.trim_start_matches('\n').trim().to_string()
+}
+
+fn normalize_claude_tools_to_openai(tools: Option<&Value>) -> Option<Option<Vec<Value>>> {
+    let Some(tools) = tools else {
+        return Some(None);
+    };
+    let tools = tools.as_array()?;
+    let mut normalized = Vec::new();
+    for tool in tools {
+        let tool = tool.as_object()?;
+        let name = tool
+            .get("name")
+            .and_then(Value::as_str)
+            .map(str::trim)
+            .filter(|value| !value.is_empty())?;
+        let mut function = Map::new();
+        function.insert("name".to_string(), Value::String(name.to_string()));
+        if let Some(description) = tool.get("description").and_then(Value::as_str) {
+            if !description.trim().is_empty() {
+                function.insert(
+                    "description".to_string(),
+                    Value::String(description.trim().to_string()),
+                );
+            }
+        }
+        function.insert(
+            "parameters".to_string(),
+            tool.get("input_schema")
+                .cloned()
+                .unwrap_or_else(|| json!({"type": "object"})),
+        );
+        normalized.push(json!({
+            "type": "function",
+            "function": Value::Object(function),
+        }));
+    }
+    Some(Some(normalized))
+}
+
+fn normalize_claude_tool_choice_to_openai(tool_choice: Option<&Value>) -> Option<Option<Value>> {
+    let Some(tool_choice) = tool_choice else {
+        return Some(None);
+    };
+    match tool_choice {
+        Value::String(value) => match value.trim().to_ascii_lowercase().as_str() {
+            "auto" => Some(Some(Value::String("auto".to_string()))),
+            "any" => Some(Some(Value::String("required".to_string()))),
+            "none" => Some(Some(Value::String("none".to_string()))),
+            _ => Some(None),
+        },
+        Value::Object(value) => {
+            if let Some(name) = value.get("name").and_then(Value::as_str) {
+                return Some(Some(json!({
+                    "type": "function",
+                    "function": { "name": name }
+                })));
+            }
+            let kind = value
+                .get("type")
+                .and_then(Value::as_str)
+                .unwrap_or_default();
+            match kind.trim().to_ascii_lowercase().as_str() {
+                "auto" => Some(Some(Value::String("auto".to_string()))),
+                "any" => Some(Some(Value::String("required".to_string()))),
+                "none" => Some(Some(Value::String("none".to_string()))),
+                "tool" => value
+                    .get("name")
+                    .and_then(Value::as_str)
+                    .map(|name| {
+                        Some(json!({
+                            "type": "function",
+                            "function": { "name": name }
+                        }))
+                    })
+                    .or(Some(None)),
+                _ => Some(None),
+            }
+        }
+        _ => Some(None),
+    }
+}
--- a/apps/aether-gateway/src/ai_pipeline/conversion/request/to_openai_chat/gemini.rs
+++ b/apps/aether-gateway/src/ai_pipeline/conversion/request/to_openai_chat/gemini.rs
@@ -0,0 +1,292 @@
+use serde_json::{json, Map, Value};
+
+use super::shared::canonical_json_string;
+
+pub(crate) fn normalize_gemini_request_to_openai_chat_request(
+    body_json: &Value,
+    request_path: &str,
+) -> Option<Value> {
+    let request = body_json.as_object()?;
+    let mut output = Map::new();
+    if let Some(model) = request
+        .get("model")
+        .and_then(Value::as_str)
+        .map(str::trim)
+        .filter(|value| !value.is_empty())
+    {
+        output.insert("model".to_string(), Value::String(model.to_string()));
+    } else if let Some(model) = extract_gemini_model_from_path(request_path) {
+        output.insert("model".to_string(), Value::String(model));
+    }
+
+    let mut messages = Vec::new();
+    if let Some(system_text) = extract_gemini_system_text(
+        request
+            .get("systemInstruction")
+            .or_else(|| request.get("system_instruction")),
+    ) {
+        if !system_text.trim().is_empty() {
+            messages.push(json!({
+                "role": "system",
+                "content": system_text,
+            }));
+        }
+    }
+
+    if let Some(contents) = request.get("contents").and_then(Value::as_array) {
+        for content in contents {
+            let content_object = content.as_object()?;
+            let role = content_object
+                .get("role")
+                .and_then(Value::as_str)
+                .unwrap_or("user")
+                .trim()
+                .to_ascii_lowercase();
+            let parts = content_object.get("parts").and_then(Value::as_array)?;
+            match role.as_str() {
+                "model" => {
+                    let mut text_segments = Vec::new();
+                    let mut tool_calls = Vec::new();
+                    for (index, part) in parts.iter().enumerate() {
+                        let part = part.as_object()?;
+                        if let Some(text) = part.get("text").and_then(Value::as_str) {
+                            if !text.trim().is_empty() {
+                                text_segments.push(text.to_string());
+                            }
+                        } else if let Some(function_call) =
+                            part.get("functionCall").and_then(Value::as_object)
+                        {
+                            let name = function_call
+                                .get("name")
+                                .and_then(Value::as_str)
+                                .map(str::trim)
+                                .filter(|value| !value.is_empty())?;
+                            let id = function_call
+                                .get("id")
+                                .and_then(Value::as_str)
+                                .filter(|value| !value.is_empty())
+                                .map(ToOwned::to_owned)
+                                .unwrap_or_else(|| format!("toolu_{}_{}", name, index));
+                            tool_calls.push(json!({
+                                "id": id,
+                                "type": "function",
+                                "function": {
+                                    "name": name,
+                                    "arguments": canonical_json_string(function_call.get("args").cloned().unwrap_or(Value::Object(Map::new()))),
+                                }
+                            }));
+                        }
+                    }
+                    let mut assistant = Map::new();
+                    assistant.insert("role".to_string(), Value::String("assistant".to_string()));
+                    assistant.insert(
+                        "content".to_string(),
+                        if text_segments.is_empty() && !tool_calls.is_empty() {
+                            Value::Null
+                        } else {
+                            Value::String(text_segments.join("\n\n"))
+                        },
+                    );
+                    if !tool_calls.is_empty() {
+                        assistant.insert("tool_calls".to_string(), Value::Array(tool_calls));
+                    }
+                    messages.push(Value::Object(assistant));
+                }
+                _ => {
+                    let mut text_segments = Vec::new();
+                    for part in parts {
+                        let part = part.as_object()?;
+                        if let Some(text) = part.get("text").and_then(Value::as_str) {
+                            if !text.trim().is_empty() {
+                                text_segments.push(text.to_string());
+                            }
+                        } else if let Some(function_response) =
+                            part.get("functionResponse").and_then(Value::as_object)
+                        {
+                            let name = function_response
+                                .get("name")
+                                .and_then(Value::as_str)
+                                .unwrap_or("tool");
+                            let response_value = function_response
+                                .get("response")
+                                .cloned()
+                                .unwrap_or(Value::Object(Map::new()));
+                            let tool_call_id = function_response
+                                .get("id")
+                                .and_then(Value::as_str)
+                                .filter(|value| !value.is_empty())
+                                .map(ToOwned::to_owned)
+                                .unwrap_or_else(|| format!("toolu_{}", name));
+                            messages.push(json!({
+                                "role": "tool",
+                                "tool_call_id": tool_call_id,
+                                "content": response_value,
+                            }));
+                        }
+                    }
+                    let text = text_segments.join("\n\n");
+                    if !text.trim().is_empty() {
+                        messages.push(json!({
+                            "role": "user",
+                            "content": text,
+                        }));
+                    }
+                }
+            }
+        }
+    }
+    output.insert("messages".to_string(), Value::Array(messages));
+
+    let generation_config = request
+        .get("generationConfig")
+        .or_else(|| request.get("generation_config"))
+        .and_then(Value::as_object);
+    if let Some(generation_config) = generation_config {
+        if let Some(value) = generation_config.get("maxOutputTokens").cloned() {
+            output.insert("max_completion_tokens".to_string(), value);
+        }
+        if let Some(value) = generation_config.get("temperature").cloned() {
+            output.insert("temperature".to_string(), value);
+        }
+        if let Some(value) = generation_config.get("topP").cloned() {
+            output.insert("top_p".to_string(), value);
+        }
+        if let Some(value) = generation_config.get("candidateCount").cloned() {
+            output.insert("n".to_string(), value);
+        }
+        if let Some(value) = generation_config.get("stopSequences").cloned() {
+            output.insert("stop".to_string(), value);
+        }
+    }
+    if let Some(value) = request.get("stream").cloned() {
+        output.insert("stream".to_string(), value);
+    }
+    if let Some(tools) = normalize_gemini_tools_to_openai(request.get("tools"))? {
+        output.insert("tools".to_string(), Value::Array(tools));
+    }
+    if let Some(tool_choice) = normalize_gemini_tool_choice_to_openai(request.get("toolConfig"))? {
+        output.insert("tool_choice".to_string(), tool_choice);
+    }
+
+    Some(Value::Object(output))
+}
+
+fn extract_gemini_system_text(system_instruction: Option<&Value>) -> Option<String> {
+    let system_instruction = system_instruction?;
+    match system_instruction {
+        Value::String(text) => Some(text.trim().to_string()),
+        Value::Object(object) => {
+            let parts = object.get("parts")?.as_array()?;
+            let mut segments = Vec::new();
+            for part in parts {
+                let part = part.as_object()?;
+                if let Some(text) = part.get("text").and_then(Value::as_str) {
+                    if !text.trim().is_empty() {
+                        segments.push(text.to_string());
+                    }
+                }
+            }
+            Some(segments.join("\n\n"))
+        }
+        _ => None,
+    }
+}
+
+fn normalize_gemini_tools_to_openai(tools: Option<&Value>) -> Option<Option<Vec<Value>>> {
+    let Some(tools) = tools else {
+        return Some(None);
+    };
+    let tools = tools.as_array()?;
+    let mut normalized = Vec::new();
+    for tool in tools {
+        let tool = tool.as_object()?;
+        let declarations = tool
+            .get("functionDeclarations")
+            .or_else(|| tool.get("function_declarations"))
+            .and_then(Value::as_array);
+        let Some(declarations) = declarations else {
+            continue;
+        };
+        for declaration in declarations {
+            let declaration = declaration.as_object()?;
+            let name = declaration
+                .get("name")
+                .and_then(Value::as_str)
+                .map(str::trim)
+                .filter(|value| !value.is_empty())?;
+            let mut function = Map::new();
+            function.insert("name".to_string(), Value::String(name.to_string()));
+            if let Some(description) = declaration.get("description").and_then(Value::as_str) {
+                if !description.trim().is_empty() {
+                    function.insert(
+                        "description".to_string(),
+                        Value::String(description.trim().to_string()),
+                    );
+                }
+            }
+            function.insert(
+                "parameters".to_string(),
+                declaration
+                    .get("parameters")
+                    .cloned()
+                    .unwrap_or_else(|| json!({"type": "object"})),
+            );
+            normalized.push(json!({
+                "type": "function",
+                "function": Value::Object(function),
+            }));
+        }
+    }
+    Some(Some(normalized))
+}
+
+fn normalize_gemini_tool_choice_to_openai(tool_config: Option<&Value>) -> Option<Option<Value>> {
+    let Some(tool_config) = tool_config else {
+        return Some(None);
+    };
+    let tool_config = tool_config.as_object()?;
+    let function_config = tool_config
+        .get("functionCallingConfig")
+        .or_else(|| tool_config.get("function_calling_config"))
+        .and_then(Value::as_object)?;
+    let mode = function_config
+        .get("mode")
+        .and_then(Value::as_str)
+        .unwrap_or_default()
+        .trim()
+        .to_ascii_uppercase();
+    match mode.as_str() {
+        "NONE" => Some(Some(Value::String("none".to_string()))),
+        "AUTO" => Some(Some(Value::String("auto".to_string()))),
+        "ANY" | "REQUIRED" => Some(Some(Value::String("required".to_string()))),
+        _ => {
+            if let Some(name) = function_config
+                .get("allowedFunctionNames")
+                .or_else(|| function_config.get("allowed_function_names"))
+                .and_then(Value::as_array)
+                .and_then(|values| values.first())
+                .and_then(Value::as_str)
+            {
+                Some(Some(json!({
+                    "type": "function",
+                    "function": { "name": name }
+                })))
+            } else {
+                Some(None)
+            }
+        }
+    }
+}
+
+fn extract_gemini_model_from_path(path: &str) -> Option<String> {
+    let marker = "/models/";
+    let start = path.find(marker)? + marker.len();
+    let tail = &path[start..];
+    let end = tail.find(':').unwrap_or(tail.len());
+    let model = tail[..end].trim();
+    if model.is_empty() {
+        None
+    } else {
+        Some(model.to_string())
+    }
+}
--- a/apps/aether-gateway/src/ai_pipeline/conversion/request/to_openai_chat/mod.rs
+++ b/apps/aether-gateway/src/ai_pipeline/conversion/request/to_openai_chat/mod.rs
@@ -0,0 +1,9 @@
+mod claude;
+mod gemini;
+mod openai_cli;
+mod shared;
+
+pub(crate) use claude::normalize_claude_request_to_openai_chat_request;
+pub(crate) use gemini::normalize_gemini_request_to_openai_chat_request;
+pub(crate) use openai_cli::normalize_openai_cli_request_to_openai_chat_request;
+pub(crate) use shared::{extract_openai_text_content, parse_openai_tool_result_content};
--- a/apps/aether-gateway/src/ai_pipeline/conversion/request/to_openai_chat/openai_cli.rs
+++ b/apps/aether-gateway/src/ai_pipeline/conversion/request/to_openai_chat/openai_cli.rs
@@ -0,0 +1,310 @@
+use serde_json::{json, Map, Value};
+
+use super::shared::{extract_openai_text_content, parse_openai_tool_result_content};
+
+pub(crate) fn normalize_openai_cli_request_to_openai_chat_request(
+    body_json: &Value,
+) -> Option<Value> {
+    let request = body_json.as_object()?;
+    let mut output = Map::new();
+    if let Some(model) = request.get("model") {
+        output.insert("model".to_string(), model.clone());
+    }
+
+    let mut messages = Vec::new();
+    if let Some(instructions) = request.get("instructions") {
+        let text = extract_openai_text_content(Some(instructions))?;
+        if !text.trim().is_empty() {
+            messages.push(json!({
+                "role": "system",
+                "content": text,
+            }));
+        }
+    }
+    messages.extend(normalize_openai_cli_input_to_openai_chat_messages(
+        request.get("input"),
+    )?);
+    output.insert("messages".to_string(), Value::Array(messages));
+
+    if let Some(max_output_tokens) = request.get("max_output_tokens").cloned() {
+        output.insert("max_completion_tokens".to_string(), max_output_tokens);
+    }
+    for passthrough_key in [
+        "temperature",
+        "top_p",
+        "metadata",
+        "store",
+        "previous_response_id",
+        "service_tier",
+        "reasoning",
+        "stop",
+        "stream",
+    ] {
+        if let Some(value) = request.get(passthrough_key) {
+            output.insert(passthrough_key.to_string(), value.clone());
+        }
+    }
+    if let Some(tools) = normalize_openai_cli_tools_to_openai_chat(request.get("tools"))? {
+        output.insert("tools".to_string(), Value::Array(tools));
+    }
+    if let Some(tool_choice) =
+        normalize_openai_cli_tool_choice_to_openai_chat(request.get("tool_choice"))?
+    {
+        output.insert("tool_choice".to_string(), tool_choice);
+    }
+
+    Some(Value::Object(output))
+}
+
+fn normalize_openai_cli_input_to_openai_chat_messages(input: Option<&Value>) -> Option<Vec<Value>> {
+    let Some(input) = input else {
+        return Some(Vec::new());
+    };
+    match input {
+        Value::Null => Some(Vec::new()),
+        Value::String(text) => {
+            if text.trim().is_empty() {
+                Some(Vec::new())
+            } else {
+                Some(vec![json!({
+                    "role": "user",
+                    "content": text,
+                })])
+            }
+        }
+        Value::Array(items) => {
+            let mut messages = Vec::new();
+            let mut next_generated_tool_call_index = 0usize;
+            for item in items {
+                if let Some(item_text) = item.as_str() {
+                    if !item_text.trim().is_empty() {
+                        messages.push(json!({
+                            "role": "user",
+                            "content": item_text,
+                        }));
+                    }
+                    continue;
+                }
+                let item_object = item.as_object()?;
+                let item_type = item_object
+                    .get("type")
+                    .and_then(Value::as_str)
+                    .unwrap_or("message")
+                    .trim()
+                    .to_ascii_lowercase();
+                match item_type.as_str() {
+                    "message" => {
+                        let role = item_object
+                            .get("role")
+                            .and_then(Value::as_str)
+                            .unwrap_or("user")
+                            .trim()
+                            .to_ascii_lowercase();
+                        if role == "system" || role == "developer" {
+                            let text = extract_openai_text_content(item_object.get("content"))?;
+                            if !text.trim().is_empty() {
+                                messages.push(json!({
+                                    "role": "system",
+                                    "content": text,
+                                }));
+                            }
+                            continue;
+                        }
+                        let normalized_content =
+                            normalize_openai_cli_message_content(item_object.get("content"))?;
+                        messages.push(json!({
+                            "role": role,
+                            "content": normalized_content,
+                        }));
+                    }
+                    "function_call" => {
+                        let tool_name = item_object
+                            .get("name")
+                            .and_then(Value::as_str)
+                            .map(str::trim)
+                            .filter(|value| !value.is_empty())?;
+                        let call_id = item_object
+                            .get("call_id")
+                            .or_else(|| item_object.get("id"))
+                            .and_then(Value::as_str)
+                            .map(str::trim)
+                            .filter(|value| !value.is_empty())
+                            .map(ToOwned::to_owned)
+                            .unwrap_or_else(|| {
+                                let generated =
+                                    format!("call_auto_{next_generated_tool_call_index}");
+                                next_generated_tool_call_index += 1;
+                                generated
+                            });
+                        let arguments = item_object
+                            .get("arguments")
+                            .and_then(Value::as_str)
+                            .map(ToOwned::to_owned)
+                            .unwrap_or_else(|| "{}".to_string());
+                        messages.push(json!({
+                            "role": "assistant",
+                            "content": Value::Array(Vec::new()),
+                            "tool_calls": [{
+                                "id": call_id,
+                                "type": "function",
+                                "function": {
+                                    "name": tool_name,
+                                    "arguments": arguments,
+                                }
+                            }]
+                        }));
+                    }
+                    "function_call_output" => {
+                        let tool_call_id = item_object
+                            .get("call_id")
+                            .or_else(|| item_object.get("tool_call_id"))
+                            .or_else(|| item_object.get("id"))
+                            .and_then(Value::as_str)
+                            .map(str::trim)
+                            .filter(|value| !value.is_empty())
+                            .map(ToOwned::to_owned)
+                            .unwrap_or_else(|| {
+                                let generated =
+                                    format!("call_auto_{next_generated_tool_call_index}");
+                                next_generated_tool_call_index += 1;
+                                generated
+                            });
+                        messages.push(json!({
+                            "role": "tool",
+                            "tool_call_id": tool_call_id,
+                            "content": parse_openai_tool_result_content(item_object.get("output")),
+                        }));
+                    }
+                    _ => {}
+                }
+            }
+            Some(messages)
+        }
+        _ => None,
+    }
+}
+
+fn normalize_openai_cli_message_content(content: Option<&Value>) -> Option<Value> {
+    let Some(content) = content else {
+        return Some(Value::Array(Vec::new()));
+    };
+    match content {
+        Value::String(text) => Some(Value::String(text.clone())),
+        Value::Array(parts) => {
+            let mut normalized = Vec::new();
+            for part in parts {
+                let part_object = part.as_object()?;
+                let part_type = part_object
+                    .get("type")
+                    .and_then(Value::as_str)
+                    .unwrap_or_default()
+                    .trim()
+                    .to_ascii_lowercase();
+                match part_type.as_str() {
+                    "input_text" | "output_text" | "text" => {
+                        if let Some(text) = part_object.get("text").and_then(Value::as_str) {
+                            normalized.push(json!({
+                                "type": "text",
+                                "text": text,
+                            }));
+                        }
+                    }
+                    "input_image" | "output_image" | "image_url" => {
+                        let image_url = part_object
+                            .get("image_url")
+                            .and_then(|value| {
+                                value.as_str().map(ToOwned::to_owned).or_else(|| {
+                                    value
+                                        .as_object()
+                                        .and_then(|object| object.get("url"))
+                                        .and_then(Value::as_str)
+                                        .map(ToOwned::to_owned)
+                                })
+                            })
+                            .or_else(|| {
+                                part_object
+                                    .get("url")
+                                    .and_then(Value::as_str)
+                                    .map(ToOwned::to_owned)
+                            })?;
+                        normalized.push(json!({
+                            "type": "input_image",
+                            "image_url": image_url,
+                        }));
+                    }
+                    _ => {}
+                }
+            }
+            Some(Value::Array(normalized))
+        }
+        _ => Some(content.clone()),
+    }
+}
+
+fn normalize_openai_cli_tools_to_openai_chat(tools: Option<&Value>) -> Option<Option<Vec<Value>>> {
+    let Some(Value::Array(tool_values)) = tools else {
+        return Some(None);
+    };
+    let mut normalized = Vec::new();
+    for tool in tool_values {
+        let tool_object = tool.as_object()?;
+        let tool_type = tool_object
+            .get("type")
+            .and_then(Value::as_str)
+            .unwrap_or("function")
+            .trim()
+            .to_ascii_lowercase();
+        if tool_object.get("function").is_some() || tool_type != "function" {
+            normalized.push(tool.clone());
+            continue;
+        }
+        let name = tool_object
+            .get("name")
+            .and_then(Value::as_str)
+            .map(str::trim)
+            .filter(|value| !value.is_empty())?;
+        let mut function = Map::new();
+        function.insert("name".to_string(), Value::String(name.to_string()));
+        if let Some(description) = tool_object.get("description") {
+            function.insert("description".to_string(), description.clone());
+        }
+        if let Some(parameters) = tool_object.get("parameters") {
+            function.insert("parameters".to_string(), parameters.clone());
+        }
+        normalized.push(json!({
+            "type": "function",
+            "function": function,
+        }));
+    }
+    Some((!normalized.is_empty()).then_some(normalized))
+}
+
+fn normalize_openai_cli_tool_choice_to_openai_chat(
+    tool_choice: Option<&Value>,
+) -> Option<Option<Value>> {
+    let Some(tool_choice) = tool_choice else {
+        return Some(None);
+    };
+    match tool_choice {
+        Value::Object(object)
+            if object.get("function").is_none()
+                && object
+                    .get("type")
+                    .and_then(Value::as_str)
+                    .is_some_and(|value| value.eq_ignore_ascii_case("function")) =>
+        {
+            let name = object
+                .get("name")
+                .and_then(Value::as_str)
+                .map(str::trim)
+                .filter(|value| !value.is_empty())?;
+            Some(Some(json!({
+                "type": "function",
+                "function": {
+                    "name": name,
+                }
+            })))
+        }
+        _ => Some(Some(tool_choice.clone())),
+    }
+}
--- a/apps/aether-gateway/src/ai_pipeline/conversion/request/to_openai_chat/shared.rs
+++ b/apps/aether-gateway/src/ai_pipeline/conversion/request/to_openai_chat/shared.rs
@@ -0,0 +1,66 @@
+use serde_json::Value;
+
+pub(crate) fn extract_openai_text_content(content: Option<&Value>) -> Option<String> {
+    match content {
+        None | Some(Value::Null) => Some(String::new()),
+        Some(Value::String(text)) => Some(text.clone()),
+        Some(Value::Array(parts)) => {
+            let mut collected = Vec::new();
+            for part in parts {
+                let part_object = part.as_object()?;
+                let part_type = part_object
+                    .get("type")
+                    .and_then(Value::as_str)
+                    .unwrap_or_default();
+                if matches!(part_type, "text" | "input_text") {
+                    if let Some(text) = part_object.get("text").and_then(Value::as_str) {
+                        if !text.trim().is_empty() {
+                            collected.push(text.to_string());
+                        }
+                    }
+                }
+            }
+            Some(collected.join("\n"))
+        }
+        _ => None,
+    }
+}
+
+pub(crate) fn parse_openai_tool_result_content(content: Option<&Value>) -> Value {
+    match content {
+        Some(Value::String(raw)) => {
+            let trimmed = raw.trim();
+            if trimmed.is_empty() {
+                Value::String(String::new())
+            } else {
+                serde_json::from_str::<Value>(trimmed)
+                    .unwrap_or_else(|_| Value::String(raw.clone()))
+            }
+        }
+        Some(Value::Array(parts)) => {
+            let texts = parts
+                .iter()
+                .filter_map(|part| {
+                    part.as_object()
+                        .and_then(|object| object.get("text"))
+                        .and_then(Value::as_str)
+                        .map(ToOwned::to_owned)
+                })
+                .collect::<Vec<_>>();
+            if texts.is_empty() {
+                Value::Array(parts.clone())
+            } else {
+                Value::String(texts.join("\n"))
+            }
+        }
+        Some(value) => value.clone(),
+        None => Value::String(String::new()),
+    }
+}
+
+pub(super) fn canonical_json_string(value: Value) -> String {
+    match value {
+        Value::String(text) => text,
+        other => serde_json::to_string(&other).unwrap_or_else(|_| "null".to_string()),
+    }
+}
--- a/apps/aether-gateway/src/ai_pipeline/conversion/response/from_openai_chat/claude_chat.rs
+++ b/apps/aether-gateway/src/ai_pipeline/conversion/response/from_openai_chat/claude_chat.rs
@@ -0,0 +1,95 @@
+use serde_json::{json, Value};
+
+use super::shared::{
+    build_generated_tool_call_id, extract_openai_assistant_text, parse_openai_function_arguments,
+};
+
+pub(crate) fn convert_openai_chat_response_to_claude_chat(
+    body_json: &Value,
+    report_context: &Value,
+) -> Option<Value> {
+    let body = body_json.as_object()?;
+    let choices = body.get("choices")?.as_array()?;
+    let first_choice = choices.first()?.as_object()?;
+    let message = first_choice.get("message")?.as_object()?;
+    let mut content = Vec::new();
+
+    if let Some(text) = extract_openai_assistant_text(message.get("content")) {
+        if !text.trim().is_empty() {
+            content.push(json!({
+                "type": "text",
+                "text": text,
+            }));
+        }
+    }
+    if let Some(tool_call_values) = message.get("tool_calls").and_then(Value::as_array) {
+        for (index, tool_call) in tool_call_values.iter().enumerate() {
+            let tool_call = tool_call.as_object()?;
+            let function = tool_call.get("function")?.as_object()?;
+            let tool_name = function
+                .get("name")
+                .and_then(Value::as_str)
+                .map(str::trim)
+                .filter(|value| !value.is_empty())?;
+            let tool_id = tool_call
+                .get("id")
+                .and_then(Value::as_str)
+                .filter(|value| !value.is_empty())
+                .map(ToOwned::to_owned)
+                .unwrap_or_else(|| build_generated_tool_call_id(index));
+            let input = parse_openai_function_arguments(function.get("arguments"))?;
+            content.push(json!({
+                "type": "tool_use",
+                "id": tool_id,
+                "name": tool_name,
+                "input": input,
+            }));
+        }
+    }
+    if content.is_empty() {
+        content.push(json!({
+            "type": "text",
+            "text": "",
+        }));
+    }
+
+    let stop_reason = match first_choice.get("finish_reason").and_then(Value::as_str) {
+        Some("stop") | None => "end_turn",
+        Some("length") => "max_tokens",
+        Some("tool_calls") | Some("function_call") => "tool_use",
+        Some("content_filter") => "content_filtered",
+        Some(other) => other,
+    };
+    let usage = body.get("usage").and_then(Value::as_object);
+    let input_tokens = usage
+        .and_then(|value| value.get("prompt_tokens"))
+        .and_then(Value::as_u64)
+        .unwrap_or(0);
+    let output_tokens = usage
+        .and_then(|value| value.get("completion_tokens"))
+        .and_then(Value::as_u64)
+        .unwrap_or(0);
+    let model = body
+        .get("model")
+        .and_then(Value::as_str)
+        .or_else(|| report_context.get("mapped_model").and_then(Value::as_str))
+        .or_else(|| report_context.get("model").and_then(Value::as_str))
+        .unwrap_or("unknown");
+    let id = body
+        .get("id")
+        .and_then(Value::as_str)
+        .unwrap_or("msg-local-finalize");
+
+    Some(json!({
+        "id": id,
+        "type": "message",
+        "role": "assistant",
+        "model": model,
+        "content": content,
+        "stop_reason": stop_reason,
+        "usage": {
+            "input_tokens": input_tokens,
+            "output_tokens": output_tokens,
+        }
+    }))
+}
--- a/apps/aether-gateway/src/ai_pipeline/conversion/response/from_openai_chat/gemini_chat.rs
+++ b/apps/aether-gateway/src/ai_pipeline/conversion/response/from_openai_chat/gemini_chat.rs
@@ -0,0 +1,101 @@
+use serde_json::{json, Value};
+
+use super::shared::{
+    build_generated_tool_call_id, extract_openai_assistant_text, parse_openai_function_arguments,
+};
+
+pub(crate) fn convert_openai_chat_response_to_gemini_chat(
+    body_json: &Value,
+    report_context: &Value,
+) -> Option<Value> {
+    let body = body_json.as_object()?;
+    let choices = body.get("choices")?.as_array()?;
+    let first_choice = choices.first()?.as_object()?;
+    let message = first_choice.get("message")?.as_object()?;
+    let mut parts = Vec::new();
+
+    if let Some(text) = extract_openai_assistant_text(message.get("content")) {
+        if !text.trim().is_empty() {
+            parts.push(json!({ "text": text }));
+        }
+    }
+    if let Some(tool_call_values) = message.get("tool_calls").and_then(Value::as_array) {
+        for (index, tool_call) in tool_call_values.iter().enumerate() {
+            let tool_call = tool_call.as_object()?;
+            let function = tool_call.get("function")?.as_object()?;
+            let tool_name = function
+                .get("name")
+                .and_then(Value::as_str)
+                .map(str::trim)
+                .filter(|value| !value.is_empty())?;
+            let call_id = tool_call
+                .get("id")
+                .and_then(Value::as_str)
+                .filter(|value| !value.is_empty())
+                .map(ToOwned::to_owned)
+                .unwrap_or_else(|| build_generated_tool_call_id(index));
+            parts.push(json!({
+                "functionCall": {
+                    "id": call_id,
+                    "name": tool_name,
+                    "args": parse_openai_function_arguments(function.get("arguments"))?,
+                }
+            }));
+        }
+    }
+    if parts.is_empty() {
+        parts.push(json!({ "text": "" }));
+    }
+
+    let usage = body.get("usage").and_then(Value::as_object);
+    let prompt_tokens = usage
+        .and_then(|value| value.get("prompt_tokens"))
+        .and_then(Value::as_u64)
+        .unwrap_or(0);
+    let completion_tokens = usage
+        .and_then(|value| value.get("completion_tokens"))
+        .and_then(Value::as_u64)
+        .unwrap_or(0);
+    let total_tokens = usage
+        .and_then(|value| value.get("total_tokens"))
+        .and_then(Value::as_u64)
+        .unwrap_or(prompt_tokens + completion_tokens);
+    let mut finish_reason = match first_choice.get("finish_reason").and_then(Value::as_str) {
+        Some("stop") | None => "STOP",
+        Some("length") => "MAX_TOKENS",
+        Some("content_filter") => "SAFETY",
+        Some("tool_calls") | Some("function_call") => "STOP",
+        Some(other) => other,
+    };
+    if parts.iter().any(|part| part.get("functionCall").is_some()) {
+        finish_reason = "STOP";
+    }
+    let model = body
+        .get("model")
+        .and_then(Value::as_str)
+        .or_else(|| report_context.get("mapped_model").and_then(Value::as_str))
+        .or_else(|| report_context.get("model").and_then(Value::as_str))
+        .unwrap_or("unknown");
+    let response_id = body
+        .get("id")
+        .and_then(Value::as_str)
+        .unwrap_or("resp-local-finalize");
+
+    Some(json!({
+        "responseId": response_id,
+        "modelVersion": model,
+        "candidates": [{
+            "content": {
+                "role": "model",
+                "parts": parts,
+            },
+            "finishReason": finish_reason,
+            "index": 0,
+        }],
+        "usageMetadata": {
+            "promptTokenCount": prompt_tokens,
+            "candidatesTokenCount": completion_tokens,
+            "totalTokenCount": total_tokens,
+        }
+    }))
+}
--- a/apps/aether-gateway/src/ai_pipeline/conversion/response/from_openai_chat/mod.rs
+++ b/apps/aether-gateway/src/ai_pipeline/conversion/response/from_openai_chat/mod.rs
@@ -0,0 +1,9 @@
+mod claude_chat;
+mod gemini_chat;
+mod openai_cli;
+mod shared;
+
+pub(crate) use claude_chat::convert_openai_chat_response_to_claude_chat;
+pub(crate) use gemini_chat::convert_openai_chat_response_to_gemini_chat;
+pub(crate) use openai_cli::convert_openai_chat_response_to_openai_cli;
+pub(crate) use shared::build_openai_cli_response;
--- a/apps/aether-gateway/src/ai_pipeline/conversion/response/from_openai_chat/openai_cli.rs
+++ b/apps/aether-gateway/src/ai_pipeline/conversion/response/from_openai_chat/openai_cli.rs
@@ -0,0 +1,99 @@
+use serde_json::{json, Value};
+
+use super::shared::{
+    build_openai_cli_response, canonicalize_tool_arguments,
+};
+
+pub(crate) fn convert_openai_chat_response_to_openai_cli(
+    body_json: &Value,
+    report_context: &Value,
+    compact: bool,
+) -> Option<Value> {
+    let body = body_json.as_object()?;
+    let choices = body.get("choices")?.as_array()?;
+    let first_choice = choices.first()?.as_object()?;
+    let message = first_choice.get("message")?.as_object()?;
+    let mut text = String::new();
+    match message.get("content") {
+        Some(Value::String(value)) => text.push_str(value),
+        Some(Value::Array(parts)) => {
+            for part in parts {
+                let part = part.as_object()?;
+                let part_type = part
+                    .get("type")
+                    .and_then(Value::as_str)
+                    .unwrap_or_default()
+                    .trim()
+                    .to_ascii_lowercase();
+                if matches!(part_type.as_str(), "text" | "output_text") {
+                    if let Some(piece) = part.get("text").and_then(Value::as_str) {
+                        text.push_str(piece);
+                    }
+                }
+            }
+        }
+        Some(Value::Null) | None => {}
+        _ => return None,
+    }
+
+    let mut function_calls = Vec::new();
+    if let Some(tool_call_values) = message.get("tool_calls").and_then(Value::as_array) {
+        for tool_call in tool_call_values {
+            let tool_call = tool_call.as_object()?;
+            let function = tool_call.get("function")?.as_object()?;
+            let tool_name = function
+                .get("name")
+                .and_then(Value::as_str)
+                .map(str::trim)
+                .filter(|value| !value.is_empty())?;
+            function_calls.push(json!({
+                "type": "function_call",
+                "id": tool_call.get("id").cloned().unwrap_or(Value::Null),
+                "call_id": tool_call.get("id").cloned().unwrap_or(Value::Null),
+                "name": tool_name,
+                "arguments": canonicalize_tool_arguments(function.get("arguments").cloned()),
+            }));
+        }
+    }
+
+    let usage = body.get("usage").and_then(Value::as_object);
+    let prompt_tokens = usage
+        .and_then(|value| value.get("prompt_tokens"))
+        .and_then(Value::as_u64)
+        .unwrap_or(0);
+    let output_tokens = usage
+        .and_then(|value| value.get("completion_tokens"))
+        .and_then(Value::as_u64)
+        .unwrap_or(0);
+    let total_tokens = usage
+        .and_then(|value| value.get("total_tokens"))
+        .and_then(Value::as_u64)
+        .unwrap_or(prompt_tokens + output_tokens);
+    let response_id = if compact {
+        body.get("id")
+            .and_then(Value::as_str)
+            .map(|value| value.replace("chatcmpl", "resp"))
+            .unwrap_or_else(|| "resp-local-finalize".to_string())
+    } else {
+        body.get("id")
+            .and_then(Value::as_str)
+            .map(|value| value.replace("chatcmpl", "resp"))
+            .unwrap_or_else(|| "resp-local-finalize".to_string())
+    };
+    let model = body
+        .get("model")
+        .and_then(Value::as_str)
+        .or_else(|| report_context.get("mapped_model").and_then(Value::as_str))
+        .or_else(|| report_context.get("model").and_then(Value::as_str))
+        .unwrap_or("unknown");
+
+    Some(build_openai_cli_response(
+        &response_id,
+        model,
+        &text,
+        function_calls,
+        prompt_tokens,
+        output_tokens,
+        total_tokens,
+    ))
+}
--- a/apps/aether-gateway/src/ai_pipeline/conversion/response/from_openai_chat/shared.rs
+++ b/apps/aether-gateway/src/ai_pipeline/conversion/response/from_openai_chat/shared.rs
@@ -0,0 +1,86 @@
+use serde_json::{json, Map, Value};
+
+pub(crate) fn build_openai_cli_response(
+    response_id: &str,
+    model: &str,
+    text: &str,
+    function_calls: Vec<Value>,
+    prompt_tokens: u64,
+    output_tokens: u64,
+    total_tokens: u64,
+) -> Value {
+    let mut output = Vec::new();
+    if !text.is_empty() {
+        output.push(json!({
+            "type": "message",
+            "id": format!("{response_id}_msg"),
+            "role": "assistant",
+            "status": "completed",
+            "content": [{
+                "type": "output_text",
+                "text": text,
+                "annotations": []
+            }]
+        }));
+    }
+    output.extend(function_calls);
+    json!({
+        "id": response_id,
+        "object": "response",
+        "status": "completed",
+        "model": model,
+        "output": output,
+        "usage": {
+            "input_tokens": prompt_tokens,
+            "output_tokens": output_tokens,
+            "total_tokens": total_tokens,
+        }
+    })
+}
+
+pub(super) fn extract_openai_assistant_text(content: Option<&Value>) -> Option<String> {
+    match content? {
+        Value::Null => Some(String::new()),
+        Value::String(text) => Some(text.clone()),
+        Value::Array(parts) => {
+            let mut text = String::new();
+            for part in parts {
+                let part = part.as_object()?;
+                let part_type = part
+                    .get("type")
+                    .and_then(Value::as_str)
+                    .unwrap_or_default()
+                    .trim()
+                    .to_ascii_lowercase();
+                if matches!(part_type.as_str(), "text" | "output_text") {
+                    if let Some(piece) = part.get("text").and_then(Value::as_str) {
+                        text.push_str(piece);
+                    }
+                }
+            }
+            Some(text)
+        }
+        _ => None,
+    }
+}
+
+pub(super) fn parse_openai_function_arguments(arguments: Option<&Value>) -> Option<Value> {
+    match arguments.cloned().unwrap_or(Value::Object(Map::new())) {
+        Value::String(text) => serde_json::from_str(&text)
+            .ok()
+            .or(Some(Value::String(text))),
+        other => Some(other),
+    }
+}
+
+pub(super) fn build_generated_tool_call_id(index: usize) -> String {
+    format!("call_auto_{index}")
+}
+
+pub(super) fn canonicalize_tool_arguments(value: Option<Value>) -> String {
+    match value {
+        Some(Value::String(text)) => text,
+        Some(other) => serde_json::to_string(&other).unwrap_or_else(|_| "null".to_string()),
+        None => "{}".to_string(),
+    }
+}
--- a/apps/aether-gateway/src/ai_pipeline/conversion/response/mod.rs
+++ b/apps/aether-gateway/src/ai_pipeline/conversion/response/mod.rs
@@ -0,0 +1,12 @@
+mod from_openai_chat;
+mod to_openai_chat;
+
+pub(crate) use from_openai_chat::{
+    build_openai_cli_response, convert_openai_chat_response_to_claude_chat,
+    convert_openai_chat_response_to_gemini_chat, convert_openai_chat_response_to_openai_cli,
+};
+pub(crate) use to_openai_chat::{
+    convert_claude_chat_response_to_openai_chat, convert_claude_cli_response_to_openai_cli,
+    convert_gemini_chat_response_to_openai_chat, convert_gemini_cli_response_to_openai_cli,
+    convert_openai_cli_response_to_openai_chat,
+};
--- a/apps/aether-gateway/src/ai_pipeline/conversion/response/to_openai_chat/claude_chat.rs
+++ b/apps/aether-gateway/src/ai_pipeline/conversion/response/to_openai_chat/claude_chat.rs
@@ -0,0 +1,96 @@
+use serde_json::{json, Map, Value};
+
+use super::shared::{build_generated_tool_call_id, canonicalize_tool_arguments};
+
+pub(crate) fn convert_claude_chat_response_to_openai_chat(
+    body_json: &Value,
+    report_context: &Value,
+) -> Option<Value> {
+    let body = body_json.as_object()?;
+    let content = body.get("content")?.as_array()?;
+    let mut text = String::new();
+    let mut tool_calls = Vec::new();
+    for (index, block) in content.iter().enumerate() {
+        let block = block.as_object()?;
+        match block.get("type")?.as_str()? {
+            "text" => {
+                text.push_str(block.get("text")?.as_str()?);
+            }
+            "tool_use" => {
+                let tool_name = block.get("name")?.as_str()?;
+                let tool_id = block
+                    .get("id")
+                    .and_then(Value::as_str)
+                    .filter(|value| !value.is_empty())
+                    .map(ToOwned::to_owned)
+                    .unwrap_or_else(|| build_generated_tool_call_id(index));
+                let arguments = canonicalize_tool_arguments(block.get("input").cloned());
+                tool_calls.push(json!({
+                    "id": tool_id,
+                    "type": "function",
+                    "function": {
+                        "name": tool_name,
+                        "arguments": arguments,
+                    }
+                }));
+            }
+            _ => return None,
+        }
+    }
+    let mut finish_reason = match body.get("stop_reason").and_then(Value::as_str) {
+        Some("end_turn") | Some("stop_sequence") => Some("stop"),
+        Some("max_tokens") => Some("length"),
+        Some("tool_use") => Some("tool_calls"),
+        Some(other) if !other.is_empty() => Some(other),
+        _ => None,
+    };
+    if !tool_calls.is_empty() && finish_reason.is_none_or(|reason| reason == "stop") {
+        finish_reason = Some("tool_calls");
+    }
+    let usage = body.get("usage").and_then(Value::as_object);
+    let prompt_tokens = usage
+        .and_then(|value| value.get("input_tokens"))
+        .and_then(Value::as_u64)
+        .unwrap_or(0);
+    let completion_tokens = usage
+        .and_then(|value| value.get("output_tokens"))
+        .and_then(Value::as_u64)
+        .unwrap_or(0);
+    let total_tokens = prompt_tokens + completion_tokens;
+    let model = body
+        .get("model")
+        .and_then(Value::as_str)
+        .or_else(|| report_context.get("mapped_model").and_then(Value::as_str))
+        .or_else(|| report_context.get("model").and_then(Value::as_str))
+        .unwrap_or("unknown");
+    let id = body
+        .get("id")
+        .and_then(Value::as_str)
+        .unwrap_or("chatcmpl-local-finalize");
+    let message_content = if text.is_empty() && !tool_calls.is_empty() {
+        Value::Null
+    } else {
+        Value::String(text)
+    };
+    let mut message = Map::new();
+    message.insert("role".to_string(), Value::String("assistant".to_string()));
+    message.insert("content".to_string(), message_content);
+    if !tool_calls.is_empty() {
+        message.insert("tool_calls".to_string(), Value::Array(tool_calls));
+    }
+    Some(json!({
+        "id": id,
+        "object": "chat.completion",
+        "model": model,
+        "choices": [{
+            "index": 0,
+            "message": Value::Object(message),
+            "finish_reason": finish_reason,
+        }],
+        "usage": {
+            "prompt_tokens": prompt_tokens,
+            "completion_tokens": completion_tokens,
+            "total_tokens": total_tokens,
+        }
+    }))
+}
--- a/apps/aether-gateway/src/ai_pipeline/conversion/response/to_openai_chat/claude_cli.rs
+++ b/apps/aether-gateway/src/ai_pipeline/conversion/response/to_openai_chat/claude_cli.rs
@@ -0,0 +1,70 @@
+use serde_json::{json, Value};
+
+use super::shared::{build_generated_tool_call_id, canonicalize_tool_arguments};
+use super::super::from_openai_chat::build_openai_cli_response;
+
+pub(crate) fn convert_claude_cli_response_to_openai_cli(
+    body_json: &Value,
+    report_context: &Value,
+) -> Option<Value> {
+    let body = body_json.as_object()?;
+    let content = body.get("content")?.as_array()?;
+    let mut text = String::new();
+    let mut function_calls = Vec::new();
+    for (index, block) in content.iter().enumerate() {
+        let block = block.as_object()?;
+        match block.get("type")?.as_str()? {
+            "text" => {
+                text.push_str(block.get("text")?.as_str()?);
+            }
+            "tool_use" => {
+                let tool_name = block.get("name")?.as_str()?;
+                let call_id = block
+                    .get("id")
+                    .and_then(Value::as_str)
+                    .filter(|value| !value.is_empty())
+                    .map(ToOwned::to_owned)
+                    .unwrap_or_else(|| build_generated_tool_call_id(index));
+                let arguments = canonicalize_tool_arguments(block.get("input").cloned());
+                function_calls.push(json!({
+                    "type": "function_call",
+                    "call_id": call_id,
+                    "name": tool_name,
+                    "arguments": arguments,
+                }));
+            }
+            _ => return None,
+        }
+    }
+
+    let usage = body.get("usage").and_then(Value::as_object);
+    let prompt_tokens = usage
+        .and_then(|value| value.get("input_tokens"))
+        .and_then(Value::as_u64)
+        .unwrap_or(0);
+    let output_tokens = usage
+        .and_then(|value| value.get("output_tokens"))
+        .and_then(Value::as_u64)
+        .unwrap_or(0);
+    let total_tokens = prompt_tokens + output_tokens;
+    let model = body
+        .get("model")
+        .and_then(Value::as_str)
+        .or_else(|| report_context.get("mapped_model").and_then(Value::as_str))
+        .or_else(|| report_context.get("model").and_then(Value::as_str))
+        .unwrap_or("unknown");
+    let response_id = body
+        .get("id")
+        .and_then(Value::as_str)
+        .unwrap_or("resp-local-finalize");
+
+    Some(build_openai_cli_response(
+        response_id,
+        model,
+        &text,
+        function_calls,
+        prompt_tokens,
+        output_tokens,
+        total_tokens,
+    ))
+}
--- a/apps/aether-gateway/src/ai_pipeline/conversion/response/to_openai_chat/gemini_chat.rs
+++ b/apps/aether-gateway/src/ai_pipeline/conversion/response/to_openai_chat/gemini_chat.rs
@@ -0,0 +1,100 @@
+use serde_json::{json, Map, Value};
+
+use super::shared::{build_generated_tool_call_id, canonicalize_tool_arguments};
+
+pub(crate) fn convert_gemini_chat_response_to_openai_chat(
+    body_json: &Value,
+    report_context: &Value,
+) -> Option<Value> {
+    let body = body_json.as_object()?;
+    let candidates = body.get("candidates")?.as_array()?;
+    let first_candidate = candidates.first()?.as_object()?;
+    let content = first_candidate.get("content")?.as_object()?;
+    let parts = content.get("parts")?.as_array()?;
+    let mut text = String::new();
+    let mut tool_calls = Vec::new();
+    for (index, part) in parts.iter().enumerate() {
+        let part = part.as_object()?;
+        if let Some(piece) = part.get("text").and_then(Value::as_str) {
+            text.push_str(piece);
+        } else if let Some(function_call) = part.get("functionCall").and_then(Value::as_object) {
+            let tool_name = function_call.get("name")?.as_str()?;
+            let tool_id = function_call
+                .get("id")
+                .and_then(Value::as_str)
+                .filter(|value| !value.is_empty())
+                .map(ToOwned::to_owned)
+                .unwrap_or_else(|| build_generated_tool_call_id(index));
+            let arguments = canonicalize_tool_arguments(function_call.get("args").cloned());
+            tool_calls.push(json!({
+                "id": tool_id,
+                "type": "function",
+                "function": {
+                    "name": tool_name,
+                    "arguments": arguments,
+                }
+            }));
+        } else {
+            return None;
+        }
+    }
+    let mut finish_reason = match first_candidate.get("finishReason").and_then(Value::as_str) {
+        Some("STOP") => Some("stop"),
+        Some("MAX_TOKENS") => Some("length"),
+        Some("SAFETY") => Some("content_filter"),
+        Some(other) if !other.is_empty() => Some(other),
+        _ => None,
+    };
+    if !tool_calls.is_empty() && finish_reason.is_none_or(|reason| reason == "stop") {
+        finish_reason = Some("tool_calls");
+    }
+    let usage = body.get("usageMetadata").and_then(Value::as_object);
+    let prompt_tokens = usage
+        .and_then(|value| value.get("promptTokenCount"))
+        .and_then(Value::as_u64)
+        .unwrap_or(0);
+    let completion_tokens = usage
+        .and_then(|value| value.get("candidatesTokenCount"))
+        .and_then(Value::as_u64)
+        .unwrap_or(0);
+    let total_tokens = usage
+        .and_then(|value| value.get("totalTokenCount"))
+        .and_then(Value::as_u64)
+        .unwrap_or(prompt_tokens + completion_tokens);
+    let model = body
+        .get("modelVersion")
+        .and_then(Value::as_str)
+        .or_else(|| report_context.get("mapped_model").and_then(Value::as_str))
+        .or_else(|| report_context.get("model").and_then(Value::as_str))
+        .unwrap_or("unknown");
+    let id = body
+        .get("responseId")
+        .and_then(Value::as_str)
+        .unwrap_or("chatcmpl-local-finalize");
+    let message_content = if text.is_empty() && !tool_calls.is_empty() {
+        Value::Null
+    } else {
+        Value::String(text)
+    };
+    let mut message = Map::new();
+    message.insert("role".to_string(), Value::String("assistant".to_string()));
+    message.insert("content".to_string(), message_content);
+    if !tool_calls.is_empty() {
+        message.insert("tool_calls".to_string(), Value::Array(tool_calls));
+    }
+    Some(json!({
+        "id": id,
+        "object": "chat.completion",
+        "model": model,
+        "choices": [{
+            "index": first_candidate.get("index").and_then(Value::as_u64).unwrap_or(0),
+            "message": Value::Object(message),
+            "finish_reason": finish_reason,
+        }],
+        "usage": {
+            "prompt_tokens": prompt_tokens,
+            "completion_tokens": completion_tokens,
+            "total_tokens": total_tokens,
+        }
+    }))
+}
--- a/apps/aether-gateway/src/ai_pipeline/conversion/response/to_openai_chat/gemini_cli.rs
+++ b/apps/aether-gateway/src/ai_pipeline/conversion/response/to_openai_chat/gemini_cli.rs
@@ -0,0 +1,83 @@
+use serde_json::{json, Value};
+
+use super::shared::{build_generated_tool_call_id, canonicalize_tool_arguments};
+use super::super::from_openai_chat::build_openai_cli_response;
+
+pub(crate) fn convert_gemini_cli_response_to_openai_cli(
+    body_json: &Value,
+    report_context: &Value,
+) -> Option<Value> {
+    let body = body_json.as_object()?;
+    let candidates = body.get("candidates")?.as_array()?;
+    let first_candidate = candidates.first()?.as_object()?;
+    let content = first_candidate.get("content")?.as_object()?;
+    let parts = content.get("parts")?.as_array()?;
+    let mut text = String::new();
+    let mut function_calls = Vec::new();
+    for (index, part) in parts.iter().enumerate() {
+        let part = part.as_object()?;
+        if let Some(piece) = part.get("text").and_then(Value::as_str) {
+            text.push_str(piece);
+        } else if let Some(function_call) = part.get("functionCall").and_then(Value::as_object) {
+            let tool_name = function_call.get("name")?.as_str()?;
+            let call_id = function_call
+                .get("id")
+                .and_then(Value::as_str)
+                .filter(|value| !value.is_empty())
+                .map(ToOwned::to_owned)
+                .unwrap_or_else(|| build_generated_tool_call_id(index));
+            let arguments = canonicalize_tool_arguments(function_call.get("args").cloned());
+            function_calls.push(json!({
+                "type": "function_call",
+                "call_id": call_id,
+                "name": tool_name,
+                "arguments": arguments,
+            }));
+        } else {
+            return None;
+        }
+    }
+
+    let usage = body.get("usageMetadata").and_then(Value::as_object);
+    let prompt_tokens = usage
+        .and_then(|value| value.get("promptTokenCount"))
+        .and_then(Value::as_u64)
+        .unwrap_or(0);
+    let output_tokens = usage
+        .map(|value| {
+            value
+                .get("candidatesTokenCount")
+                .and_then(Value::as_u64)
+                .unwrap_or(0)
+                + value
+                    .get("thoughtsTokenCount")
+                    .and_then(Value::as_u64)
+                    .unwrap_or(0)
+        })
+        .unwrap_or(0);
+    let total_tokens = usage
+        .and_then(|value| value.get("totalTokenCount"))
+        .and_then(Value::as_u64)
+        .unwrap_or(prompt_tokens + output_tokens);
+    let model = body
+        .get("modelVersion")
+        .and_then(Value::as_str)
+        .or_else(|| report_context.get("mapped_model").and_then(Value::as_str))
+        .or_else(|| report_context.get("model").and_then(Value::as_str))
+        .unwrap_or("unknown");
+    let response_id = body
+        .get("responseId")
+        .or_else(|| body.get("_v1internal_response_id"))
+        .and_then(Value::as_str)
+        .unwrap_or("resp-local-finalize");
+
+    Some(build_openai_cli_response(
+        response_id,
+        model,
+        &text,
+        function_calls,
+        prompt_tokens,
+        output_tokens,
+        total_tokens,
+    ))
+}
--- a/apps/aether-gateway/src/ai_pipeline/conversion/response/to_openai_chat/mod.rs
+++ b/apps/aether-gateway/src/ai_pipeline/conversion/response/to_openai_chat/mod.rs
@@ -0,0 +1,12 @@
+mod claude_chat;
+mod claude_cli;
+mod gemini_chat;
+mod gemini_cli;
+mod openai_cli;
+mod shared;
+
+pub(crate) use claude_chat::convert_claude_chat_response_to_openai_chat;
+pub(crate) use claude_cli::convert_claude_cli_response_to_openai_cli;
+pub(crate) use gemini_chat::convert_gemini_chat_response_to_openai_chat;
+pub(crate) use gemini_cli::convert_gemini_cli_response_to_openai_cli;
+pub(crate) use openai_cli::convert_openai_cli_response_to_openai_chat;
--- a/apps/aether-gateway/src/ai_pipeline/conversion/response/to_openai_chat/openai_cli.rs
+++ b/apps/aether-gateway/src/ai_pipeline/conversion/response/to_openai_chat/openai_cli.rs
@@ -0,0 +1,135 @@
+use serde_json::{json, Map, Value};
+
+use super::shared::{build_generated_tool_call_id, canonicalize_tool_arguments};
+
+pub(crate) fn convert_openai_cli_response_to_openai_chat(
+    body_json: &Value,
+    report_context: &Value,
+) -> Option<Value> {
+    let body = body_json.as_object()?;
+    let mut text = String::new();
+    let mut tool_calls = Vec::new();
+
+    if let Some(output_items) = body.get("output").and_then(Value::as_array) {
+        for (index, item) in output_items.iter().enumerate() {
+            let item_object = item.as_object()?;
+            let item_type = item_object
+                .get("type")
+                .and_then(Value::as_str)
+                .unwrap_or_default()
+                .trim()
+                .to_ascii_lowercase();
+            match item_type.as_str() {
+                "message" => {
+                    if let Some(content) = item_object.get("content").and_then(Value::as_array) {
+                        for part in content {
+                            let part_object = part.as_object()?;
+                            let part_type = part_object
+                                .get("type")
+                                .and_then(Value::as_str)
+                                .unwrap_or_default()
+                                .trim()
+                                .to_ascii_lowercase();
+                            if matches!(part_type.as_str(), "output_text" | "text") {
+                                if let Some(piece) = part_object.get("text").and_then(Value::as_str)
+                                {
+                                    text.push_str(piece);
+                                }
+                            }
+                        }
+                    }
+                }
+                "function_call" => {
+                    let tool_name = item_object
+                        .get("name")
+                        .and_then(Value::as_str)
+                        .map(str::trim)
+                        .filter(|value| !value.is_empty())?;
+                    let tool_id = item_object
+                        .get("call_id")
+                        .and_then(Value::as_str)
+                        .filter(|value| !value.is_empty())
+                        .or_else(|| {
+                            item_object
+                                .get("id")
+                                .and_then(Value::as_str)
+                                .filter(|value| !value.is_empty())
+                        })
+                        .map(ToOwned::to_owned)
+                        .unwrap_or_else(|| build_generated_tool_call_id(index));
+                    tool_calls.push(json!({
+                        "id": tool_id,
+                        "type": "function",
+                        "function": {
+                            "name": tool_name,
+                            "arguments": canonicalize_tool_arguments(item_object.get("arguments").cloned()),
+                        }
+                    }));
+                }
+                "output_text" | "text" => {
+                    if let Some(piece) = item_object.get("text").and_then(Value::as_str) {
+                        text.push_str(piece);
+                    }
+                }
+                _ => {}
+            }
+        }
+    }
+
+    let finish_reason = if tool_calls.is_empty() {
+        Some("stop")
+    } else {
+        Some("tool_calls")
+    };
+    let model = body
+        .get("model")
+        .and_then(Value::as_str)
+        .or_else(|| report_context.get("mapped_model").and_then(Value::as_str))
+        .or_else(|| report_context.get("model").and_then(Value::as_str))
+        .unwrap_or("unknown");
+    let id = body
+        .get("id")
+        .and_then(Value::as_str)
+        .unwrap_or("chatcmpl-local-openai-cli");
+
+    let usage = body.get("usage").and_then(Value::as_object);
+    let prompt_tokens = usage
+        .and_then(|value| value.get("input_tokens"))
+        .and_then(Value::as_u64)
+        .unwrap_or(0);
+    let completion_tokens = usage
+        .and_then(|value| value.get("output_tokens"))
+        .and_then(Value::as_u64)
+        .unwrap_or(0);
+    let total_tokens = usage
+        .and_then(|value| value.get("total_tokens"))
+        .and_then(Value::as_u64)
+        .unwrap_or(prompt_tokens + completion_tokens);
+
+    let mut message = Map::new();
+    message.insert("role".to_string(), Value::String("assistant".to_string()));
+    if text.is_empty() && !tool_calls.is_empty() {
+        message.insert("content".to_string(), Value::Null);
+    } else {
+        message.insert("content".to_string(), Value::String(text));
+    }
+    if !tool_calls.is_empty() {
+        message.insert("tool_calls".to_string(), Value::Array(tool_calls));
+    }
+
+    Some(json!({
+        "id": id,
+        "object": "chat.completion",
+        "model": model,
+        "choices": [{
+            "index": 0,
+            "message": Value::Object(message),
+            "finish_reason": finish_reason,
+        }],
+        "usage": {
+            "prompt_tokens": prompt_tokens,
+            "completion_tokens": completion_tokens,
+            "total_tokens": total_tokens,
+        }
+    }))
+}
--- a/apps/aether-gateway/src/ai_pipeline/conversion/response/to_openai_chat/shared.rs
+++ b/apps/aether-gateway/src/ai_pipeline/conversion/response/to_openai_chat/shared.rs
@@ -0,0 +1,13 @@
+use serde_json::Value;
+
+pub(super) fn build_generated_tool_call_id(index: usize) -> String {
+    format!("call_auto_{index}")
+}
+
+pub(super) fn canonicalize_tool_arguments(value: Option<Value>) -> String {
+    match value {
+        Some(Value::String(text)) => text,
+        Some(other) => serde_json::to_string(&other).unwrap_or_else(|_| "null".to_string()),
+        None => "{}".to_string(),
+    }
+}
--- a/apps/aether-gateway/src/ai_pipeline/finalize/common.rs
+++ b/apps/aether-gateway/src/ai_pipeline/finalize/common.rs
@@ -1,156 +1,24 @@
-use super::*;
+use std::collections::BTreeMap;
+
+use axum::body::Body;
+use axum::http::Response;
+use serde_json::Value;
+
+pub(crate) use crate::gateway::ai_pipeline::runtime::{
+    normalize_provider_private_response_value as unwrap_local_finalize_response_value,
+    provider_private_response_allows_sync_finalize as local_finalize_allows_envelope,
+};
+use crate::gateway::{
+    build_client_response_from_parts, GatewayControlDecision, GatewayError,
+    GatewaySyncReportRequest,
+};

 pub(crate) struct LocalCoreSyncFinalizeOutcome {
    pub(crate) response: Response<Body>,
    pub(crate) background_report: Option<GatewaySyncReportRequest>,
 }

-pub(super) fn local_finalize_allows_envelope(report_context: &Value) -> bool {
-    let has_envelope = report_context
-        .get("has_envelope")
-        .and_then(Value::as_bool)
-        .unwrap_or(false);
-    if !has_envelope {
-        return true;
-    }
-    matches!(
-        report_context.get("envelope_name").and_then(Value::as_str),
-        Some("gemini_cli:v1internal")
-            | Some("antigravity:v1internal")
-            | Some("claude:cli")
-            | Some("kiro:generateAssistantResponse")
-    )
-}
-
-fn local_finalize_response_model(report_context: &Value) -> &str {
-    report_context
-        .get("mapped_model")
-        .and_then(Value::as_str)
-        .or_else(|| report_context.get("model").and_then(Value::as_str))
-        .unwrap_or_default()
-}
-
-fn inject_antigravity_claude_tool_ids_response(response: &mut Value, model: &str) {
-    if !model.to_ascii_lowercase().contains("claude") {
-        return;
-    }
-
-    let Some(candidates) = response.get_mut("candidates").and_then(Value::as_array_mut) else {
-        return;
-    };
-
-    for candidate in candidates {
-        let Some(parts) = candidate
-            .get_mut("content")
-            .and_then(Value::as_object_mut)
-            .and_then(|content| content.get_mut("parts"))
-            .and_then(Value::as_array_mut)
-        else {
-            continue;
-        };
-
-        let mut name_counters: BTreeMap<String, usize> = BTreeMap::new();
-        for part in parts {
-            let function_call = if let Some(function_call) =
-                part.get_mut("functionCall").and_then(Value::as_object_mut)
-            {
-                function_call
-            } else if let Some(function_call) =
-                part.get_mut("function_call").and_then(Value::as_object_mut)
-            {
-                function_call
-            } else {
-                continue;
-            };
-            let has_id = function_call
-                .get("id")
-                .and_then(Value::as_str)
-                .is_some_and(|value| !value.is_empty());
-            if has_id {
-                continue;
-            }
-            let function_name = function_call
-                .get("name")
-                .and_then(Value::as_str)
-                .filter(|value| !value.is_empty())
-                .unwrap_or("unknown")
-                .to_string();
-            let count = name_counters.entry(function_name.clone()).or_insert(0);
-            function_call.insert(
-                "id".to_string(),
-                Value::String(format!("call_{function_name}_{count}")),
-            );
-            *count += 1;
-        }
-    }
-}
-
-fn postprocess_local_finalize_response_value(data: &mut Value, report_context: &Value) {
-    if !matches!(
-        report_context.get("envelope_name").and_then(Value::as_str),
-        Some("antigravity:v1internal")
-    ) {
-        return;
-    }
-    if let Some(object) = data.as_object_mut() {
-        if !object.contains_key("_v1internal_response_id") {
-            if let Some(response_id) = object.remove("responseId") {
-                object.insert("_v1internal_response_id".to_string(), response_id);
-            }
-        }
-    }
-    inject_antigravity_claude_tool_ids_response(
-        data,
-        local_finalize_response_model(report_context),
-    );
-}
-
-pub(crate) fn unwrap_local_finalize_response_value(
-    data: Value,
-    report_context: &Value,
-) -> Result<Option<Value>, GatewayError> {
-    if !report_context
-        .get("has_envelope")
-        .and_then(Value::as_bool)
-        .unwrap_or(false)
-    {
-        return Ok(Some(data));
-    }
-    let mut unwrapped = match report_context.get("envelope_name").and_then(Value::as_str) {
-        Some("claude:cli") | Some("kiro:generateAssistantResponse") => data,
-        Some("gemini_cli:v1internal") => {
-            if let Some(response) = data
-                .get("response")
-                .and_then(Value::as_object)
-                .filter(|response| !response.contains_key("response"))
-            {
-                Value::Object(response.clone())
-            } else {
-                data
-            }
-        }
-        Some("antigravity:v1internal") => {
-            if let Some(response) = data
-                .get("response")
-                .and_then(Value::as_object)
-                .filter(|response| !response.contains_key("response"))
-            {
-                let mut unwrapped = response.clone();
-                if let Some(response_id) = data.get("responseId").cloned() {
-                    unwrapped.insert("_v1internal_response_id".to_string(), response_id);
-                }
-                Value::Object(unwrapped)
-            } else {
-                data
-            }
-        }
-        _ => return Ok(None),
-    };
-    postprocess_local_finalize_response_value(&mut unwrapped, report_context);
-    Ok(Some(unwrapped))
-}
-
-pub(super) fn build_local_success_outcome(
+pub(crate) fn build_local_success_outcome(
    trace_id: &str,
    decision: &GatewayControlDecision,
    payload: &GatewaySyncReportRequest,
@@ -169,7 +37,7 @@ pub(super) fn build_local_success_outcome(
    )
 }

-pub(super) fn build_local_success_outcome_with_report(
+pub(crate) fn build_local_success_outcome_with_report(
    trace_id: &str,
    decision: &GatewayControlDecision,
    status_code: u16,
@@ -196,7 +64,7 @@ pub(super) fn build_local_success_outcome_with_report(
    })
 }

-pub(super) fn build_local_success_outcome_with_conversion_report(
+pub(crate) fn build_local_success_outcome_with_conversion_report(
    trace_id: &str,
    decision: &GatewayControlDecision,
    payload: &GatewaySyncReportRequest,
@@ -284,7 +152,7 @@ pub(crate) fn build_generated_tool_call_id(index: usize) -> String {
    format!("call_auto_{index}")
 }

-pub(super) fn parse_stream_json_events(body: &[u8]) -> Option<Vec<Value>> {
+pub(crate) fn parse_stream_json_events(body: &[u8]) -> Option<Vec<Value>> {
    let text = std::str::from_utf8(body).ok()?;
    let trimmed = text.trim();
    if trimmed.is_empty() {
--- a/apps/aether-gateway/src/ai_pipeline/finalize/internal/mod.rs
+++ b/apps/aether-gateway/src/ai_pipeline/finalize/internal/mod.rs
@@ -0,0 +1,38 @@
+use axum::body::Body;
+use axum::http::Response;
+use serde_json::Value;
+
+use crate::gateway::{GatewayControlDecision, GatewayError, GatewaySyncReportRequest};
+
+#[path = "stream_rewrite.rs"]
+pub(crate) mod stream;
+#[path = "sync_finalize.rs"]
+pub(crate) mod sync;
+
+pub(crate) use stream::LocalStreamRewriter;
+pub(crate) use sync::LocalCoreSyncFinalizeOutcome;
+
+pub(crate) fn maybe_build_sync_finalize_outcome(
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    payload: &GatewaySyncReportRequest,
+) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
+    sync::maybe_build_local_core_sync_finalize_response(trace_id, decision, payload)
+}
+
+pub(crate) fn maybe_compile_sync_finalize_response(
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    payload: &GatewaySyncReportRequest,
+) -> Result<Option<Response<Body>>, GatewayError> {
+    Ok(
+        maybe_build_sync_finalize_outcome(trace_id, decision, payload)?
+            .map(|outcome| outcome.response),
+    )
+}
+
+pub(crate) fn maybe_build_stream_response_rewriter(
+    report_context: Option<&Value>,
+) -> Option<LocalStreamRewriter> {
+    stream::maybe_build_local_stream_rewriter(report_context)
+}
--- a/apps/aether-gateway/src/ai_pipeline/finalize/internal/stream_rewrite.rs
+++ b/apps/aether-gateway/src/ai_pipeline/finalize/internal/stream_rewrite.rs
@@ -1,44 +1,39 @@
 use serde_json::Value;

+use crate::gateway::ai_pipeline::finalize::standard::{
+    BufferedCliConversionStreamState, BufferedStandardConversionStreamState,
+    ClaudeToOpenAIChatStreamState, GeminiToOpenAIChatStreamState, OpenAICliToOpenAIChatStreamState,
+};
+use crate::gateway::ai_pipeline::finalize::sse::{encode_done_sse, encode_json_sse};
+use crate::gateway::ai_pipeline::private_response::transform_provider_private_stream_line as transform_envelope_line;
+use crate::gateway::ai_pipeline::runtime::KiroToClaudeCliStreamState;
 use crate::gateway::GatewayError;

-use super::kiro_stream::KiroToClaudeCliStreamState;
-use super::local_finalize::{
+use super::sync::{
    aggregate_claude_stream_sync_response, aggregate_gemini_stream_sync_response,
    convert_claude_cli_response_to_openai_cli, convert_gemini_cli_response_to_openai_cli,
 };
-
-#[path = "local_stream/chat.rs"]
-mod chat;
-#[path = "local_stream/cli.rs"]
-mod cli;
-#[path = "local_stream/common.rs"]
-mod common;
-#[path = "local_stream/envelope.rs"]
-mod envelope;
-
-use chat::{ClaudeToOpenAIChatStreamState, GeminiToOpenAIChatStreamState};
-use cli::BufferedCliConversionStreamState;
-use envelope::transform_envelope_line;
-
 enum RewriteMode {
    EnvelopeUnwrap,
    ClaudeToOpenAIChat(ClaudeToOpenAIChatStreamState),
    GeminiToOpenAIChat(GeminiToOpenAIChatStreamState),
+    OpenAICliToOpenAIChat(OpenAICliToOpenAIChatStreamState),
    ClaudeToOpenAICli(BufferedCliConversionStreamState),
    GeminiToOpenAICli(BufferedCliConversionStreamState),
    AntigravityGeminiToOpenAIChat(GeminiToOpenAIChatStreamState),
    AntigravityGeminiToOpenAICli(BufferedCliConversionStreamState),
    KiroToClaudeCli(KiroToClaudeCliStreamState),
+    StandardChat(BufferedStandardConversionStreamState),
+    StandardCli(BufferedStandardConversionStreamState),
 }

-pub(super) struct LocalStreamRewriter {
+pub(crate) struct LocalStreamRewriter {
    report_context: Value,
    buffered: Vec<u8>,
    mode: RewriteMode,
 }

-pub(super) fn maybe_build_local_stream_rewriter(
+pub(crate) fn maybe_build_local_stream_rewriter(
    report_context: Option<&Value>,
 ) -> Option<LocalStreamRewriter> {
    let report_context = report_context?;
@@ -77,6 +72,9 @@ pub(super) fn maybe_build_local_stream_rewriter(
            ("", "gemini:chat", "openai:chat") => {
                RewriteMode::GeminiToOpenAIChat(GeminiToOpenAIChatStreamState::default())
            }
+            ("", "openai:cli", "openai:chat") | ("", "openai:compact", "openai:chat") => {
+                RewriteMode::OpenAICliToOpenAIChat(OpenAICliToOpenAIChatStreamState::default())
+            }
            ("", "claude:cli", "openai:cli") => {
                RewriteMode::ClaudeToOpenAICli(BufferedCliConversionStreamState::default())
            }
@@ -102,6 +100,16 @@ pub(super) fn maybe_build_local_stream_rewriter(
                    BufferedCliConversionStreamState::default(),
                )
            }
+            _ if is_standard_chat_client_api_format(client_api_format.as_str())
+                && is_standard_provider_api_format(provider_api_format.as_str()) =>
+            {
+                RewriteMode::StandardChat(BufferedStandardConversionStreamState::default())
+            }
+            _ if is_standard_cli_client_api_format(client_api_format.as_str())
+                && is_standard_provider_api_format(provider_api_format.as_str()) =>
+            {
+                RewriteMode::StandardCli(BufferedStandardConversionStreamState::default())
+            }
            _ => return None,
        }
    } else {
@@ -141,7 +149,7 @@ pub(super) fn maybe_build_local_stream_rewriter(
 }

 impl LocalStreamRewriter {
-    pub(super) fn push_chunk(&mut self, chunk: &[u8]) -> Result<Vec<u8>, GatewayError> {
+    pub(crate) fn push_chunk(&mut self, chunk: &[u8]) -> Result<Vec<u8>, GatewayError> {
        if let RewriteMode::KiroToClaudeCli(state) = &mut self.mode {
            return state.push_chunk(&self.report_context, chunk);
        }
@@ -154,7 +162,7 @@ impl LocalStreamRewriter {
        Ok(output)
    }

-    pub(super) fn finish(&mut self) -> Result<Vec<u8>, GatewayError> {
+    pub(crate) fn finish(&mut self) -> Result<Vec<u8>, GatewayError> {
        if let RewriteMode::KiroToClaudeCli(state) = &mut self.mode {
            return state.finish(&self.report_context);
        }
@@ -164,6 +172,9 @@ impl LocalStreamRewriter {
                RewriteMode::GeminiToOpenAIChat(state) => {
                    return state.finish(&self.report_context);
                }
+                RewriteMode::OpenAICliToOpenAIChat(state) => {
+                    return state.finish(&self.report_context);
+                }
                RewriteMode::ClaudeToOpenAICli(state) => {
                    return state.finish(
                        &self.report_context,
@@ -189,6 +200,12 @@ impl LocalStreamRewriter {
                    );
                }
                RewriteMode::KiroToClaudeCli(_) => {}
+                RewriteMode::StandardChat(state) => {
+                    return state.finish_as_chat(&self.report_context)
+                }
+                RewriteMode::StandardCli(state) => {
+                    return state.finish_as_cli(&self.report_context)
+                }
                RewriteMode::EnvelopeUnwrap => {}
            }
            return Ok(Vec::new());
@@ -202,6 +219,9 @@ impl LocalStreamRewriter {
            RewriteMode::GeminiToOpenAIChat(state) => {
                output.extend(state.finish(&self.report_context)?);
            }
+            RewriteMode::OpenAICliToOpenAIChat(state) => {
+                output.extend(state.finish(&self.report_context)?);
+            }
            RewriteMode::ClaudeToOpenAICli(state) => {
                output.extend(state.finish(
                    &self.report_context,
@@ -227,6 +247,12 @@ impl LocalStreamRewriter {
                )?);
            }
            RewriteMode::KiroToClaudeCli(_) => {}
+            RewriteMode::StandardChat(state) => {
+                output.extend(state.finish_as_chat(&self.report_context)?);
+            }
+            RewriteMode::StandardCli(state) => {
+                output.extend(state.finish_as_cli(&self.report_context)?);
+            }
            RewriteMode::EnvelopeUnwrap => {}
        }
        Ok(output)
@@ -241,6 +267,9 @@ impl LocalStreamRewriter {
            RewriteMode::GeminiToOpenAIChat(state) => {
                state.transform_line(&self.report_context, line)
            }
+            RewriteMode::OpenAICliToOpenAIChat(state) => {
+                state.transform_line(&self.report_context, line)
+            }
            RewriteMode::ClaudeToOpenAICli(state) | RewriteMode::GeminiToOpenAICli(state) => {
                state.transform_line(line)
            }
@@ -260,11 +289,38 @@ impl LocalStreamRewriter {
                    state.transform_line(unwrapped)
                }
            }
+            RewriteMode::StandardChat(state) | RewriteMode::StandardCli(state) => {
+                state.transform_line(&self.report_context, line)
+            }
            RewriteMode::KiroToClaudeCli(_) => Ok(Vec::new()),
        }
    }
 }

+fn is_standard_provider_api_format(api_format: &str) -> bool {
+    matches!(
+        api_format,
+        "openai:chat"
+            | "openai:cli"
+            | "openai:compact"
+            | "claude:chat"
+            | "claude:cli"
+            | "gemini:chat"
+            | "gemini:cli"
+    )
+}
+
+fn is_standard_chat_client_api_format(api_format: &str) -> bool {
+    matches!(api_format, "openai:chat" | "claude:chat" | "gemini:chat")
+}
+
+fn is_standard_cli_client_api_format(api_format: &str) -> bool {
+    matches!(
+        api_format,
+        "openai:cli" | "openai:compact" | "claude:cli" | "gemini:cli"
+    )
+}
+
 #[cfg(test)]
-#[path = "local_stream/tests.rs"]
+#[path = "../tests_stream.rs"]
 mod tests;
--- a/apps/aether-gateway/src/ai_pipeline/finalize/internal/sync_finalize.rs
+++ b/apps/aether-gateway/src/ai_pipeline/finalize/internal/sync_finalize.rs
@@ -0,0 +1,377 @@
+use std::collections::BTreeMap;
+
+use axum::body::Body;
+use axum::http::Response;
+use base64::Engine as _;
+use serde_json::{json, Map, Value};
+
+use crate::gateway::ai_pipeline::conversion::{
+    sync_chat_response_conversion_kind, sync_cli_response_conversion_kind,
+};
+use crate::gateway::{
+    build_client_response_from_parts, GatewayControlDecision, GatewayError,
+    GatewaySyncReportRequest,
+};
+
+pub(crate) use crate::gateway::ai_pipeline::finalize::common::{
+    build_generated_tool_call_id, build_local_success_outcome,
+    build_local_success_outcome_with_conversion_report, canonicalize_tool_arguments,
+    local_finalize_allows_envelope, parse_stream_json_events, unwrap_local_finalize_response_value,
+    LocalCoreSyncFinalizeOutcome,
+};
+pub(crate) use crate::gateway::ai_pipeline::finalize::standard::{
+    aggregate_claude_stream_sync_response, aggregate_gemini_stream_sync_response,
+    aggregate_openai_chat_stream_sync_response, aggregate_openai_cli_stream_sync_response,
+    aggregate_standard_chat_stream_sync_response, aggregate_standard_cli_stream_sync_response,
+    convert_claude_chat_response_to_openai_chat, convert_claude_cli_response_to_openai_cli,
+    convert_gemini_chat_response_to_openai_chat, convert_gemini_cli_response_to_openai_cli,
+    convert_openai_chat_response_to_claude_chat, convert_openai_chat_response_to_gemini_chat,
+    convert_openai_chat_response_to_openai_cli, convert_openai_cli_response_to_openai_chat,
+    convert_standard_chat_response, convert_standard_cli_response,
+    maybe_build_local_claude_cli_stream_sync_response,
+    maybe_build_local_claude_stream_sync_response, maybe_build_local_claude_sync_response,
+    maybe_build_local_gemini_cli_stream_sync_response,
+    maybe_build_local_gemini_stream_sync_response, maybe_build_local_gemini_sync_response,
+    maybe_build_local_openai_chat_cross_format_stream_sync_response,
+    maybe_build_local_openai_chat_cross_format_sync_response,
+    maybe_build_local_openai_chat_stream_sync_response,
+    maybe_build_local_openai_chat_sync_response,
+    maybe_build_local_openai_cli_cross_format_stream_sync_response,
+    maybe_build_local_openai_cli_cross_format_sync_response,
+    maybe_build_local_openai_cli_stream_sync_response,
+};
+
+pub(crate) fn maybe_build_local_core_sync_finalize_response(
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    payload: &GatewaySyncReportRequest,
+) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
+    let Some(normalized_payload) =
+        crate::gateway::ai_pipeline::private_response::maybe_normalize_provider_private_sync_report_payload(payload)?
+    else {
+        return Ok(None);
+    };
+    let payload = &normalized_payload;
+    if let Some(response) =
+        maybe_build_local_openai_chat_stream_sync_response(trace_id, decision, payload)?
+    {
+        return Ok(Some(response));
+    }
+    if let Some(response) =
+        maybe_build_local_openai_chat_sync_response(trace_id, decision, payload)?
+    {
+        return Ok(Some(response));
+    }
+    if let Some(response) = maybe_build_local_openai_chat_cross_format_stream_sync_response(
+        trace_id, decision, payload,
+    )? {
+        return Ok(Some(response));
+    }
+    if let Some(response) =
+        maybe_build_local_openai_cli_stream_sync_response(trace_id, decision, payload)?
+    {
+        return Ok(Some(response));
+    }
+    if let Some(response) =
+        maybe_build_local_openai_cli_cross_format_stream_sync_response(trace_id, decision, payload)?
+    {
+        return Ok(Some(response));
+    }
+    if let Some(response) =
+        maybe_build_local_claude_cli_stream_sync_response(trace_id, decision, payload)?
+    {
+        return Ok(Some(response));
+    }
+    if let Some(response) =
+        maybe_build_local_gemini_cli_stream_sync_response(trace_id, decision, payload)?
+    {
+        return Ok(Some(response));
+    }
+    if let Some(response) =
+        maybe_build_local_claude_stream_sync_response(trace_id, decision, payload)?
+    {
+        return Ok(Some(response));
+    }
+    if let Some(response) = maybe_build_local_claude_sync_response(trace_id, decision, payload)? {
+        return Ok(Some(response));
+    }
+    if let Some(response) =
+        maybe_build_local_gemini_stream_sync_response(trace_id, decision, payload)?
+    {
+        return Ok(Some(response));
+    }
+    if let Some(response) = maybe_build_local_gemini_sync_response(trace_id, decision, payload)? {
+        return Ok(Some(response));
+    }
+    if let Some(response) =
+        maybe_build_local_openai_chat_cross_format_sync_response(trace_id, decision, payload)?
+    {
+        return Ok(Some(response));
+    }
+    if let Some(response) =
+        maybe_build_local_openai_cli_cross_format_sync_response(trace_id, decision, payload)?
+    {
+        return Ok(Some(response));
+    }
+    if let Some(response) = maybe_build_local_standard_chat_cross_format_stream_sync_response(
+        trace_id, decision, payload,
+    )? {
+        return Ok(Some(response));
+    }
+    if let Some(response) =
+        maybe_build_local_standard_chat_cross_format_sync_response(trace_id, decision, payload)?
+    {
+        return Ok(Some(response));
+    }
+    if let Some(response) = maybe_build_local_standard_cli_cross_format_stream_sync_response(
+        trace_id, decision, payload,
+    )? {
+        return Ok(Some(response));
+    }
+    if let Some(response) =
+        maybe_build_local_standard_cli_cross_format_sync_response(trace_id, decision, payload)?
+    {
+        return Ok(Some(response));
+    }
+    Ok(None)
+}
+
+fn maybe_build_local_standard_chat_cross_format_stream_sync_response(
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    payload: &GatewaySyncReportRequest,
+) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
+    if !matches!(
+        payload.report_kind.as_str(),
+        "openai_chat_sync_finalize" | "claude_chat_sync_finalize" | "gemini_chat_sync_finalize"
+    ) || payload.status_code >= 400
+    {
+        return Ok(None);
+    }
+
+    let Some(report_context) = payload.report_context.as_ref() else {
+        return Ok(None);
+    };
+    let provider_api_format = report_context
+        .get("provider_api_format")
+        .and_then(Value::as_str)
+        .unwrap_or_default()
+        .trim()
+        .to_ascii_lowercase();
+    let client_api_format = report_context
+        .get("client_api_format")
+        .and_then(Value::as_str)
+        .unwrap_or_default()
+        .trim()
+        .to_ascii_lowercase();
+    if !local_finalize_allows_envelope(report_context)
+        || sync_chat_response_conversion_kind(&provider_api_format, &client_api_format).is_none()
+    {
+        return Ok(None);
+    }
+    let Some(body_base64) = payload.body_base64.as_deref() else {
+        return Ok(None);
+    };
+    let body_bytes = base64::engine::general_purpose::STANDARD
+        .decode(body_base64)
+        .map_err(|err| GatewayError::Internal(err.to_string()))?;
+    let Some(aggregated) =
+        aggregate_standard_chat_stream_sync_response(&body_bytes, &provider_api_format)
+    else {
+        return Ok(None);
+    };
+    let Some(aggregated) = unwrap_local_finalize_response_value(aggregated, report_context)? else {
+        return Ok(None);
+    };
+    let Some(converted) = convert_standard_chat_response(
+        &aggregated,
+        &provider_api_format,
+        &client_api_format,
+        report_context,
+    ) else {
+        return Ok(None);
+    };
+
+    Ok(Some(build_local_success_outcome_with_conversion_report(
+        trace_id, decision, payload, converted, aggregated,
+    )?))
+}
+
+fn maybe_build_local_standard_chat_cross_format_sync_response(
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    payload: &GatewaySyncReportRequest,
+) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
+    if !matches!(
+        payload.report_kind.as_str(),
+        "openai_chat_sync_finalize" | "claude_chat_sync_finalize" | "gemini_chat_sync_finalize"
+    ) || payload.status_code >= 400
+    {
+        return Ok(None);
+    }
+
+    let Some(report_context) = payload.report_context.as_ref() else {
+        return Ok(None);
+    };
+    let provider_api_format = report_context
+        .get("provider_api_format")
+        .and_then(Value::as_str)
+        .unwrap_or_default()
+        .trim()
+        .to_ascii_lowercase();
+    let client_api_format = report_context
+        .get("client_api_format")
+        .and_then(Value::as_str)
+        .unwrap_or_default()
+        .trim()
+        .to_ascii_lowercase();
+    if !local_finalize_allows_envelope(report_context)
+        || sync_chat_response_conversion_kind(&provider_api_format, &client_api_format).is_none()
+    {
+        return Ok(None);
+    }
+    let Some(body_json) = payload.body_json.as_ref() else {
+        return Ok(None);
+    };
+    let Some(body_json) = unwrap_local_finalize_response_value(body_json.clone(), report_context)?
+    else {
+        return Ok(None);
+    };
+    let Some(converted) = convert_standard_chat_response(
+        &body_json,
+        &provider_api_format,
+        &client_api_format,
+        report_context,
+    ) else {
+        return Ok(None);
+    };
+
+    Ok(Some(build_local_success_outcome_with_conversion_report(
+        trace_id, decision, payload, converted, body_json,
+    )?))
+}
+
+fn maybe_build_local_standard_cli_cross_format_stream_sync_response(
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    payload: &GatewaySyncReportRequest,
+) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
+    if !matches!(
+        payload.report_kind.as_str(),
+        "openai_cli_sync_finalize"
+            | "openai_compact_sync_finalize"
+            | "claude_cli_sync_finalize"
+            | "gemini_cli_sync_finalize"
+    ) || payload.status_code >= 400
+    {
+        return Ok(None);
+    }
+
+    let Some(report_context) = payload.report_context.as_ref() else {
+        return Ok(None);
+    };
+    let provider_api_format = report_context
+        .get("provider_api_format")
+        .and_then(Value::as_str)
+        .unwrap_or_default()
+        .trim()
+        .to_ascii_lowercase();
+    let client_api_format = report_context
+        .get("client_api_format")
+        .and_then(Value::as_str)
+        .unwrap_or_default()
+        .trim()
+        .to_ascii_lowercase();
+    if !local_finalize_allows_envelope(report_context)
+        || sync_cli_response_conversion_kind(&provider_api_format, &client_api_format).is_none()
+    {
+        return Ok(None);
+    }
+    let Some(body_base64) = payload.body_base64.as_deref() else {
+        return Ok(None);
+    };
+    let body_bytes = base64::engine::general_purpose::STANDARD
+        .decode(body_base64)
+        .map_err(|err| GatewayError::Internal(err.to_string()))?;
+    let Some(aggregated) =
+        aggregate_standard_cli_stream_sync_response(&body_bytes, &provider_api_format)
+    else {
+        return Ok(None);
+    };
+    let Some(aggregated) = unwrap_local_finalize_response_value(aggregated, report_context)? else {
+        return Ok(None);
+    };
+    let Some(converted) = convert_standard_cli_response(
+        &aggregated,
+        &provider_api_format,
+        &client_api_format,
+        report_context,
+    ) else {
+        return Ok(None);
+    };
+
+    Ok(Some(build_local_success_outcome_with_conversion_report(
+        trace_id, decision, payload, converted, aggregated,
+    )?))
+}
+
+fn maybe_build_local_standard_cli_cross_format_sync_response(
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    payload: &GatewaySyncReportRequest,
+) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
+    if !matches!(
+        payload.report_kind.as_str(),
+        "openai_cli_sync_finalize"
+            | "openai_compact_sync_finalize"
+            | "claude_cli_sync_finalize"
+            | "gemini_cli_sync_finalize"
+    ) || payload.status_code >= 400
+    {
+        return Ok(None);
+    }
+
+    let Some(report_context) = payload.report_context.as_ref() else {
+        return Ok(None);
+    };
+    let provider_api_format = report_context
+        .get("provider_api_format")
+        .and_then(Value::as_str)
+        .unwrap_or_default()
+        .trim()
+        .to_ascii_lowercase();
+    let client_api_format = report_context
+        .get("client_api_format")
+        .and_then(Value::as_str)
+        .unwrap_or_default()
+        .trim()
+        .to_ascii_lowercase();
+    if !local_finalize_allows_envelope(report_context)
+        || sync_cli_response_conversion_kind(&provider_api_format, &client_api_format).is_none()
+    {
+        return Ok(None);
+    }
+    let Some(body_json) = payload.body_json.as_ref() else {
+        return Ok(None);
+    };
+    let Some(body_json) = unwrap_local_finalize_response_value(body_json.clone(), report_context)?
+    else {
+        return Ok(None);
+    };
+    let Some(converted) = convert_standard_cli_response(
+        &body_json,
+        &provider_api_format,
+        &client_api_format,
+        report_context,
+    ) else {
+        return Ok(None);
+    };
+
+    Ok(Some(build_local_success_outcome_with_conversion_report(
+        trace_id, decision, payload, converted, body_json,
+    )?))
+}
+
+#[cfg(test)]
+#[path = "../tests_sync.rs"]
+mod tests;
--- a/apps/aether-gateway/src/ai_pipeline/finalize/mod.rs
+++ b/apps/aether-gateway/src/ai_pipeline/finalize/mod.rs
@@ -0,0 +1,13 @@
+pub(crate) mod common;
+pub(crate) mod sse;
+pub(crate) mod standard;
+pub(crate) use common::build_local_success_outcome;
+pub(crate) use internal::{
+    maybe_build_stream_response_rewriter, maybe_build_sync_finalize_outcome,
+    maybe_compile_sync_finalize_response, LocalCoreSyncFinalizeOutcome,
+};
+pub(crate) use crate::gateway::build_client_response;
+pub(crate) use crate::gateway::build_client_response_from_parts;
+pub(crate) use crate::gateway::execution_runtime::maybe_build_local_sync_finalize_response;
+
+pub(crate) mod internal;
--- a/apps/aether-gateway/src/ai_pipeline/finalize/sse.rs
+++ b/apps/aether-gateway/src/ai_pipeline/finalize/sse.rs
@@ -2,7 +2,7 @@ use serde_json::Value;

 use crate::gateway::GatewayError;

-pub(super) fn map_claude_stop_reason(
+pub(crate) fn map_claude_stop_reason(
    stop_reason: Option<&str>,
    has_tool_calls: bool,
 ) -> Option<&'static str> {
@@ -20,11 +20,11 @@ pub(super) fn map_claude_stop_reason(
    }
 }

-pub(super) fn encode_done_sse() -> Vec<u8> {
+pub(crate) fn encode_done_sse() -> Vec<u8> {
    b"data: [DONE]\n\n".to_vec()
 }

-pub(super) fn encode_json_sse(event: Option<&str>, value: &Value) -> Result<Vec<u8>, GatewayError> {
+pub(crate) fn encode_json_sse(event: Option<&str>, value: &Value) -> Result<Vec<u8>, GatewayError> {
    let mut out = Vec::new();
    if let Some(event) = event.filter(|value| !value.trim().is_empty()) {
        out.extend_from_slice(b"event: ");
--- a/apps/aether-gateway/src/ai_pipeline/finalize/standard/claude/chat.rs
+++ b/apps/aether-gateway/src/ai_pipeline/finalize/standard/claude/chat.rs
@@ -1,11 +1,11 @@
 use base64::Engine as _;

-use super::super::common::{
+use super::*;
+use crate::gateway::ai_pipeline::finalize::common::{
    build_generated_tool_call_id, build_local_success_outcome, canonicalize_tool_arguments,
    local_finalize_allows_envelope, parse_stream_json_events, unwrap_local_finalize_response_value,
    LocalCoreSyncFinalizeOutcome,
 };
-use super::super::*;

 #[derive(Debug, Default)]
 struct ClaudeContentBlockState {
@@ -215,6 +215,131 @@ pub(crate) fn convert_claude_chat_response_to_openai_chat(
    }))
 }

+pub(crate) fn convert_openai_chat_response_to_claude_chat(
+    body_json: &Value,
+    report_context: &Value,
+) -> Option<Value> {
+    let body = body_json.as_object()?;
+    let choices = body.get("choices")?.as_array()?;
+    let first_choice = choices.first()?.as_object()?;
+    let message = first_choice.get("message")?.as_object()?;
+    let mut content = Vec::new();
+
+    if let Some(text) = extract_openai_assistant_text(message.get("content")) {
+        if !text.trim().is_empty() {
+            content.push(json!({
+                "type": "text",
+                "text": text,
+            }));
+        }
+    }
+    if let Some(tool_call_values) = message.get("tool_calls").and_then(Value::as_array) {
+        for (index, tool_call) in tool_call_values.iter().enumerate() {
+            let tool_call = tool_call.as_object()?;
+            let function = tool_call.get("function")?.as_object()?;
+            let tool_name = function
+                .get("name")
+                .and_then(Value::as_str)
+                .map(str::trim)
+                .filter(|value| !value.is_empty())?;
+            let tool_id = tool_call
+                .get("id")
+                .and_then(Value::as_str)
+                .filter(|value| !value.is_empty())
+                .map(ToOwned::to_owned)
+                .unwrap_or_else(|| build_generated_tool_call_id(index));
+            let input = parse_openai_function_arguments(function.get("arguments"))?;
+            content.push(json!({
+                "type": "tool_use",
+                "id": tool_id,
+                "name": tool_name,
+                "input": input,
+            }));
+        }
+    }
+    if content.is_empty() {
+        content.push(json!({
+            "type": "text",
+            "text": "",
+        }));
+    }
+
+    let stop_reason = match first_choice.get("finish_reason").and_then(Value::as_str) {
+        Some("stop") | None => "end_turn",
+        Some("length") => "max_tokens",
+        Some("tool_calls") | Some("function_call") => "tool_use",
+        Some("content_filter") => "content_filtered",
+        Some(other) => other,
+    };
+    let usage = body.get("usage").and_then(Value::as_object);
+    let input_tokens = usage
+        .and_then(|value| value.get("prompt_tokens"))
+        .and_then(Value::as_u64)
+        .unwrap_or(0);
+    let output_tokens = usage
+        .and_then(|value| value.get("completion_tokens"))
+        .and_then(Value::as_u64)
+        .unwrap_or(0);
+    let model = body
+        .get("model")
+        .and_then(Value::as_str)
+        .or_else(|| report_context.get("mapped_model").and_then(Value::as_str))
+        .or_else(|| report_context.get("model").and_then(Value::as_str))
+        .unwrap_or("unknown");
+    let id = body
+        .get("id")
+        .and_then(Value::as_str)
+        .unwrap_or("msg-local-finalize");
+
+    Some(json!({
+        "id": id,
+        "type": "message",
+        "role": "assistant",
+        "model": model,
+        "content": content,
+        "stop_reason": stop_reason,
+        "usage": {
+            "input_tokens": input_tokens,
+            "output_tokens": output_tokens,
+        }
+    }))
+}
+
+fn extract_openai_assistant_text(content: Option<&Value>) -> Option<String> {
+    match content? {
+        Value::Null => Some(String::new()),
+        Value::String(text) => Some(text.clone()),
+        Value::Array(parts) => {
+            let mut text = String::new();
+            for part in parts {
+                let part = part.as_object()?;
+                let part_type = part
+                    .get("type")
+                    .and_then(Value::as_str)
+                    .unwrap_or_default()
+                    .trim()
+                    .to_ascii_lowercase();
+                if matches!(part_type.as_str(), "text" | "output_text") {
+                    if let Some(piece) = part.get("text").and_then(Value::as_str) {
+                        text.push_str(piece);
+                    }
+                }
+            }
+            Some(text)
+        }
+        _ => None,
+    }
+}
+
+fn parse_openai_function_arguments(arguments: Option<&Value>) -> Option<Value> {
+    match arguments.cloned().unwrap_or(Value::Object(Map::new())) {
+        Value::String(text) => serde_json::from_str(&text)
+            .ok()
+            .or(Some(Value::String(text))),
+        other => Some(other),
+    }
+}
+
 pub(crate) fn aggregate_claude_stream_sync_response(body: &[u8]) -> Option<Value> {
    let events = parse_stream_json_events(body)?;
    if events.is_empty() {
--- a/apps/aether-gateway/src/ai_pipeline/finalize/standard/claude/cli.rs
+++ b/apps/aether-gateway/src/ai_pipeline/finalize/standard/claude/cli.rs
@@ -1,15 +1,13 @@
 use base64::Engine as _;

-use crate::gateway::kiro_stream::KiroToClaudeCliStreamState;
-
-use super::super::chat::aggregate_claude_stream_sync_response;
-use super::super::common::{
+use super::aggregate_claude_stream_sync_response;
+use super::*;
+use crate::gateway::ai_pipeline::finalize::common::{
    build_generated_tool_call_id, build_local_success_outcome, canonicalize_tool_arguments,
-    local_finalize_allows_envelope, parse_stream_json_events, unwrap_local_finalize_response_value,
+    local_finalize_allows_envelope, unwrap_local_finalize_response_value,
    LocalCoreSyncFinalizeOutcome,
 };
-use super::super::*;
-use super::openai::build_openai_cli_response;
+use crate::gateway::ai_pipeline::finalize::standard::build_openai_cli_response;

 pub(crate) fn maybe_build_local_claude_cli_stream_sync_response(
    trace_id: &str,
@@ -69,36 +67,11 @@ pub(crate) fn maybe_build_local_claude_cli_stream_sync_response(

 fn aggregate_provider_claude_cli_stream_sync_response(
    body_bytes: &[u8],
-    report_context: &Value,
+    _report_context: &Value,
 ) -> Result<Option<Value>, GatewayError> {
-    if report_context
-        .get("envelope_name")
-        .and_then(Value::as_str)
-        .is_some_and(|value| value.eq_ignore_ascii_case("kiro:generateAssistantResponse"))
-    {
-        let mut rewriter = KiroToClaudeCliStreamState::new(report_context);
-        let mut rewritten = rewriter.push_chunk(report_context, body_bytes)?;
-        rewritten.extend(rewriter.finish(report_context)?);
-        if rewritten_stream_contains_error_event(&rewritten) {
-            return Ok(None);
-        }
-        return Ok(aggregate_claude_stream_sync_response(&rewritten));
-    }
-
    Ok(aggregate_claude_stream_sync_response(body_bytes))
 }

-fn rewritten_stream_contains_error_event(body: &[u8]) -> bool {
-    parse_stream_json_events(body).is_some_and(|events| {
-        events.iter().any(|event| {
-            event
-                .get("type")
-                .and_then(Value::as_str)
-                .is_some_and(|value| value.eq_ignore_ascii_case("error"))
-        })
-    })
-}
-
 pub(crate) fn convert_claude_cli_response_to_openai_cli(
    body_json: &Value,
    report_context: &Value,
--- a/apps/aether-gateway/src/ai_pipeline/finalize/standard/gemini/chat.rs
+++ b/apps/aether-gateway/src/ai_pipeline/finalize/standard/gemini/chat.rs
@@ -1,11 +1,11 @@
 use base64::Engine as _;

-use super::super::common::{
+use super::*;
+use crate::gateway::ai_pipeline::finalize::common::{
    build_generated_tool_call_id, build_local_success_outcome, canonicalize_tool_arguments,
    local_finalize_allows_envelope, parse_stream_json_events, unwrap_local_finalize_response_value,
    LocalCoreSyncFinalizeOutcome,
 };
-use super::super::*;

 pub(crate) fn maybe_build_local_gemini_stream_sync_response(
    trace_id: &str,
@@ -212,6 +212,137 @@ pub(crate) fn convert_gemini_chat_response_to_openai_chat(
    }))
 }

+pub(crate) fn convert_openai_chat_response_to_gemini_chat(
+    body_json: &Value,
+    report_context: &Value,
+) -> Option<Value> {
+    let body = body_json.as_object()?;
+    let choices = body.get("choices")?.as_array()?;
+    let first_choice = choices.first()?.as_object()?;
+    let message = first_choice.get("message")?.as_object()?;
+    let mut parts = Vec::new();
+
+    if let Some(text) = extract_openai_assistant_text(message.get("content")) {
+        if !text.trim().is_empty() {
+            parts.push(json!({ "text": text }));
+        }
+    }
+    if let Some(tool_call_values) = message.get("tool_calls").and_then(Value::as_array) {
+        for (index, tool_call) in tool_call_values.iter().enumerate() {
+            let tool_call = tool_call.as_object()?;
+            let function = tool_call.get("function")?.as_object()?;
+            let tool_name = function
+                .get("name")
+                .and_then(Value::as_str)
+                .map(str::trim)
+                .filter(|value| !value.is_empty())?;
+            let call_id = tool_call
+                .get("id")
+                .and_then(Value::as_str)
+                .filter(|value| !value.is_empty())
+                .map(ToOwned::to_owned)
+                .unwrap_or_else(|| build_generated_tool_call_id(index));
+            parts.push(json!({
+                "functionCall": {
+                    "id": call_id,
+                    "name": tool_name,
+                    "args": parse_openai_function_arguments(function.get("arguments"))?,
+                }
+            }));
+        }
+    }
+    if parts.is_empty() {
+        parts.push(json!({ "text": "" }));
+    }
+
+    let usage = body.get("usage").and_then(Value::as_object);
+    let prompt_tokens = usage
+        .and_then(|value| value.get("prompt_tokens"))
+        .and_then(Value::as_u64)
+        .unwrap_or(0);
+    let completion_tokens = usage
+        .and_then(|value| value.get("completion_tokens"))
+        .and_then(Value::as_u64)
+        .unwrap_or(0);
+    let total_tokens = usage
+        .and_then(|value| value.get("total_tokens"))
+        .and_then(Value::as_u64)
+        .unwrap_or(prompt_tokens + completion_tokens);
+    let mut finish_reason = match first_choice.get("finish_reason").and_then(Value::as_str) {
+        Some("stop") | None => "STOP",
+        Some("length") => "MAX_TOKENS",
+        Some("content_filter") => "SAFETY",
+        Some("tool_calls") | Some("function_call") => "STOP",
+        Some(other) => other,
+    };
+    if parts.iter().any(|part| part.get("functionCall").is_some()) {
+        finish_reason = "STOP";
+    }
+    let model = body
+        .get("model")
+        .and_then(Value::as_str)
+        .or_else(|| report_context.get("mapped_model").and_then(Value::as_str))
+        .or_else(|| report_context.get("model").and_then(Value::as_str))
+        .unwrap_or("unknown");
+    let response_id = body
+        .get("id")
+        .and_then(Value::as_str)
+        .unwrap_or("resp-local-finalize");
+
+    Some(json!({
+        "responseId": response_id,
+        "modelVersion": model,
+        "candidates": [{
+            "content": {
+                "role": "model",
+                "parts": parts,
+            },
+            "finishReason": finish_reason,
+            "index": 0,
+        }],
+        "usageMetadata": {
+            "promptTokenCount": prompt_tokens,
+            "candidatesTokenCount": completion_tokens,
+            "totalTokenCount": total_tokens,
+        }
+    }))
+}
+
+fn extract_openai_assistant_text(content: Option<&Value>) -> Option<String> {
+    match content? {
+        Value::Null => Some(String::new()),
+        Value::String(text) => Some(text.clone()),
+        Value::Array(parts) => {
+            let mut text = String::new();
+            for part in parts {
+                let part = part.as_object()?;
+                let part_type = part
+                    .get("type")
+                    .and_then(Value::as_str)
+                    .unwrap_or_default()
+                    .trim()
+                    .to_ascii_lowercase();
+                if matches!(part_type.as_str(), "text" | "output_text") {
+                    if let Some(piece) = part.get("text").and_then(Value::as_str) {
+                        text.push_str(piece);
+                    }
+                }
+            }
+            Some(text)
+        }
+        _ => None,
+    }
+}
+
+fn parse_openai_function_arguments(arguments: Option<&Value>) -> Option<Value> {
+    match arguments.cloned().unwrap_or(Value::Object(Map::new())) {
+        Value::String(text) => serde_json::from_str(&text)
+            .ok()
+            .or(Some(Value::String(text))),
+        other => Some(other),
+    }
+}
+
 pub(crate) fn aggregate_gemini_stream_sync_response(body: &[u8]) -> Option<Value> {
    let events = parse_stream_json_events(body)?;
    if events.is_empty() {
@@ -220,6 +351,7 @@ pub(crate) fn aggregate_gemini_stream_sync_response(body: &[u8]) -> Option<Value

    let mut candidates: BTreeMap<usize, Value> = BTreeMap::new();
    let mut response_id: Option<Value> = None;
+    let mut private_response_id: Option<Value> = None;
    let mut model_version: Option<Value> = None;
    let mut usage_metadata: Option<Value> = None;
    let mut prompt_feedback: Option<Value> = None;
@@ -230,6 +362,9 @@ pub(crate) fn aggregate_gemini_stream_sync_response(body: &[u8]) -> Option<Value
        if let Some(id) = raw_event_object.get("responseId") {
            response_id = Some(id.clone());
        }
+        if let Some(id) = raw_event_object.get("_v1internal_response_id") {
+            private_response_id = Some(id.clone());
+        }
        let event_object = if let Some(response) = raw_event_object
            .get("response")
            .and_then(Value::as_object)
@@ -242,6 +377,9 @@ pub(crate) fn aggregate_gemini_stream_sync_response(body: &[u8]) -> Option<Value
        if let Some(id) = event_object.get("responseId") {
            response_id = Some(id.clone());
        }
+        if let Some(id) = event_object.get("_v1internal_response_id") {
+            private_response_id = Some(id.clone());
+        }
        if let Some(version) = event_object.get("modelVersion") {
            model_version = Some(version.clone());
        }
@@ -277,6 +415,9 @@ pub(crate) fn aggregate_gemini_stream_sync_response(body: &[u8]) -> Option<Value
    if let Some(response_id) = response_id {
        response.insert("responseId".to_string(), response_id);
    }
+    if let Some(private_response_id) = private_response_id {
+        response.insert("_v1internal_response_id".to_string(), private_response_id);
+    }
    response.insert(
        "candidates".to_string(),
        Value::Array(candidates.into_values().collect()),
--- a/apps/aether-gateway/src/ai_pipeline/finalize/standard/gemini/cli.rs
+++ b/apps/aether-gateway/src/ai_pipeline/finalize/standard/gemini/cli.rs
@@ -1,13 +1,13 @@
 use base64::Engine as _;

-use super::super::chat::aggregate_gemini_stream_sync_response;
-use super::super::common::{
+use super::aggregate_gemini_stream_sync_response;
+use super::*;
+use crate::gateway::ai_pipeline::finalize::common::{
    build_generated_tool_call_id, build_local_success_outcome, canonicalize_tool_arguments,
    local_finalize_allows_envelope, unwrap_local_finalize_response_value,
    LocalCoreSyncFinalizeOutcome,
 };
-use super::super::*;
-use super::openai::build_openai_cli_response;
+use crate::gateway::ai_pipeline::finalize::standard::build_openai_cli_response;

 pub(crate) fn maybe_build_local_gemini_cli_stream_sync_response(
    trace_id: &str,
@@ -128,6 +128,7 @@ pub(crate) fn convert_gemini_cli_response_to_openai_cli(
        .unwrap_or("unknown");
    let response_id = body
        .get("responseId")
+        .or_else(|| body.get("_v1internal_response_id"))
        .and_then(Value::as_str)
        .unwrap_or("resp-local-finalize");

--- a/apps/aether-gateway/src/ai_pipeline/finalize/standard/mod.rs
+++ b/apps/aether-gateway/src/ai_pipeline/finalize/standard/mod.rs
@@ -0,0 +1,141 @@
+//! Standard finalize surface for standard contract sync/stream compilation.
+
+use std::collections::BTreeMap;
+
+use serde_json::Value;
+use serde_json::{json, Map};
+
+use crate::gateway::{GatewayControlDecision, GatewayError, GatewaySyncReportRequest};
+
+#[path = "claude/chat.rs"]
+mod claude_chat;
+#[path = "claude/cli.rs"]
+mod claude_cli;
+#[path = "gemini/chat.rs"]
+mod gemini_chat;
+#[path = "gemini/cli.rs"]
+mod gemini_cli;
+#[path = "openai/chat.rs"]
+mod openai_chat;
+#[path = "openai/chat_stream.rs"]
+mod openai_chat_stream;
+#[path = "openai/cli.rs"]
+mod openai_cli;
+#[path = "openai/cli_stream.rs"]
+mod openai_cli_stream;
+
+#[path = "stream.rs"]
+mod stream;
+
+pub(crate) use crate::gateway::ai_pipeline::conversion::response::{
+    build_openai_cli_response, convert_claude_chat_response_to_openai_chat,
+    convert_claude_cli_response_to_openai_cli, convert_gemini_chat_response_to_openai_chat,
+    convert_gemini_cli_response_to_openai_cli, convert_openai_chat_response_to_claude_chat,
+    convert_openai_chat_response_to_gemini_chat, convert_openai_chat_response_to_openai_cli,
+    convert_openai_cli_response_to_openai_chat,
+};
+pub(crate) use claude_chat::{
+    aggregate_claude_stream_sync_response, maybe_build_local_claude_stream_sync_response,
+    maybe_build_local_claude_sync_response,
+};
+pub(crate) use claude_cli::maybe_build_local_claude_cli_stream_sync_response;
+pub(crate) use gemini_chat::{
+    aggregate_gemini_stream_sync_response, maybe_build_local_gemini_stream_sync_response,
+    maybe_build_local_gemini_sync_response,
+};
+pub(crate) use gemini_cli::maybe_build_local_gemini_cli_stream_sync_response;
+pub(crate) use openai_chat::{
+    aggregate_openai_chat_stream_sync_response, maybe_build_local_openai_chat_cross_format_stream_sync_response,
+    maybe_build_local_openai_chat_cross_format_sync_response,
+    maybe_build_local_openai_chat_stream_sync_response,
+    maybe_build_local_openai_chat_sync_response,
+};
+pub(crate) use openai_chat_stream::{
+    ClaudeToOpenAIChatStreamState, GeminiToOpenAIChatStreamState, OpenAICliToOpenAIChatStreamState,
+};
+pub(crate) use openai_cli::{
+    aggregate_openai_cli_stream_sync_response, maybe_build_local_openai_cli_cross_format_stream_sync_response,
+    maybe_build_local_openai_cli_cross_format_sync_response,
+    maybe_build_local_openai_cli_stream_sync_response,
+};
+pub(crate) use openai_cli_stream::BufferedCliConversionStreamState;
+pub(crate) use stream::BufferedStandardConversionStreamState;
+
+pub(crate) fn aggregate_standard_chat_stream_sync_response(
+    body: &[u8],
+    provider_api_format: &str,
+) -> Option<Value> {
+    match provider_api_format.trim().to_ascii_lowercase().as_str() {
+        "openai:chat" => aggregate_openai_chat_stream_sync_response(body),
+        "openai:cli" | "openai:compact" => aggregate_openai_cli_stream_sync_response(body),
+        "claude:chat" | "claude:cli" => aggregate_claude_stream_sync_response(body),
+        "gemini:chat" | "gemini:cli" => aggregate_gemini_stream_sync_response(body),
+        _ => None,
+    }
+}
+
+pub(crate) fn convert_standard_chat_response(
+    body_json: &Value,
+    provider_api_format: &str,
+    client_api_format: &str,
+    report_context: &Value,
+) -> Option<Value> {
+    let canonical = match provider_api_format.trim().to_ascii_lowercase().as_str() {
+        "openai:chat" => body_json.clone(),
+        "openai:cli" | "openai:compact" => {
+            convert_openai_cli_response_to_openai_chat(body_json, report_context)?
+        }
+        "claude:chat" | "claude:cli" => {
+            convert_claude_chat_response_to_openai_chat(body_json, report_context)?
+        }
+        "gemini:chat" | "gemini:cli" => {
+            convert_gemini_chat_response_to_openai_chat(body_json, report_context)?
+        }
+        _ => return None,
+    };
+
+    match client_api_format.trim().to_ascii_lowercase().as_str() {
+        "openai:chat" => Some(canonical),
+        "claude:chat" => convert_openai_chat_response_to_claude_chat(&canonical, report_context),
+        "gemini:chat" => convert_openai_chat_response_to_gemini_chat(&canonical, report_context),
+        _ => None,
+    }
+}
+
+pub(crate) fn aggregate_standard_cli_stream_sync_response(
+    body: &[u8],
+    provider_api_format: &str,
+) -> Option<Value> {
+    aggregate_standard_chat_stream_sync_response(body, provider_api_format)
+}
+
+pub(crate) fn convert_standard_cli_response(
+    body_json: &Value,
+    provider_api_format: &str,
+    client_api_format: &str,
+    report_context: &Value,
+) -> Option<Value> {
+    let canonical = match provider_api_format.trim().to_ascii_lowercase().as_str() {
+        "openai:cli" | "openai:compact" => {
+            convert_openai_cli_response_to_openai_chat(body_json, report_context)?
+        }
+        _ => convert_standard_chat_response(
+            body_json,
+            provider_api_format,
+            "openai:chat",
+            report_context,
+        )?,
+    };
+
+    match client_api_format.trim().to_ascii_lowercase().as_str() {
+        "openai:cli" => {
+            convert_openai_chat_response_to_openai_cli(&canonical, report_context, false)
+        }
+        "openai:compact" => {
+            convert_openai_chat_response_to_openai_cli(&canonical, report_context, true)
+        }
+        "claude:cli" => convert_openai_chat_response_to_claude_chat(&canonical, report_context),
+        "gemini:cli" => convert_openai_chat_response_to_gemini_chat(&canonical, report_context),
+        _ => None,
+    }
+}
--- a/apps/aether-gateway/src/ai_pipeline/finalize/standard/openai/chat.rs
+++ b/apps/aether-gateway/src/ai_pipeline/finalize/standard/openai/chat.rs
@@ -1,15 +1,18 @@
 use base64::Engine as _;

-use super::super::common::{
-    build_local_success_outcome, build_local_success_outcome_with_conversion_report,
+use super::*;
+use super::{
+    aggregate_claude_stream_sync_response, aggregate_gemini_stream_sync_response,
+    aggregate_openai_cli_stream_sync_response, convert_claude_chat_response_to_openai_chat,
+    convert_gemini_chat_response_to_openai_chat,
+};
+use crate::gateway::ai_pipeline::finalize::common::{
+    build_generated_tool_call_id, build_local_success_outcome,
+    build_local_success_outcome_with_conversion_report, canonicalize_tool_arguments,
    local_finalize_allows_envelope, unwrap_local_finalize_response_value,
    LocalCoreSyncFinalizeOutcome,
 };
-use super::super::*;
-use super::{
-    aggregate_claude_stream_sync_response, aggregate_gemini_stream_sync_response,
-    convert_claude_chat_response_to_openai_chat, convert_gemini_chat_response_to_openai_chat,
-};
+use crate::gateway::ai_pipeline::conversion::sync_chat_response_conversion_kind;

 #[derive(Debug, Default)]
 struct OpenAIChatChoiceState {
@@ -162,6 +165,11 @@ pub(crate) fn maybe_build_local_openai_chat_cross_format_stream_sync_response(
    if client_api_format != "openai:chat" || !local_finalize_allows_envelope(report_context) {
        return Ok(None);
    }
+    let Some(conversion_kind) =
+        sync_chat_response_conversion_kind(&provider_api_format, &client_api_format)
+    else {
+        return Ok(None);
+    };

    let Some(body_base64) = payload.body_base64.as_deref() else {
        return Ok(None);
@@ -170,8 +178,9 @@ pub(crate) fn maybe_build_local_openai_chat_cross_format_stream_sync_response(
        .decode(body_base64)
        .map_err(|err| GatewayError::Internal(err.to_string()))?;
    let aggregated = match provider_api_format.as_str() {
-        "claude:chat" => aggregate_claude_stream_sync_response(&body_bytes),
-        "gemini:chat" => aggregate_gemini_stream_sync_response(&body_bytes),
+        "claude:chat" | "claude:cli" => aggregate_claude_stream_sync_response(&body_bytes),
+        "gemini:chat" | "gemini:cli" => aggregate_gemini_stream_sync_response(&body_bytes),
+        "openai:cli" | "openai:compact" => aggregate_openai_cli_stream_sync_response(&body_bytes),
        _ => None,
    };
    let Some(aggregated) = aggregated else {
@@ -181,8 +190,15 @@ pub(crate) fn maybe_build_local_openai_chat_cross_format_stream_sync_response(
        return Ok(None);
    };
    let converted = match provider_api_format.as_str() {
-        "claude:chat" => convert_claude_chat_response_to_openai_chat(&aggregated, report_context),
-        "gemini:chat" => convert_gemini_chat_response_to_openai_chat(&aggregated, report_context),
+        "claude:chat" | "claude:cli" => {
+            convert_claude_chat_response_to_openai_chat(&aggregated, report_context)
+        }
+        "gemini:chat" | "gemini:cli" => {
+            convert_gemini_chat_response_to_openai_chat(&aggregated, report_context)
+        }
+        "openai:cli" | "openai:compact" => {
+            convert_openai_cli_response_to_openai_chat(&aggregated, report_context)
+        }
        _ => None,
    };
    let Some(converted) = converted else {
@@ -221,6 +237,11 @@ pub(crate) fn maybe_build_local_openai_chat_cross_format_sync_response(
    if client_api_format != "openai:chat" || !local_finalize_allows_envelope(report_context) {
        return Ok(None);
    }
+    let Some(conversion_kind) =
+        sync_chat_response_conversion_kind(&provider_api_format, &client_api_format)
+    else {
+        return Ok(None);
+    };

    let Some(body_json) = payload.body_json.as_ref() else {
        return Ok(None);
@@ -230,8 +251,15 @@ pub(crate) fn maybe_build_local_openai_chat_cross_format_sync_response(
        return Ok(None);
    };
    let converted = match provider_api_format.as_str() {
-        "claude:chat" => convert_claude_chat_response_to_openai_chat(&body_json, report_context),
-        "gemini:chat" => convert_gemini_chat_response_to_openai_chat(&body_json, report_context),
+        "claude:chat" | "claude:cli" => {
+            convert_claude_chat_response_to_openai_chat(&body_json, report_context)
+        }
+        "gemini:chat" | "gemini:cli" => {
+            convert_gemini_chat_response_to_openai_chat(&body_json, report_context)
+        }
+        "openai:cli" | "openai:compact" => {
+            convert_openai_cli_response_to_openai_chat(&body_json, report_context)
+        }
        _ => None,
    };
    let Some(converted) = converted else {
@@ -243,6 +271,138 @@ pub(crate) fn maybe_build_local_openai_chat_cross_format_sync_response(
    )?))
 }

+pub(crate) fn convert_openai_cli_response_to_openai_chat(
+    body_json: &Value,
+    report_context: &Value,
+) -> Option<Value> {
+    let body = body_json.as_object()?;
+    let mut text = String::new();
+    let mut tool_calls = Vec::new();
+
+    if let Some(output_items) = body.get("output").and_then(Value::as_array) {
+        for (index, item) in output_items.iter().enumerate() {
+            let item_object = item.as_object()?;
+            let item_type = item_object
+                .get("type")
+                .and_then(Value::as_str)
+                .unwrap_or_default()
+                .trim()
+                .to_ascii_lowercase();
+            match item_type.as_str() {
+                "message" => {
+                    if let Some(content) = item_object.get("content").and_then(Value::as_array) {
+                        for part in content {
+                            let part_object = part.as_object()?;
+                            let part_type = part_object
+                                .get("type")
+                                .and_then(Value::as_str)
+                                .unwrap_or_default()
+                                .trim()
+                                .to_ascii_lowercase();
+                            if matches!(part_type.as_str(), "output_text" | "text") {
+                                if let Some(piece) = part_object.get("text").and_then(Value::as_str)
+                                {
+                                    text.push_str(piece);
+                                }
+                            }
+                        }
+                    }
+                }
+                "function_call" => {
+                    let tool_name = item_object
+                        .get("name")
+                        .and_then(Value::as_str)
+                        .map(str::trim)
+                        .filter(|value| !value.is_empty())?;
+                    let tool_id = item_object
+                        .get("call_id")
+                        .and_then(Value::as_str)
+                        .filter(|value| !value.is_empty())
+                        .or_else(|| {
+                            item_object
+                                .get("id")
+                                .and_then(Value::as_str)
+                                .filter(|value| !value.is_empty())
+                        })
+                        .map(ToOwned::to_owned)
+                        .unwrap_or_else(|| build_generated_tool_call_id(index));
+                    tool_calls.push(json!({
+                        "id": tool_id,
+                        "type": "function",
+                        "function": {
+                            "name": tool_name,
+                            "arguments": canonicalize_tool_arguments(item_object.get("arguments").cloned()),
+                        }
+                    }));
+                }
+                "output_text" | "text" => {
+                    if let Some(piece) = item_object.get("text").and_then(Value::as_str) {
+                        text.push_str(piece);
+                    }
+                }
+                _ => {}
+            }
+        }
+    }
+
+    let finish_reason = if tool_calls.is_empty() {
+        Some("stop")
+    } else {
+        Some("tool_calls")
+    };
+    let model = body
+        .get("model")
+        .and_then(Value::as_str)
+        .or_else(|| report_context.get("mapped_model").and_then(Value::as_str))
+        .or_else(|| report_context.get("model").and_then(Value::as_str))
+        .unwrap_or("unknown");
+    let id = body
+        .get("id")
+        .and_then(Value::as_str)
+        .unwrap_or("chatcmpl-local-openai-cli");
+
+    let usage = body.get("usage").and_then(Value::as_object);
+    let prompt_tokens = usage
+        .and_then(|value| value.get("input_tokens"))
+        .and_then(Value::as_u64)
+        .unwrap_or(0);
+    let completion_tokens = usage
+        .and_then(|value| value.get("output_tokens"))
+        .and_then(Value::as_u64)
+        .unwrap_or(0);
+    let total_tokens = usage
+        .and_then(|value| value.get("total_tokens"))
+        .and_then(Value::as_u64)
+        .unwrap_or(prompt_tokens + completion_tokens);
+
+    let mut message = Map::new();
+    message.insert("role".to_string(), Value::String("assistant".to_string()));
+    if text.is_empty() && !tool_calls.is_empty() {
+        message.insert("content".to_string(), Value::Null);
+    } else {
+        message.insert("content".to_string(), Value::String(text));
+    }
+    if !tool_calls.is_empty() {
+        message.insert("tool_calls".to_string(), Value::Array(tool_calls));
+    }
+
+    Some(json!({
+        "id": id,
+        "object": "chat.completion",
+        "model": model,
+        "choices": [{
+            "index": 0,
+            "message": Value::Object(message),
+            "finish_reason": finish_reason,
+        }],
+        "usage": {
+            "prompt_tokens": prompt_tokens,
+            "completion_tokens": completion_tokens,
+            "total_tokens": total_tokens,
+        }
+    }))
+}
+
 pub(crate) fn aggregate_openai_chat_stream_sync_response(body: &[u8]) -> Option<Value> {
    let text = std::str::from_utf8(body).ok()?;
    let mut response_id: Option<String> = None;
--- a/apps/aether-gateway/src/ai_pipeline/finalize/standard/openai/chat_stream.rs
+++ b/apps/aether-gateway/src/ai_pipeline/finalize/standard/openai/chat_stream.rs
@@ -1,19 +1,29 @@
 use serde_json::{json, Map, Value};

+use crate::gateway::ai_pipeline::finalize::sse::{
+    encode_done_sse, encode_json_sse, map_claude_stop_reason,
+};
 use crate::gateway::GatewayError;

-use super::common::{encode_done_sse, encode_json_sse, map_claude_stop_reason};
-use super::*;
+use super::{
+    aggregate_claude_stream_sync_response, aggregate_gemini_stream_sync_response,
+    convert_openai_cli_response_to_openai_chat,
+};

 #[derive(Default)]
-pub(super) struct ClaudeToOpenAIChatStreamState {
+pub(crate) struct ClaudeToOpenAIChatStreamState {
    raw: Vec<u8>,
    message_id: Option<String>,
    model: Option<String>,
 }

 #[derive(Default)]
-pub(super) struct GeminiToOpenAIChatStreamState {
+pub(crate) struct GeminiToOpenAIChatStreamState {
+    raw: Vec<u8>,
+}
+
+#[derive(Default)]
+pub(crate) struct OpenAICliToOpenAIChatStreamState {
    raw: Vec<u8>,
 }

@@ -243,7 +253,7 @@ fn convert_gemini_aggregated_to_openai_chunk(
 }

 impl ClaudeToOpenAIChatStreamState {
-    pub(super) fn transform_line(
+    pub(crate) fn transform_line(
        &mut self,
        report_context: &Value,
        line: Vec<u8>,
@@ -379,7 +389,7 @@ impl ClaudeToOpenAIChatStreamState {
        }
    }

-    pub(super) fn finish(&mut self) -> Vec<u8> {
+    pub(crate) fn finish(&mut self) -> Vec<u8> {
        if self.raw.is_empty() {
            return Vec::new();
        }
@@ -399,7 +409,7 @@ impl ClaudeToOpenAIChatStreamState {
 }

 impl GeminiToOpenAIChatStreamState {
-    pub(super) fn transform_line(
+    pub(crate) fn transform_line(
        &mut self,
        _report_context: &Value,
        line: Vec<u8>,
@@ -408,7 +418,7 @@ impl GeminiToOpenAIChatStreamState {
        Ok(Vec::new())
    }

-    pub(super) fn finish(&mut self, report_context: &Value) -> Result<Vec<u8>, GatewayError> {
+    pub(crate) fn finish(&mut self, report_context: &Value) -> Result<Vec<u8>, GatewayError> {
        if self.raw.is_empty() {
            return Ok(Vec::new());
        }
@@ -426,3 +436,64 @@ impl GeminiToOpenAIChatStreamState {
        Ok(out)
    }
 }
+
+impl OpenAICliToOpenAIChatStreamState {
+    pub(crate) fn transform_line(
+        &mut self,
+        _report_context: &Value,
+        line: Vec<u8>,
+    ) -> Result<Vec<u8>, GatewayError> {
+        self.raw.extend_from_slice(&line);
+        Ok(Vec::new())
+    }
+
+    pub(crate) fn finish(&mut self, report_context: &Value) -> Result<Vec<u8>, GatewayError> {
+        if self.raw.is_empty() {
+            return Ok(Vec::new());
+        }
+        let aggregated = crate::gateway::ai_pipeline::finalize::standard::aggregate_openai_cli_stream_sync_response(&self.raw);
+        self.raw.clear();
+        let Some(aggregated) = aggregated else {
+            return Ok(Vec::new());
+        };
+        let Some(chat_response) =
+            convert_openai_cli_response_to_openai_chat(&aggregated, report_context)
+        else {
+            return Ok(Vec::new());
+        };
+        let Some(chat_object) = chat_response.as_object() else {
+            return Ok(Vec::new());
+        };
+        let Some(choice) = chat_object
+            .get("choices")
+            .and_then(Value::as_array)
+            .and_then(|choices| choices.first())
+            .and_then(Value::as_object)
+        else {
+            return Ok(Vec::new());
+        };
+        let Some(message) = choice.get("message").and_then(Value::as_object) else {
+            return Ok(Vec::new());
+        };
+        let content = message
+            .get("content")
+            .and_then(Value::as_str)
+            .unwrap_or_default()
+            .to_string();
+        let tool_calls = message.get("tool_calls").and_then(Value::as_array).cloned();
+        let finish_reason = choice.get("finish_reason").and_then(Value::as_str);
+        let id = chat_object
+            .get("id")
+            .and_then(Value::as_str)
+            .unwrap_or("chatcmpl-local-stream");
+        let model = chat_object
+            .get("model")
+            .and_then(Value::as_str)
+            .unwrap_or("unknown");
+
+        let chunk = build_openai_chat_chunk(id, model, content, tool_calls, finish_reason);
+        let mut out = encode_json_sse(None, &chunk)?;
+        out.extend(encode_done_sse());
+        Ok(out)
+    }
+}
--- a/apps/aether-gateway/src/ai_pipeline/finalize/standard/openai/cli.rs
+++ b/apps/aether-gateway/src/ai_pipeline/finalize/standard/openai/cli.rs
@@ -0,0 +1,692 @@
+use base64::Engine as _;
+
+use super::*;
+use super::{
+    aggregate_claude_stream_sync_response, aggregate_gemini_stream_sync_response,
+    convert_claude_cli_response_to_openai_cli, convert_gemini_cli_response_to_openai_cli,
+};
+use crate::gateway::ai_pipeline::finalize::common::{
+    build_local_success_outcome, build_local_success_outcome_with_conversion_report,
+    canonicalize_tool_arguments, local_finalize_allows_envelope,
+    unwrap_local_finalize_response_value, LocalCoreSyncFinalizeOutcome,
+};
+use crate::gateway::ai_pipeline::conversion::sync_cli_response_conversion_kind;
+
+pub(crate) fn maybe_build_local_openai_cli_stream_sync_response(
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    payload: &GatewaySyncReportRequest,
+) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
+    if let Some(response) =
+        maybe_build_local_openai_cli_direct_stream_sync_response(trace_id, decision, payload)?
+    {
+        return Ok(Some(response));
+    }
+    if let Some(response) = maybe_build_local_openai_cli_openai_family_stream_sync_response(
+        trace_id, decision, payload,
+    )? {
+        return Ok(Some(response));
+    }
+    maybe_build_local_openai_cli_direct_sync_response(trace_id, decision, payload)
+}
+
+pub(crate) fn maybe_build_local_openai_cli_cross_format_stream_sync_response(
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    payload: &GatewaySyncReportRequest,
+) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
+    if let Some(response) =
+        maybe_build_local_openai_cli_antigravity_cross_format_stream_sync_response(
+            trace_id, decision, payload,
+        )?
+    {
+        return Ok(Some(response));
+    }
+
+    if !matches!(
+        payload.report_kind.as_str(),
+        "openai_cli_sync_finalize" | "openai_compact_sync_finalize"
+    ) || payload.status_code >= 400
+    {
+        return Ok(None);
+    }
+
+    let Some(report_context) = payload.report_context.as_ref() else {
+        return Ok(None);
+    };
+    let provider_api_format = report_context
+        .get("provider_api_format")
+        .and_then(Value::as_str)
+        .unwrap_or_default()
+        .trim()
+        .to_ascii_lowercase();
+    let client_api_format = report_context
+        .get("client_api_format")
+        .and_then(Value::as_str)
+        .unwrap_or_default()
+        .trim()
+        .to_ascii_lowercase();
+    let _has_envelope = report_context
+        .get("has_envelope")
+        .and_then(Value::as_bool)
+        .unwrap_or(false);
+    if !matches!(client_api_format.as_str(), "openai:cli" | "openai:compact")
+        || !local_finalize_allows_envelope(report_context)
+    {
+        return Ok(None);
+    }
+    let Some(conversion_kind) =
+        sync_cli_response_conversion_kind(&provider_api_format, &client_api_format)
+    else {
+        return Ok(None);
+    };
+
+    let Some(body_base64) = payload.body_base64.as_deref() else {
+        return Ok(None);
+    };
+    let body_bytes = base64::engine::general_purpose::STANDARD
+        .decode(body_base64)
+        .map_err(|err| GatewayError::Internal(err.to_string()))?;
+    let aggregated = match provider_api_format.as_str() {
+        "openai:cli" | "openai:compact" => aggregate_openai_cli_stream_sync_response(&body_bytes),
+        "claude:chat" | "claude:cli" => aggregate_claude_stream_sync_response(&body_bytes),
+        "gemini:chat" | "gemini:cli" => aggregate_gemini_stream_sync_response(&body_bytes),
+        _ => None,
+    };
+    let Some(aggregated) = aggregated else {
+        return Ok(None);
+    };
+    let Some(aggregated) = unwrap_local_finalize_response_value(aggregated, report_context)? else {
+        return Ok(None);
+    };
+    let converted = match provider_api_format.as_str() {
+        "openai:cli" | "openai:compact" => Some(aggregated.clone()),
+        "claude:chat" | "claude:cli" => {
+            convert_claude_cli_response_to_openai_cli(&aggregated, report_context)
+        }
+        "gemini:chat" | "gemini:cli" => {
+            convert_gemini_cli_response_to_openai_cli(&aggregated, report_context)
+        }
+        _ => None,
+    };
+    let Some(converted) = converted else {
+        return Ok(None);
+    };
+
+    Ok(Some(build_local_success_outcome_with_conversion_report(
+        trace_id, decision, payload, converted, aggregated,
+    )?))
+}
+
+pub(crate) fn maybe_build_local_openai_cli_cross_format_sync_response(
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    payload: &GatewaySyncReportRequest,
+) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
+    if let Some(response) = maybe_build_local_openai_cli_antigravity_cross_format_sync_response(
+        trace_id, decision, payload,
+    )? {
+        return Ok(Some(response));
+    }
+
+    if !matches!(
+        payload.report_kind.as_str(),
+        "openai_cli_sync_finalize" | "openai_compact_sync_finalize"
+    ) || payload.status_code >= 400
+    {
+        return Ok(None);
+    }
+
+    let Some(report_context) = payload.report_context.as_ref() else {
+        return Ok(None);
+    };
+    let provider_api_format = report_context
+        .get("provider_api_format")
+        .and_then(Value::as_str)
+        .unwrap_or_default()
+        .trim()
+        .to_ascii_lowercase();
+    let client_api_format = report_context
+        .get("client_api_format")
+        .and_then(Value::as_str)
+        .unwrap_or_default()
+        .trim()
+        .to_ascii_lowercase();
+    if !matches!(client_api_format.as_str(), "openai:cli" | "openai:compact")
+        || !local_finalize_allows_envelope(report_context)
+    {
+        return Ok(None);
+    }
+    let Some(conversion_kind) =
+        sync_cli_response_conversion_kind(&provider_api_format, &client_api_format)
+    else {
+        return Ok(None);
+    };
+
+    let Some(body_json) = payload.body_json.as_ref() else {
+        return Ok(None);
+    };
+    let Some(body_json) = unwrap_local_finalize_response_value(body_json.clone(), report_context)?
+    else {
+        return Ok(None);
+    };
+    let converted = match provider_api_format.as_str() {
+        "openai:cli" | "openai:compact" => Some(body_json.clone()),
+        "claude:chat" | "claude:cli" => {
+            convert_claude_cli_response_to_openai_cli(&body_json, report_context)
+        }
+        "gemini:chat" | "gemini:cli" => {
+            convert_gemini_cli_response_to_openai_cli(&body_json, report_context)
+        }
+        _ => None,
+    };
+    let Some(converted) = converted else {
+        return Ok(None);
+    };
+
+    Ok(Some(build_local_success_outcome_with_conversion_report(
+        trace_id, decision, payload, converted, body_json,
+    )?))
+}
+
+fn maybe_build_local_openai_cli_direct_sync_response(
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    payload: &GatewaySyncReportRequest,
+) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
+    if !matches!(
+        payload.report_kind.as_str(),
+        "openai_cli_sync_finalize" | "openai_compact_sync_finalize"
+    ) || payload.status_code >= 400
+    {
+        return Ok(None);
+    }
+
+    let Some(report_context) = payload.report_context.as_ref() else {
+        return Ok(None);
+    };
+    let provider_api_format = report_context
+        .get("provider_api_format")
+        .and_then(Value::as_str)
+        .unwrap_or_default()
+        .trim()
+        .to_ascii_lowercase();
+    let client_api_format = report_context
+        .get("client_api_format")
+        .and_then(Value::as_str)
+        .unwrap_or_default()
+        .trim()
+        .to_ascii_lowercase();
+
+    if !local_finalize_allows_envelope(report_context)
+        || !is_openai_cli_family_api_format(provider_api_format.as_str())
+        || !is_openai_cli_family_api_format(client_api_format.as_str())
+    {
+        return Ok(None);
+    }
+
+    let Some(body_json) = payload.body_json.as_ref() else {
+        return Ok(None);
+    };
+    let Some(body_json) = unwrap_local_finalize_response_value(body_json.clone(), report_context)?
+    else {
+        return Ok(None);
+    };
+
+    Ok(Some(build_local_success_outcome(
+        trace_id, decision, payload, body_json,
+    )?))
+}
+
+fn maybe_build_local_openai_cli_direct_stream_sync_response(
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    payload: &GatewaySyncReportRequest,
+) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
+    if !matches!(
+        payload.report_kind.as_str(),
+        "openai_cli_sync_finalize" | "openai_compact_sync_finalize"
+    ) || payload.status_code >= 400
+    {
+        return Ok(None);
+    }
+
+    let Some(report_context) = payload.report_context.as_ref() else {
+        return Ok(None);
+    };
+    let provider_api_format = report_context
+        .get("provider_api_format")
+        .and_then(Value::as_str)
+        .unwrap_or_default()
+        .trim()
+        .to_ascii_lowercase();
+    let client_api_format = report_context
+        .get("client_api_format")
+        .and_then(Value::as_str)
+        .unwrap_or_default()
+        .trim()
+        .to_ascii_lowercase();
+    let needs_conversion = report_context
+        .get("needs_conversion")
+        .and_then(Value::as_bool)
+        .unwrap_or(false);
+    if !local_finalize_allows_envelope(report_context) {
+        return Ok(None);
+    }
+    if !matches!(
+        provider_api_format.as_str(),
+        "openai:cli" | "openai:compact"
+    ) || provider_api_format != client_api_format
+        || needs_conversion
+    {
+        return Ok(None);
+    }
+
+    let Some(body_base64) = payload.body_base64.as_deref() else {
+        return Ok(None);
+    };
+    let body_bytes = base64::engine::general_purpose::STANDARD
+        .decode(body_base64)
+        .map_err(|err| GatewayError::Internal(err.to_string()))?;
+    let body_json = match aggregate_openai_cli_stream_sync_response(&body_bytes) {
+        Some(body_json) => body_json,
+        None => return Ok(None),
+    };
+    let Some(body_json) = unwrap_local_finalize_response_value(body_json, report_context)? else {
+        return Ok(None);
+    };
+
+    Ok(Some(build_local_success_outcome(
+        trace_id, decision, payload, body_json,
+    )?))
+}
+
+fn maybe_build_local_openai_cli_openai_family_stream_sync_response(
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    payload: &GatewaySyncReportRequest,
+) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
+    if !matches!(
+        payload.report_kind.as_str(),
+        "openai_cli_sync_finalize" | "openai_compact_sync_finalize"
+    ) || payload.status_code >= 400
+    {
+        return Ok(None);
+    }
+
+    let Some(report_context) = payload.report_context.as_ref() else {
+        return Ok(None);
+    };
+    let provider_api_format = report_context
+        .get("provider_api_format")
+        .and_then(Value::as_str)
+        .unwrap_or_default()
+        .trim()
+        .to_ascii_lowercase();
+    let client_api_format = report_context
+        .get("client_api_format")
+        .and_then(Value::as_str)
+        .unwrap_or_default()
+        .trim()
+        .to_ascii_lowercase();
+    if !local_finalize_allows_envelope(report_context)
+        || !is_openai_cli_family_api_format(provider_api_format.as_str())
+        || !is_openai_cli_family_api_format(client_api_format.as_str())
+        || provider_api_format == client_api_format
+    {
+        return Ok(None);
+    }
+
+    let Some(body_base64) = payload.body_base64.as_deref() else {
+        return Ok(None);
+    };
+    let body_bytes = base64::engine::general_purpose::STANDARD
+        .decode(body_base64)
+        .map_err(|err| GatewayError::Internal(err.to_string()))?;
+    let Some(body_json) = aggregate_openai_cli_stream_sync_response(&body_bytes) else {
+        return Ok(None);
+    };
+    let Some(body_json) = unwrap_local_finalize_response_value(body_json, report_context)? else {
+        return Ok(None);
+    };
+
+    Ok(Some(build_local_success_outcome(
+        trace_id, decision, payload, body_json,
+    )?))
+}
+
+fn maybe_build_local_openai_cli_antigravity_cross_format_stream_sync_response(
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    payload: &GatewaySyncReportRequest,
+) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
+    if !matches!(
+        payload.report_kind.as_str(),
+        "openai_cli_sync_finalize" | "openai_compact_sync_finalize"
+    ) || payload.status_code >= 400
+    {
+        return Ok(None);
+    }
+
+    let Some(report_context) = payload.report_context.as_ref() else {
+        return Ok(None);
+    };
+    let provider_api_format = report_context
+        .get("provider_api_format")
+        .and_then(Value::as_str)
+        .unwrap_or_default()
+        .trim()
+        .to_ascii_lowercase();
+    let client_api_format = report_context
+        .get("client_api_format")
+        .and_then(Value::as_str)
+        .unwrap_or_default()
+        .trim()
+        .to_ascii_lowercase();
+    if provider_api_format != "gemini:cli"
+        || !is_openai_cli_family_api_format(client_api_format.as_str())
+        || !is_antigravity_v1internal_envelope(report_context)
+        || !local_finalize_allows_envelope(report_context)
+    {
+        return Ok(None);
+    }
+
+    let Some(body_base64) = payload.body_base64.as_deref() else {
+        return Ok(None);
+    };
+    let body_bytes = base64::engine::general_purpose::STANDARD
+        .decode(body_base64)
+        .map_err(|err| GatewayError::Internal(err.to_string()))?;
+    let Some(aggregated) = aggregate_gemini_stream_sync_response(&body_bytes) else {
+        return Ok(None);
+    };
+    let Some(provider_body_json) =
+        unwrap_cli_conversion_response_value(aggregated, report_context)?
+    else {
+        return Ok(None);
+    };
+    let Some(converted) =
+        convert_gemini_cli_response_to_openai_cli(&provider_body_json, report_context)
+    else {
+        return Ok(None);
+    };
+
+    Ok(Some(build_local_success_outcome_with_conversion_report(
+        trace_id,
+        decision,
+        payload,
+        converted,
+        provider_body_json,
+    )?))
+}
+
+fn maybe_build_local_openai_cli_antigravity_cross_format_sync_response(
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    payload: &GatewaySyncReportRequest,
+) -> Result<Option<LocalCoreSyncFinalizeOutcome>, GatewayError> {
+    if !matches!(
+        payload.report_kind.as_str(),
+        "openai_cli_sync_finalize" | "openai_compact_sync_finalize"
+    ) || payload.status_code >= 400
+    {
+        return Ok(None);
+    }
+
+    let Some(report_context) = payload.report_context.as_ref() else {
+        return Ok(None);
+    };
+    let provider_api_format = report_context
+        .get("provider_api_format")
+        .and_then(Value::as_str)
+        .unwrap_or_default()
+        .trim()
+        .to_ascii_lowercase();
+    let client_api_format = report_context
+        .get("client_api_format")
+        .and_then(Value::as_str)
+        .unwrap_or_default()
+        .trim()
+        .to_ascii_lowercase();
+    if provider_api_format != "gemini:cli"
+        || !is_openai_cli_family_api_format(client_api_format.as_str())
+        || !is_antigravity_v1internal_envelope(report_context)
+        || !local_finalize_allows_envelope(report_context)
+    {
+        return Ok(None);
+    }
+
+    let Some(body_json) = payload.body_json.as_ref() else {
+        return Ok(None);
+    };
+    let Some(provider_body_json) =
+        unwrap_cli_conversion_response_value(body_json.clone(), report_context)?
+    else {
+        return Ok(None);
+    };
+    let Some(converted) =
+        convert_gemini_cli_response_to_openai_cli(&provider_body_json, report_context)
+    else {
+        return Ok(None);
+    };
+
+    Ok(Some(build_local_success_outcome_with_conversion_report(
+        trace_id,
+        decision,
+        payload,
+        converted,
+        provider_body_json,
+    )?))
+}
+
+fn unwrap_cli_conversion_response_value(
+    data: Value,
+    report_context: &Value,
+) -> Result<Option<Value>, GatewayError> {
+    if !is_antigravity_v1internal_envelope(report_context) {
+        return unwrap_local_finalize_response_value(data, report_context);
+    }
+
+    let mut unwrapped = if let Some(response) = data
+        .get("response")
+        .and_then(Value::as_object)
+        .filter(|response| !response.contains_key("response"))
+    {
+        let mut response = response.clone();
+        if let Some(response_id) = data.get("responseId").cloned() {
+            response
+                .entry("responseId".to_string())
+                .or_insert(response_id);
+        }
+        Value::Object(response)
+    } else {
+        data
+    };
+
+    if let Some(object) = unwrapped.as_object_mut() {
+        if !object.contains_key("responseId") {
+            if let Some(response_id) = object.get("_v1internal_response_id").cloned() {
+                object.insert("responseId".to_string(), response_id);
+            }
+        }
+    }
+
+    Ok(Some(unwrapped))
+}
+
+fn is_antigravity_v1internal_envelope(report_context: &Value) -> bool {
+    report_context
+        .get("has_envelope")
+        .and_then(Value::as_bool)
+        .unwrap_or(false)
+        && report_context
+            .get("envelope_name")
+            .and_then(Value::as_str)
+            .is_some_and(|value| value.eq_ignore_ascii_case("antigravity:v1internal"))
+}
+
+fn is_openai_cli_family_api_format(api_format: &str) -> bool {
+    matches!(api_format, "openai:cli" | "openai:compact")
+}
+
+pub(crate) fn build_openai_cli_response(
+    response_id: &str,
+    model: &str,
+    text: &str,
+    function_calls: Vec<Value>,
+    prompt_tokens: u64,
+    output_tokens: u64,
+    total_tokens: u64,
+) -> Value {
+    let mut output = Vec::new();
+    if !text.is_empty() {
+        output.push(json!({
+            "type": "message",
+            "id": format!("{response_id}_msg"),
+            "role": "assistant",
+            "status": "completed",
+            "content": [{
+                "type": "output_text",
+                "text": text,
+                "annotations": []
+            }]
+        }));
+    }
+    output.extend(function_calls);
+    json!({
+        "id": response_id,
+        "object": "response",
+        "status": "completed",
+        "model": model,
+        "output": output,
+        "usage": {
+            "input_tokens": prompt_tokens,
+            "output_tokens": output_tokens,
+            "total_tokens": total_tokens,
+        }
+    })
+}
+
+pub(crate) fn convert_openai_chat_response_to_openai_cli(
+    body_json: &Value,
+    report_context: &Value,
+    compact: bool,
+) -> Option<Value> {
+    let body = body_json.as_object()?;
+    let choices = body.get("choices")?.as_array()?;
+    let first_choice = choices.first()?.as_object()?;
+    let message = first_choice.get("message")?.as_object()?;
+    let mut text = String::new();
+    match message.get("content") {
+        Some(Value::String(value)) => text.push_str(value),
+        Some(Value::Array(parts)) => {
+            for part in parts {
+                let part = part.as_object()?;
+                let part_type = part
+                    .get("type")
+                    .and_then(Value::as_str)
+                    .unwrap_or_default()
+                    .trim()
+                    .to_ascii_lowercase();
+                if matches!(part_type.as_str(), "text" | "output_text") {
+                    if let Some(piece) = part.get("text").and_then(Value::as_str) {
+                        text.push_str(piece);
+                    }
+                }
+            }
+        }
+        Some(Value::Null) | None => {}
+        _ => return None,
+    }
+
+    let mut function_calls = Vec::new();
+    if let Some(tool_call_values) = message.get("tool_calls").and_then(Value::as_array) {
+        for tool_call in tool_call_values {
+            let tool_call = tool_call.as_object()?;
+            let function = tool_call.get("function")?.as_object()?;
+            let tool_name = function
+                .get("name")
+                .and_then(Value::as_str)
+                .map(str::trim)
+                .filter(|value| !value.is_empty())?;
+            function_calls.push(json!({
+                "type": "function_call",
+                "id": tool_call.get("id").cloned().unwrap_or(Value::Null),
+                "call_id": tool_call.get("id").cloned().unwrap_or(Value::Null),
+                "name": tool_name,
+                "arguments": canonicalize_tool_arguments(function.get("arguments").cloned()),
+            }));
+        }
+    }
+
+    let usage = body.get("usage").and_then(Value::as_object);
+    let prompt_tokens = usage
+        .and_then(|value| value.get("prompt_tokens"))
+        .and_then(Value::as_u64)
+        .unwrap_or(0);
+    let output_tokens = usage
+        .and_then(|value| value.get("completion_tokens"))
+        .and_then(Value::as_u64)
+        .unwrap_or(0);
+    let total_tokens = usage
+        .and_then(|value| value.get("total_tokens"))
+        .and_then(Value::as_u64)
+        .unwrap_or(prompt_tokens + output_tokens);
+    let response_id = if compact {
+        body.get("id")
+            .and_then(Value::as_str)
+            .map(|value| value.replace("chatcmpl", "resp"))
+            .unwrap_or_else(|| "resp-local-finalize".to_string())
+    } else {
+        body.get("id")
+            .and_then(Value::as_str)
+            .map(|value| value.replace("chatcmpl", "resp"))
+            .unwrap_or_else(|| "resp-local-finalize".to_string())
+    };
+    let model = body
+        .get("model")
+        .and_then(Value::as_str)
+        .or_else(|| report_context.get("mapped_model").and_then(Value::as_str))
+        .or_else(|| report_context.get("model").and_then(Value::as_str))
+        .unwrap_or("unknown");
+
+    Some(build_openai_cli_response(
+        &response_id,
+        model,
+        &text,
+        function_calls,
+        prompt_tokens,
+        output_tokens,
+        total_tokens,
+    ))
+}
+
+pub(crate) fn aggregate_openai_cli_stream_sync_response(body: &[u8]) -> Option<Value> {
+    let text = std::str::from_utf8(body).ok()?;
+
+    for raw_line in text.lines() {
+        let line = raw_line.trim_matches('\r').trim();
+        if line.is_empty() || line.starts_with(':') || line.starts_with("event:") {
+            continue;
+        }
+        let Some(data_line) = line.strip_prefix("data:") else {
+            continue;
+        };
+        let data_line = data_line.trim();
+        if data_line.is_empty() || data_line == "[DONE]" {
+            continue;
+        }
+
+        let event: Value = serde_json::from_str(data_line).ok()?;
+        let event_type = event
+            .get("type")
+            .and_then(Value::as_str)
+            .unwrap_or_default();
+        if event_type == "response.completed" {
+            let response = event.get("response")?.as_object()?.clone();
+            return Some(Value::Object(response));
+        }
+    }
+
+    None
+}
--- a/apps/aether-gateway/src/ai_pipeline/finalize/standard/openai/cli_stream.rs
+++ b/apps/aether-gateway/src/ai_pipeline/finalize/standard/openai/cli_stream.rs
@@ -1,21 +1,20 @@
 use serde_json::{json, Value};

+use crate::gateway::ai_pipeline::finalize::sse::encode_json_sse;
 use crate::gateway::GatewayError;

-use super::common::encode_json_sse;
-
 #[derive(Default)]
-pub(super) struct BufferedCliConversionStreamState {
+pub(crate) struct BufferedCliConversionStreamState {
    raw: Vec<u8>,
 }

 impl BufferedCliConversionStreamState {
-    pub(super) fn transform_line(&mut self, line: Vec<u8>) -> Result<Vec<u8>, GatewayError> {
+    pub(crate) fn transform_line(&mut self, line: Vec<u8>) -> Result<Vec<u8>, GatewayError> {
        self.raw.extend_from_slice(&line);
        Ok(Vec::new())
    }

-    pub(super) fn finish<AggregateFn, ConvertFn>(
+    pub(crate) fn finish<AggregateFn, ConvertFn>(
        &mut self,
        report_context: &Value,
        aggregate: AggregateFn,
--- a/apps/aether-gateway/src/ai_pipeline/finalize/standard/stream.rs
+++ b/apps/aether-gateway/src/ai_pipeline/finalize/standard/stream.rs
@@ -0,0 +1,347 @@
+use serde_json::{json, Value};
+
+use crate::gateway::ai_pipeline::finalize::standard::{
+    aggregate_standard_chat_stream_sync_response, aggregate_standard_cli_stream_sync_response,
+    convert_standard_chat_response, convert_standard_cli_response,
+};
+use crate::gateway::ai_pipeline::finalize::sse::{encode_done_sse, encode_json_sse};
+use crate::gateway::ai_pipeline::private_response::transform_provider_private_stream_line as transform_envelope_line;
+use crate::gateway::ai_pipeline::private_surfaces::provider_adaptation_should_unwrap_stream_envelope;
+use crate::gateway::GatewayError;
+
+#[derive(Default)]
+pub(crate) struct BufferedStandardConversionStreamState {
+    raw: Vec<u8>,
+}
+
+impl BufferedStandardConversionStreamState {
+    pub(crate) fn transform_line(
+        &mut self,
+        report_context: &Value,
+        line: Vec<u8>,
+    ) -> Result<Vec<u8>, GatewayError> {
+        if should_unwrap_envelope(report_context) {
+            self.raw
+                .extend(transform_envelope_line(report_context, line)?);
+        } else {
+            self.raw.extend_from_slice(&line);
+        }
+        Ok(Vec::new())
+    }
+
+    pub(crate) fn finish_as_chat(
+        &mut self,
+        report_context: &Value,
+    ) -> Result<Vec<u8>, GatewayError> {
+        let provider_api_format = report_context
+            .get("provider_api_format")
+            .and_then(Value::as_str)
+            .unwrap_or_default()
+            .trim()
+            .to_ascii_lowercase();
+        let client_api_format = report_context
+            .get("client_api_format")
+            .and_then(Value::as_str)
+            .unwrap_or_default()
+            .trim()
+            .to_ascii_lowercase();
+        if self.raw.is_empty() {
+            return Ok(Vec::new());
+        }
+        let aggregated =
+            aggregate_standard_chat_stream_sync_response(&self.raw, provider_api_format.as_str());
+        self.raw.clear();
+        let Some(aggregated) = aggregated else {
+            return Ok(Vec::new());
+        };
+        let Some(converted) = convert_standard_chat_response(
+            &aggregated,
+            provider_api_format.as_str(),
+            client_api_format.as_str(),
+            report_context,
+        ) else {
+            return Ok(Vec::new());
+        };
+        emit_chat_stream_for_client_format(&converted, client_api_format.as_str())
+    }
+
+    pub(crate) fn finish_as_cli(
+        &mut self,
+        report_context: &Value,
+    ) -> Result<Vec<u8>, GatewayError> {
+        let provider_api_format = report_context
+            .get("provider_api_format")
+            .and_then(Value::as_str)
+            .unwrap_or_default()
+            .trim()
+            .to_ascii_lowercase();
+        let client_api_format = report_context
+            .get("client_api_format")
+            .and_then(Value::as_str)
+            .unwrap_or_default()
+            .trim()
+            .to_ascii_lowercase();
+        if self.raw.is_empty() {
+            return Ok(Vec::new());
+        }
+        let aggregated =
+            aggregate_standard_cli_stream_sync_response(&self.raw, provider_api_format.as_str());
+        self.raw.clear();
+        let Some(aggregated) = aggregated else {
+            return Ok(Vec::new());
+        };
+        let Some(converted) = convert_standard_cli_response(
+            &aggregated,
+            provider_api_format.as_str(),
+            client_api_format.as_str(),
+            report_context,
+        ) else {
+            return Ok(Vec::new());
+        };
+        emit_cli_stream_for_client_format(&converted, client_api_format.as_str())
+    }
+}
+
+fn should_unwrap_envelope(report_context: &Value) -> bool {
+    let envelope_name = report_context
+        .get("envelope_name")
+        .and_then(Value::as_str)
+        .unwrap_or_default();
+    let provider_api_format = report_context
+        .get("provider_api_format")
+        .and_then(Value::as_str)
+        .unwrap_or_default();
+    provider_adaptation_should_unwrap_stream_envelope(envelope_name, provider_api_format)
+}
+
+fn emit_chat_stream_for_client_format(
+    response_body: &Value,
+    client_api_format: &str,
+) -> Result<Vec<u8>, GatewayError> {
+    match client_api_format {
+        "openai:chat" => emit_openai_chat_stream(response_body),
+        "claude:chat" | "claude:cli" => emit_claude_message_stream(response_body),
+        "gemini:chat" | "gemini:cli" => encode_json_sse(None, response_body),
+        _ => Ok(Vec::new()),
+    }
+}
+
+fn emit_cli_stream_for_client_format(
+    response_body: &Value,
+    client_api_format: &str,
+) -> Result<Vec<u8>, GatewayError> {
+    match client_api_format {
+        "openai:cli" | "openai:compact" => encode_json_sse(
+            Some("response.completed"),
+            &json!({
+                "type": "response.completed",
+                "response": response_body,
+            }),
+        ),
+        "claude:cli" => emit_claude_message_stream(response_body),
+        "gemini:cli" => encode_json_sse(None, response_body),
+        _ => Ok(Vec::new()),
+    }
+}
+
+fn emit_openai_chat_stream(response_body: &Value) -> Result<Vec<u8>, GatewayError> {
+    let body = match response_body.as_object() {
+        Some(body) => body,
+        None => return Ok(Vec::new()),
+    };
+    let choice = match body
+        .get("choices")
+        .and_then(Value::as_array)
+        .and_then(|choices| choices.first())
+        .and_then(Value::as_object)
+    {
+        Some(choice) => choice,
+        None => return Ok(Vec::new()),
+    };
+    let message = match choice.get("message").and_then(Value::as_object) {
+        Some(message) => message,
+        None => return Ok(Vec::new()),
+    };
+    let content = match extract_openai_chat_content_text(message.get("content")) {
+        Some(content) => content,
+        None => return Ok(Vec::new()),
+    };
+    let mut delta = serde_json::Map::new();
+    delta.insert("role".to_string(), Value::String("assistant".to_string()));
+    if !content.is_empty() {
+        delta.insert("content".to_string(), Value::String(content));
+    } else if message.get("tool_calls").is_none() {
+        delta.insert("content".to_string(), Value::String(String::new()));
+    }
+    if let Some(tool_calls) = message.get("tool_calls").and_then(Value::as_array) {
+        delta.insert("tool_calls".to_string(), Value::Array(tool_calls.clone()));
+    }
+    let chunk = json!({
+        "id": body.get("id").cloned().unwrap_or_else(|| Value::String("chatcmpl-local-stream".to_string())),
+        "object": "chat.completion.chunk",
+        "model": body.get("model").cloned().unwrap_or_else(|| Value::String("unknown".to_string())),
+        "choices": [{
+            "index": choice.get("index").cloned().unwrap_or_else(|| Value::from(0_u64)),
+            "delta": Value::Object(delta),
+            "finish_reason": choice.get("finish_reason").cloned().unwrap_or(Value::Null),
+        }]
+    });
+    let mut out = encode_json_sse(None, &chunk)?;
+    out.extend(encode_done_sse());
+    Ok(out)
+}
+
+fn emit_claude_message_stream(response_body: &Value) -> Result<Vec<u8>, GatewayError> {
+    let body = match response_body.as_object() {
+        Some(body) => body,
+        None => return Ok(Vec::new()),
+    };
+    let message_id = body
+        .get("id")
+        .cloned()
+        .unwrap_or_else(|| Value::String("msg-local-stream".to_string()));
+    let model = body
+        .get("model")
+        .cloned()
+        .unwrap_or_else(|| Value::String("unknown".to_string()));
+    let content_blocks = match body.get("content").and_then(Value::as_array) {
+        Some(content) => content,
+        None => return Ok(Vec::new()),
+    };
+
+    let mut out = encode_json_sse(
+        Some("message_start"),
+        &json!({
+            "type": "message_start",
+            "message": {
+                "id": message_id,
+                "type": "message",
+                "role": "assistant",
+                "model": model,
+                "content": [],
+                "stop_reason": Value::Null,
+                "stop_sequence": Value::Null,
+            }
+        }),
+    )?;
+
+    for (index, block) in content_blocks.iter().enumerate() {
+        let Some(block_object) = block.as_object() else {
+            continue;
+        };
+        match block_object
+            .get("type")
+            .and_then(Value::as_str)
+            .unwrap_or("text")
+        {
+            "text" => {
+                out.extend(encode_json_sse(
+                    Some("content_block_start"),
+                    &json!({
+                        "type": "content_block_start",
+                        "index": index,
+                        "content_block": {
+                            "type": "text",
+                            "text": "",
+                        }
+                    }),
+                )?);
+                if let Some(text) = block_object.get("text").and_then(Value::as_str) {
+                    if !text.is_empty() {
+                        out.extend(encode_json_sse(
+                            Some("content_block_delta"),
+                            &json!({
+                                "type": "content_block_delta",
+                                "index": index,
+                                "delta": {
+                                    "type": "text_delta",
+                                    "text": text,
+                                }
+                            }),
+                        )?);
+                    }
+                }
+                out.extend(encode_json_sse(
+                    Some("content_block_stop"),
+                    &json!({
+                        "type": "content_block_stop",
+                        "index": index,
+                    }),
+                )?);
+            }
+            "tool_use" => {
+                out.extend(encode_json_sse(
+                    Some("content_block_start"),
+                    &json!({
+                        "type": "content_block_start",
+                        "index": index,
+                        "content_block": block_object,
+                    }),
+                )?);
+                out.extend(encode_json_sse(
+                    Some("content_block_stop"),
+                    &json!({
+                        "type": "content_block_stop",
+                        "index": index,
+                    }),
+                )?);
+            }
+            _ => {}
+        }
+    }
+
+    let mut delta = serde_json::Map::new();
+    delta.insert(
+        "stop_reason".to_string(),
+        body.get("stop_reason").cloned().unwrap_or(Value::Null),
+    );
+    if let Some(stop_sequence) = body.get("stop_sequence").cloned() {
+        delta.insert("stop_sequence".to_string(), stop_sequence);
+    }
+    let mut message_delta = serde_json::Map::new();
+    message_delta.insert(
+        "type".to_string(),
+        Value::String("message_delta".to_string()),
+    );
+    message_delta.insert("delta".to_string(), Value::Object(delta));
+    if let Some(usage) = body.get("usage").cloned() {
+        message_delta.insert("usage".to_string(), usage);
+    }
+    out.extend(encode_json_sse(
+        Some("message_delta"),
+        &Value::Object(message_delta),
+    )?);
+    out.extend(encode_json_sse(
+        Some("message_stop"),
+        &json!({
+            "type": "message_stop",
+        }),
+    )?);
+    Ok(out)
+}
+
+fn extract_openai_chat_content_text(content: Option<&Value>) -> Option<String> {
+    match content? {
+        Value::Null => Some(String::new()),
+        Value::String(text) => Some(text.clone()),
+        Value::Array(parts) => {
+            let mut text = String::new();
+            for part in parts {
+                let part = part.as_object()?;
+                let part_type = part
+                    .get("type")
+                    .and_then(Value::as_str)
+                    .unwrap_or_default()
+                    .trim()
+                    .to_ascii_lowercase();
+                if matches!(part_type.as_str(), "text" | "output_text") {
+                    if let Some(piece) = part.get("text").and_then(Value::as_str) {
+                        text.push_str(piece);
+                    }
+                }
+            }
+            Some(text)
+        }
+        _ => None,
+    }
+}
--- a/apps/aether-gateway/src/ai_pipeline/finalize/tests_stream.rs
+++ b/apps/aether-gateway/src/ai_pipeline/finalize/tests_stream.rs
@@ -180,6 +180,35 @@ fn gemini_to_openai_chat_stream_rewriter_buffers_and_converts_function_call() {
    assert!(output_text.contains("data: [DONE]"));
 }

+#[test]
+fn openai_cli_to_openai_chat_stream_rewriter_buffers_and_converts_completed_event() {
+    let report_context = json!({
+        "provider_api_format": "openai:cli",
+        "client_api_format": "openai:chat",
+        "needs_conversion": true,
+        "mapped_model": "gpt-5.4",
+    });
+    let mut rewriter =
+        maybe_build_local_stream_rewriter(Some(&report_context)).expect("rewriter should exist");
+    let output = rewriter
+        .push_chunk(
+            concat!(
+                "event: response.completed\n",
+                "data: {\"type\":\"response.completed\",\"response\":{\"id\":\"resp_cli_stream_123\",\"object\":\"response\",\"model\":\"gpt-5.4\",\"status\":\"completed\",\"output\":[{\"type\":\"message\",\"id\":\"msg_cli_stream_123\",\"role\":\"assistant\",\"status\":\"completed\",\"content\":[{\"type\":\"output_text\",\"text\":\"Hello Codex\",\"annotations\":[]}]}],\"usage\":{\"input_tokens\":1,\"output_tokens\":2,\"total_tokens\":3}}}\n\n"
+            )
+            .as_bytes(),
+        )
+        .expect("rewrite should succeed");
+    assert!(output.is_empty());
+    let output_text = String::from_utf8(rewriter.finish().expect("finish should succeed"))
+        .expect("utf8 should decode");
+    assert!(output_text.contains("\"object\":\"chat.completion.chunk\""));
+    assert!(output_text.contains("\"role\":\"assistant\""));
+    assert!(output_text.contains("\"content\":\"Hello Codex\""));
+    assert!(output_text.contains("\"finish_reason\":\"stop\""));
+    assert!(output_text.contains("data: [DONE]"));
+}
+
 #[test]
 fn antigravity_gemini_to_openai_chat_stream_rewriter_unwraps_and_converts_function_call() {
    let report_context = json!({
@@ -389,3 +418,60 @@ fn gemini_to_openai_compact_stream_rewriter_converts_function_call_to_completed_
    assert!(output_text.contains("\"name\":\"get_weather\""));
    assert!(output_text.contains("\\\"location\\\":\\\"Tokyo\\\""));
 }
+
+#[test]
+fn openai_chat_to_claude_chat_stream_rewriter_converts_via_standard_matrix() {
+    let report_context = json!({
+        "provider_api_format": "openai:chat",
+        "client_api_format": "claude:chat",
+        "needs_conversion": true,
+        "mapped_model": "gpt-5",
+    });
+    let mut rewriter =
+        maybe_build_local_stream_rewriter(Some(&report_context)).expect("rewriter should exist");
+    let output = rewriter
+        .push_chunk(
+            concat!(
+                "data: {\"id\":\"chatcmpl_std_claude_123\",\"object\":\"chat.completion.chunk\",\"created\":1,\"model\":\"gpt-5\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"Hello Claude\"},\"finish_reason\":null}]}\n\n",
+                "data: {\"id\":\"chatcmpl_std_claude_123\",\"object\":\"chat.completion.chunk\",\"model\":\"gpt-5\",\"choices\":[{\"index\":0,\"delta\":{},\"finish_reason\":\"stop\"}],\"usage\":{\"prompt_tokens\":1,\"completion_tokens\":2,\"total_tokens\":3}}\n\n",
+                "data: [DONE]\n\n"
+            )
+            .as_bytes(),
+        )
+        .expect("rewrite should succeed");
+    assert!(output.is_empty());
+    let output_text = String::from_utf8(rewriter.finish().expect("finish should succeed"))
+        .expect("utf8 should decode");
+    assert!(output_text.contains("event: message_start"));
+    assert!(output_text.contains("event: content_block_delta"));
+    assert!(output_text.contains("\"text\":\"Hello Claude\""));
+    assert!(output_text.contains("event: message_stop"));
+}
+
+#[test]
+fn openai_chat_to_gemini_cli_stream_rewriter_converts_via_standard_matrix() {
+    let report_context = json!({
+        "provider_api_format": "openai:chat",
+        "client_api_format": "gemini:cli",
+        "needs_conversion": true,
+        "mapped_model": "gpt-5",
+    });
+    let mut rewriter =
+        maybe_build_local_stream_rewriter(Some(&report_context)).expect("rewriter should exist");
+    let output = rewriter
+        .push_chunk(
+            concat!(
+                "data: {\"id\":\"chatcmpl_std_gemini_cli_123\",\"object\":\"chat.completion.chunk\",\"created\":1,\"model\":\"gpt-5\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"Hello Gemini CLI\"},\"finish_reason\":null}]}\n\n",
+                "data: {\"id\":\"chatcmpl_std_gemini_cli_123\",\"object\":\"chat.completion.chunk\",\"model\":\"gpt-5\",\"choices\":[{\"index\":0,\"delta\":{},\"finish_reason\":\"stop\"}],\"usage\":{\"prompt_tokens\":2,\"completion_tokens\":3,\"total_tokens\":5}}\n\n",
+                "data: [DONE]\n\n"
+            )
+            .as_bytes(),
+        )
+        .expect("rewrite should succeed");
+    assert!(output.is_empty());
+    let output_text = String::from_utf8(rewriter.finish().expect("finish should succeed"))
+        .expect("utf8 should decode");
+    assert!(output_text.contains("\"responseId\":\"chatcmpl_std_gemini_cli_123\""));
+    assert!(output_text.contains("\"candidates\""));
+    assert!(output_text.contains("\"text\":\"Hello Gemini CLI\""));
+}
--- a/apps/aether-gateway/src/ai_pipeline/finalize/tests_sync.rs
+++ b/apps/aether-gateway/src/ai_pipeline/finalize/tests_sync.rs
@@ -4,19 +4,13 @@ use base64::Engine as _;
 use serde_json::json;

 use super::{
-    chat::{
-        aggregate_claude_stream_sync_response, aggregate_gemini_stream_sync_response,
-        aggregate_openai_chat_stream_sync_response, convert_claude_chat_response_to_openai_chat,
-        convert_gemini_chat_response_to_openai_chat,
-    },
-    cli::{
-        aggregate_openai_cli_stream_sync_response, convert_claude_cli_response_to_openai_cli,
-        convert_gemini_cli_response_to_openai_cli,
-    },
+    aggregate_claude_stream_sync_response, aggregate_gemini_stream_sync_response,
+    aggregate_openai_chat_stream_sync_response, aggregate_openai_cli_stream_sync_response,
+    convert_claude_chat_response_to_openai_chat, convert_claude_cli_response_to_openai_cli,
+    convert_gemini_chat_response_to_openai_chat, convert_gemini_cli_response_to_openai_cli,
    maybe_build_local_core_sync_finalize_response,
 };
-use crate::gateway::executor::GatewaySyncReportRequest;
-use crate::gateway::GatewayControlDecision;
+use crate::gateway::{GatewayControlDecision, GatewaySyncReportRequest};

 fn test_decision() -> GatewayControlDecision {
    GatewayControlDecision {
@@ -26,7 +20,7 @@ fn test_decision() -> GatewayControlDecision {
        route_family: Some("openai".to_string()),
        route_kind: Some("compact".to_string()),
        auth_endpoint_signature: Some("openai:compact".to_string()),
-        executor_candidate: true,
+        execution_runtime_candidate: true,
        auth_context: None,
        admin_principal: None,
        local_auth_rejection: None,
@@ -898,6 +892,144 @@ fn local_finalize_handles_openai_cli_openai_family_sync_response_even_when_conve
    );
 }

+#[test]
+fn local_finalize_handles_openai_chat_stream_response_from_openai_chat() {
+    let body = concat!(
+        "data: {\"id\":\"chatcmpl_stream_direct_123\",\"object\":\"chat.completion.chunk\",\"created\":1,",
+        "\"model\":\"gpt-5\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"Hello\"},\"finish_reason\":null}]}\n\n",
+        "data: {\"id\":\"chatcmpl_stream_direct_123\",\"object\":\"chat.completion.chunk\",",
+        "\"model\":\"gpt-5\",\"choices\":[{\"index\":0,\"delta\":{\"content\":\" world\"},\"finish_reason\":null}]}\n\n",
+        "data: {\"id\":\"chatcmpl_stream_direct_123\",\"object\":\"chat.completion.chunk\",",
+        "\"model\":\"gpt-5\",\"choices\":[{\"index\":0,\"delta\":{},\"finish_reason\":\"stop\"}]}\n\n",
+        "data: [DONE]\n\n",
+    );
+    let payload = GatewaySyncReportRequest {
+        trace_id: "trace-openai-chat-stream-sync-123".to_string(),
+        report_kind: "openai_chat_sync_finalize".to_string(),
+        report_context: Some(json!({
+            "client_api_format": "openai:chat",
+            "provider_api_format": "openai:chat",
+            "model": "gpt-5",
+            "mapped_model": "gpt-5",
+            "needs_conversion": false,
+            "has_envelope": false,
+        })),
+        status_code: 200,
+        headers: BTreeMap::from([("content-type".to_string(), "text/event-stream".to_string())]),
+        body_json: None,
+        client_body_json: None,
+        body_base64: Some(base64::engine::general_purpose::STANDARD.encode(body.as_bytes())),
+        telemetry: None,
+    };
+
+    let outcome = maybe_build_local_core_sync_finalize_response(
+        "trace-openai-chat-stream-sync-123",
+        &test_decision(),
+        &payload,
+    )
+    .expect("local finalize should succeed")
+    .expect("local finalize should match");
+
+    let report = outcome
+        .background_report
+        .expect("stream finalize should downgrade to success report");
+    assert_eq!(report.report_kind, "openai_chat_sync_success");
+    let provider_body = report.body_json.expect("provider body should exist");
+    assert_eq!(provider_body["object"], "chat.completion");
+    assert_eq!(
+        provider_body["choices"][0]["message"]["content"],
+        "Hello world"
+    );
+}
+
+#[test]
+fn local_finalize_handles_openai_cli_cross_format_stream_response_from_gemini() {
+    let body = concat!(
+        "data: {\"responseId\":\"resp_cli_stream_123\",\"candidates\":[{\"content\":{\"parts\":[{\"text\":\"Hello \"}],\"role\":\"model\"},\"index\":0}],\"modelVersion\":\"gemini-2.5-pro-upstream\"}\n\n",
+        "data: {\"responseId\":\"resp_cli_stream_123\",\"candidates\":[{\"content\":{\"parts\":[{\"text\":\"Hello Gemini CLI\"}],\"role\":\"model\"},\"finishReason\":\"STOP\",\"index\":0}],\"modelVersion\":\"gemini-2.5-pro-upstream\",\"usageMetadata\":{\"promptTokenCount\":2,\"candidatesTokenCount\":3,\"totalTokenCount\":5}}\n\n",
+    );
+    let payload = GatewaySyncReportRequest {
+        trace_id: "trace-openai-cli-xfmt-stream-123".to_string(),
+        report_kind: "openai_cli_sync_finalize".to_string(),
+        report_context: Some(json!({
+            "client_api_format": "openai:cli",
+            "provider_api_format": "gemini:cli",
+            "model": "gpt-5",
+            "mapped_model": "gemini-2.5-pro-upstream",
+            "needs_conversion": true,
+            "has_envelope": false,
+        })),
+        status_code: 200,
+        headers: BTreeMap::from([("content-type".to_string(), "text/event-stream".to_string())]),
+        body_json: None,
+        client_body_json: None,
+        body_base64: Some(base64::engine::general_purpose::STANDARD.encode(body.as_bytes())),
+        telemetry: None,
+    };
+
+    let outcome = maybe_build_local_core_sync_finalize_response(
+        "trace-openai-cli-xfmt-stream-123",
+        &test_decision(),
+        &payload,
+    )
+    .expect("local finalize should succeed")
+    .expect("local finalize should match");
+
+    let report = outcome
+        .background_report
+        .expect("cross-format stream finalize should downgrade to success report");
+    assert_eq!(report.report_kind, "openai_cli_sync_success");
+    let client_body = report.client_body_json.expect("client body should exist");
+    assert_eq!(client_body["object"], "response");
+    assert_eq!(
+        client_body["output"][0]["content"][0]["text"],
+        "Hello Gemini CLI"
+    );
+}
+
+#[test]
+fn local_finalize_handles_openai_compact_openai_family_stream_response_even_when_conversion_flagged(
+) {
+    let body = concat!(
+        "event: response.completed\n",
+        "data: {\"type\":\"response.completed\",\"response\":{\"id\":\"resp_cli_family_stream_123\",\"object\":\"response\",\"model\":\"gpt-5\",\"status\":\"completed\",\"output\":[],\"usage\":{\"input_tokens\":1,\"output_tokens\":2,\"total_tokens\":3}}}\n\n",
+    );
+    let payload = GatewaySyncReportRequest {
+        trace_id: "trace-openai-compact-family-stream-123".to_string(),
+        report_kind: "openai_compact_sync_finalize".to_string(),
+        report_context: Some(json!({
+            "client_api_format": "openai:compact",
+            "provider_api_format": "openai:cli",
+            "model": "gpt-5",
+            "mapped_model": "gpt-5",
+            "needs_conversion": true,
+            "has_envelope": false,
+        })),
+        status_code: 200,
+        headers: BTreeMap::from([("content-type".to_string(), "text/event-stream".to_string())]),
+        body_json: None,
+        client_body_json: None,
+        body_base64: Some(base64::engine::general_purpose::STANDARD.encode(body.as_bytes())),
+        telemetry: None,
+    };
+
+    let outcome = maybe_build_local_core_sync_finalize_response(
+        "trace-openai-compact-family-stream-123",
+        &test_decision(),
+        &payload,
+    )
+    .expect("local finalize should succeed")
+    .expect("local finalize should match");
+
+    let report = outcome
+        .background_report
+        .expect("same-family stream finalize should downgrade to success report");
+    assert_eq!(report.report_kind, "openai_cli_sync_success");
+    let provider_body = report.body_json.expect("provider body should exist");
+    assert_eq!(provider_body["object"], "response");
+    assert_eq!(provider_body["status"], "completed");
+}
+
 #[test]
 fn local_finalize_handles_openai_chat_cross_format_sync_response_from_claude() {
    let payload = GatewaySyncReportRequest {
@@ -1017,6 +1149,229 @@ fn local_finalize_handles_openai_chat_cross_format_sync_response_from_gemini() {
    assert_eq!(client_body["usage"]["total_tokens"], 3);
 }

+#[test]
+fn local_finalize_handles_openai_chat_cross_format_sync_response_from_openai_cli() {
+    let payload = GatewaySyncReportRequest {
+        trace_id: "trace-openai-chat-xfmt-openai-cli-sync-123".to_string(),
+        report_kind: "openai_chat_sync_finalize".to_string(),
+        report_context: Some(json!({
+            "client_api_format": "openai:chat",
+            "provider_api_format": "openai:cli",
+            "model": "gpt-5.4",
+            "mapped_model": "gpt-5.4",
+            "needs_conversion": true,
+            "has_envelope": false,
+        })),
+        status_code: 200,
+        headers: BTreeMap::from([("content-type".to_string(), "application/json".to_string())]),
+        body_json: Some(json!({
+            "id": "resp_cli_direct_123",
+            "object": "response",
+            "status": "completed",
+            "model": "gpt-5.4",
+            "output": [
+                {
+                    "type": "message",
+                    "id": "msg_cli_direct_123",
+                    "role": "assistant",
+                    "status": "completed",
+                    "content": [{
+                        "type": "output_text",
+                        "text": "Hello CLI",
+                        "annotations": []
+                    }]
+                },
+                {
+                    "type": "function_call",
+                    "id": "fc_cli_direct_123",
+                    "call_id": "call_weather_123",
+                    "name": "get_weather",
+                    "arguments": "{\"city\":\"SF\"}"
+                }
+            ],
+            "usage": {
+                "input_tokens": 2,
+                "output_tokens": 3,
+                "total_tokens": 5
+            }
+        })),
+        client_body_json: None,
+        body_base64: None,
+        telemetry: None,
+    };
+
+    let outcome = maybe_build_local_core_sync_finalize_response(
+        "trace-openai-chat-xfmt-openai-cli-sync-123",
+        &test_decision(),
+        &payload,
+    )
+    .expect("local finalize should succeed")
+    .expect("local finalize should match");
+
+    let report = outcome
+        .background_report
+        .expect("cross-format finalize should downgrade to success report");
+    assert_eq!(report.report_kind, "openai_chat_sync_success");
+    assert_eq!(
+        report.body_json.expect("provider body should exist")["id"],
+        "resp_cli_direct_123"
+    );
+    let client_body = report.client_body_json.expect("client body should exist");
+    assert_eq!(client_body["object"], "chat.completion");
+    assert_eq!(client_body["choices"][0]["message"]["content"], "Hello CLI");
+    assert_eq!(
+        client_body["choices"][0]["message"]["tool_calls"][0]["id"],
+        "call_weather_123"
+    );
+    assert_eq!(
+        client_body["choices"][0]["message"]["tool_calls"][0]["function"]["name"],
+        "get_weather"
+    );
+    assert_eq!(
+        client_body["choices"][0]["message"]["tool_calls"][0]["function"]["arguments"],
+        "{\"city\":\"SF\"}"
+    );
+    assert_eq!(client_body["choices"][0]["finish_reason"], "tool_calls");
+    assert_eq!(client_body["usage"]["completion_tokens"], 3);
+    assert_eq!(client_body["usage"]["total_tokens"], 5);
+}
+
+#[test]
+fn local_finalize_handles_claude_chat_cross_format_sync_response_from_openai_chat() {
+    let payload = GatewaySyncReportRequest {
+        trace_id: "trace-claude-chat-xfmt-openai-sync-123".to_string(),
+        report_kind: "claude_chat_sync_finalize".to_string(),
+        report_context: Some(json!({
+            "client_api_format": "claude:chat",
+            "provider_api_format": "openai:chat",
+            "model": "claude-sonnet-4-5",
+            "mapped_model": "gpt-5",
+            "needs_conversion": true,
+            "has_envelope": false,
+        })),
+        status_code: 200,
+        headers: BTreeMap::from([("content-type".to_string(), "application/json".to_string())]),
+        body_json: Some(json!({
+            "id": "chatcmpl_openai_to_claude_123",
+            "object": "chat.completion",
+            "model": "gpt-5",
+            "choices": [{
+                "index": 0,
+                "message": {
+                    "role": "assistant",
+                    "content": "Hello OpenAI to Claude"
+                },
+                "finish_reason": "stop"
+            }],
+            "usage": {
+                "prompt_tokens": 2,
+                "completion_tokens": 3,
+                "total_tokens": 5
+            }
+        })),
+        client_body_json: None,
+        body_base64: None,
+        telemetry: None,
+    };
+
+    let outcome = maybe_build_local_core_sync_finalize_response(
+        "trace-claude-chat-xfmt-openai-sync-123",
+        &GatewayControlDecision {
+            public_path: "/v1/messages".to_string(),
+            public_query_string: None,
+            route_class: Some("ai_public".to_string()),
+            route_family: Some("claude".to_string()),
+            route_kind: Some("chat".to_string()),
+            auth_endpoint_signature: Some("claude:chat".to_string()),
+            execution_runtime_candidate: true,
+            auth_context: None,
+            admin_principal: None,
+            local_auth_rejection: None,
+        },
+        &payload,
+    )
+    .expect("local finalize should succeed")
+    .expect("local finalize should match");
+
+    let report = outcome
+        .background_report
+        .expect("cross-format finalize should downgrade to success report");
+    assert_eq!(report.report_kind, "claude_chat_sync_success");
+    let client_body = report.client_body_json.expect("client body should exist");
+    assert_eq!(client_body["type"], "message");
+    assert_eq!(client_body["role"], "assistant");
+    assert_eq!(client_body["content"][0]["text"], "Hello OpenAI to Claude");
+    assert_eq!(client_body["usage"]["input_tokens"], 2);
+    assert_eq!(client_body["usage"]["output_tokens"], 3);
+}
+
+#[test]
+fn local_finalize_handles_gemini_cli_cross_format_sync_response_from_claude_cli() {
+    let payload = GatewaySyncReportRequest {
+        trace_id: "trace-gemini-cli-xfmt-claude-sync-123".to_string(),
+        report_kind: "gemini_cli_sync_finalize".to_string(),
+        report_context: Some(json!({
+            "client_api_format": "gemini:cli",
+            "provider_api_format": "claude:cli",
+            "model": "gemini-cli",
+            "mapped_model": "claude-code",
+            "needs_conversion": true,
+            "has_envelope": false,
+        })),
+        status_code: 200,
+        headers: BTreeMap::from([("content-type".to_string(), "application/json".to_string())]),
+        body_json: Some(json!({
+            "id": "msg_claude_cli_to_gemini_123",
+            "type": "message",
+            "model": "claude-code",
+            "role": "assistant",
+            "content": [{
+                "type": "text",
+                "text": "Hello Claude CLI to Gemini"
+            }],
+            "stop_reason": "end_turn",
+            "usage": {
+                "input_tokens": 4,
+                "output_tokens": 6
+            }
+        })),
+        client_body_json: None,
+        body_base64: None,
+        telemetry: None,
+    };
+
+    let outcome = maybe_build_local_core_sync_finalize_response(
+        "trace-gemini-cli-xfmt-claude-sync-123",
+        &GatewayControlDecision {
+            public_path: "/v1beta/models/gemini-cli:generateContent".to_string(),
+            public_query_string: None,
+            route_class: Some("ai_public".to_string()),
+            route_family: Some("gemini".to_string()),
+            route_kind: Some("cli".to_string()),
+            auth_endpoint_signature: Some("gemini:cli".to_string()),
+            execution_runtime_candidate: true,
+            auth_context: None,
+            admin_principal: None,
+            local_auth_rejection: None,
+        },
+        &payload,
+    )
+    .expect("local finalize should succeed")
+    .expect("local finalize should match");
+
+    let report = outcome
+        .background_report
+        .expect("cross-format finalize should downgrade to success report");
+    assert_eq!(report.report_kind, "gemini_cli_sync_success");
+    let client_body = report.client_body_json.expect("client body should exist");
+    assert_eq!(
+        client_body["candidates"][0]["content"]["parts"][0]["text"],
+        "Hello Claude CLI to Gemini"
+    );
+    assert_eq!(client_body["usageMetadata"]["promptTokenCount"], 4);
+    assert_eq!(client_body["usageMetadata"]["candidatesTokenCount"], 6);
+}
+
 #[test]
 fn local_finalize_handles_kiro_claude_cli_stream_tool_use_response() {
    let payload = build_kiro_claude_cli_sync_finalize_payload(
--- a/apps/aether-gateway/src/ai_pipeline/mod.rs
+++ b/apps/aether-gateway/src/ai_pipeline/mod.rs
@@ -0,0 +1,6 @@
+pub(crate) mod conversion;
+pub(crate) mod finalize;
+pub(crate) mod planner;
+pub(crate) mod private_response;
+pub(crate) mod private_surfaces;
+pub(crate) mod runtime;
--- a/apps/aether-gateway/src/ai_pipeline/planner/candidate_affinity.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/candidate_affinity.rs
@@ -0,0 +1,294 @@
+use tracing::warn;
+
+use crate::gateway::provider_transport::resolve_transport_proxy_snapshot;
+use crate::gateway::scheduler::GatewayMinimalCandidateSelectionCandidate;
+use crate::gateway::AppState;
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq, PartialOrd, Ord)]
+enum TunnelOwnerAffinityBucket {
+    LocalTunnel = 0,
+    Neutral = 1,
+    RemoteTunnel = 2,
+}
+
+pub(crate) async fn prefer_local_tunnel_owner_candidates(
+    state: &AppState,
+    candidates: Vec<GatewayMinimalCandidateSelectionCandidate>,
+) -> Vec<GatewayMinimalCandidateSelectionCandidate> {
+    let mut ranked = Vec::with_capacity(candidates.len());
+    for (original_index, candidate) in candidates.into_iter().enumerate() {
+        let bucket = resolve_candidate_tunnel_owner_affinity(state, &candidate).await;
+        ranked.push((bucket, original_index, candidate));
+    }
+    ranked.sort_by(|left, right| left.0.cmp(&right.0).then(left.1.cmp(&right.1)));
+    ranked
+        .into_iter()
+        .map(|(_, _, candidate)| candidate)
+        .collect()
+}
+
+async fn resolve_candidate_tunnel_owner_affinity(
+    state: &AppState,
+    candidate: &GatewayMinimalCandidateSelectionCandidate,
+) -> TunnelOwnerAffinityBucket {
+    let transport = match state
+        .read_provider_transport_snapshot(
+            &candidate.provider_id,
+            &candidate.endpoint_id,
+            &candidate.key_id,
+        )
+        .await
+    {
+        Ok(Some(transport)) => transport,
+        Ok(None) => return TunnelOwnerAffinityBucket::Neutral,
+        Err(error) => {
+            warn!(
+                provider_id = %candidate.provider_id,
+                endpoint_id = %candidate.endpoint_id,
+                key_id = %candidate.key_id,
+                error = ?error,
+                "failed to load provider transport while evaluating tunnel owner affinity"
+            );
+            return TunnelOwnerAffinityBucket::Neutral;
+        }
+    };
+
+    let Some(proxy) = resolve_transport_proxy_snapshot(&transport) else {
+        return TunnelOwnerAffinityBucket::Neutral;
+    };
+    if proxy.enabled == Some(false) {
+        return TunnelOwnerAffinityBucket::Neutral;
+    }
+    let Some(node_id) = proxy
+        .node_id
+        .as_deref()
+        .map(str::trim)
+        .filter(|value| !value.is_empty())
+    else {
+        return TunnelOwnerAffinityBucket::Neutral;
+    };
+
+    if state.tunnel.has_local_proxy(node_id) {
+        return TunnelOwnerAffinityBucket::LocalTunnel;
+    }
+
+    match state
+        .tunnel
+        .lookup_attachment_owner(state.data.as_ref(), node_id)
+        .await
+    {
+        Ok(Some(owner)) if owner.gateway_instance_id == state.tunnel.local_instance_id() => {
+            TunnelOwnerAffinityBucket::LocalTunnel
+        }
+        Ok(Some(_)) => TunnelOwnerAffinityBucket::RemoteTunnel,
+        Ok(None) => TunnelOwnerAffinityBucket::Neutral,
+        Err(error) => {
+            warn!(
+                node_id = node_id,
+                error = %error,
+                "failed to load tunnel attachment owner while evaluating scheduler affinity"
+            );
+            TunnelOwnerAffinityBucket::Neutral
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use std::time::{SystemTime, UNIX_EPOCH};
+
+    use aether_data::repository::provider_catalog::{
+        InMemoryProviderCatalogReadRepository, StoredProviderCatalogEndpoint,
+        StoredProviderCatalogKey, StoredProviderCatalogProvider,
+    };
+    use serde_json::json;
+
+    use super::*;
+    use crate::gateway::tunnel::TunnelAttachmentRecord;
+    use crate::gateway::GatewayDataState;
+
+    fn sample_candidate(
+        endpoint_id: &str,
+        key_id: &str,
+    ) -> GatewayMinimalCandidateSelectionCandidate {
+        GatewayMinimalCandidateSelectionCandidate {
+            provider_id: "provider-1".to_string(),
+            provider_name: "provider-1".to_string(),
+            provider_type: "custom".to_string(),
+            provider_priority: 0,
+            endpoint_id: endpoint_id.to_string(),
+            endpoint_api_format: "openai:chat".to_string(),
+            key_id: key_id.to_string(),
+            key_name: key_id.to_string(),
+            key_auth_type: "api_key".to_string(),
+            key_internal_priority: 0,
+            key_global_priority_for_format: Some(0),
+            key_capabilities: None,
+            model_id: "model-1".to_string(),
+            global_model_id: "global-model-1".to_string(),
+            global_model_name: "gpt-4.1".to_string(),
+            selected_provider_model_name: "gpt-4.1".to_string(),
+            mapping_matched_model: None,
+        }
+    }
+
+    fn sample_provider() -> StoredProviderCatalogProvider {
+        StoredProviderCatalogProvider::new(
+            "provider-1".to_string(),
+            "provider-1".to_string(),
+            Some("https://provider.example".to_string()),
+            "custom".to_string(),
+        )
+        .expect("provider should build")
+        .with_transport_fields(true, false, false, None, None, None, None, None, None)
+    }
+
+    fn sample_endpoint(id: &str) -> StoredProviderCatalogEndpoint {
+        StoredProviderCatalogEndpoint::new(
+            id.to_string(),
+            "provider-1".to_string(),
+            "openai:chat".to_string(),
+            Some("openai".to_string()),
+            Some("chat".to_string()),
+            true,
+        )
+        .expect("endpoint should build")
+        .with_transport_fields(
+            "https://api.provider.example".to_string(),
+            None,
+            None,
+            None,
+            None,
+            None,
+            None,
+            None,
+        )
+        .expect("endpoint transport should build")
+    }
+
+    fn sample_key(id: &str, node_id: &str) -> StoredProviderCatalogKey {
+        StoredProviderCatalogKey::new(
+            id.to_string(),
+            "provider-1".to_string(),
+            id.to_string(),
+            "api_key".to_string(),
+            None,
+            true,
+        )
+        .expect("key should build")
+        .with_transport_fields(
+            Some(json!(["openai:chat"])),
+            "plain-upstream-key".to_string(),
+            None,
+            None,
+            Some(json!({"openai:chat": 1})),
+            None,
+            None,
+            Some(json!({
+                "enabled": true,
+                "mode": "tunnel",
+                "node_id": node_id,
+            })),
+            None,
+        )
+        .expect("key transport should build")
+    }
+
+    fn tunnel_attachment_key(node_id: &str) -> String {
+        format!("tunnel.attachments.{node_id}")
+    }
+
+    fn current_unix_secs() -> u64 {
+        SystemTime::now()
+            .duration_since(UNIX_EPOCH)
+            .unwrap_or_default()
+            .as_secs()
+    }
+
+    #[tokio::test]
+    async fn prefers_local_tunnel_owner_candidates_before_remote_tunnel_candidates() {
+        let provider_catalog = InMemoryProviderCatalogReadRepository::seed(
+            vec![sample_provider()],
+            vec![
+                sample_endpoint("endpoint-remote"),
+                sample_endpoint("endpoint-local"),
+            ],
+            vec![
+                sample_key("key-remote", "node-remote"),
+                sample_key("key-local", "node-local"),
+            ],
+        );
+        let observed_at_unix_secs = current_unix_secs();
+        let data_state = GatewayDataState::with_provider_transport_reader_for_tests(
+            std::sync::Arc::new(provider_catalog),
+            "development-key",
+        )
+        .with_system_config_values_for_tests(vec![
+            (
+                tunnel_attachment_key("node-remote"),
+                serde_json::to_value(TunnelAttachmentRecord {
+                    gateway_instance_id: "gateway-b".to_string(),
+                    relay_base_url: "http://gateway-b:8080".to_string(),
+                    conn_count: 1,
+                    observed_at_unix_secs,
+                })
+                .expect("remote attachment should serialize"),
+            ),
+            (
+                tunnel_attachment_key("node-local"),
+                serde_json::to_value(TunnelAttachmentRecord {
+                    gateway_instance_id: "gateway-a".to_string(),
+                    relay_base_url: "http://gateway-a:8080".to_string(),
+                    conn_count: 1,
+                    observed_at_unix_secs,
+                })
+                .expect("local attachment should serialize"),
+            ),
+        ]);
+        let state = AppState::new("http://127.0.0.1:1")
+            .expect("state should build")
+            .with_data_state_for_tests(data_state)
+            .with_tunnel_identity_for_tests("gateway-a", Some("http://gateway-a:8080"));
+
+        let reordered = prefer_local_tunnel_owner_candidates(
+            &state,
+            vec![
+                sample_candidate("endpoint-remote", "key-remote"),
+                sample_candidate("endpoint-local", "key-local"),
+            ],
+        )
+        .await;
+
+        assert_eq!(reordered[0].endpoint_id, "endpoint-local");
+        assert_eq!(reordered[1].endpoint_id, "endpoint-remote");
+    }
+
+    #[tokio::test]
+    async fn leaves_candidate_order_unchanged_when_transport_has_no_tunnel_proxy() {
+        let provider_catalog = InMemoryProviderCatalogReadRepository::seed(
+            vec![sample_provider()],
+            vec![sample_endpoint("endpoint-a"), sample_endpoint("endpoint-b")],
+            vec![sample_key("key-a", ""), sample_key("key-b", "")],
+        );
+        let data_state = GatewayDataState::with_provider_transport_reader_for_tests(
+            std::sync::Arc::new(provider_catalog),
+            "development-key",
+        );
+        let state = AppState::new("http://127.0.0.1:1")
+            .expect("state should build")
+            .with_data_state_for_tests(data_state)
+            .with_tunnel_identity_for_tests("gateway-a", Some("http://gateway-a:8080"));
+
+        let reordered = prefer_local_tunnel_owner_candidates(
+            &state,
+            vec![
+                sample_candidate("endpoint-a", "key-a"),
+                sample_candidate("endpoint-b", "key-b"),
+            ],
+        )
+        .await;
+
+        assert_eq!(reordered[0].endpoint_id, "endpoint-a");
+        assert_eq!(reordered[1].endpoint_id, "endpoint-b");
+    }
+}
--- a/apps/aether-gateway/src/ai_pipeline/planner/common.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/common.rs
@@ -0,0 +1,57 @@
+use axum::body::Bytes;
+use base64::Engine as _;
+
+use crate::gateway::headers::is_json_request;
+
+pub(crate) const GEMINI_FILES_GET_PLAN_KIND: &str = "gemini_files_get";
+pub(crate) const GEMINI_FILES_UPLOAD_PLAN_KIND: &str = "gemini_files_upload";
+pub(crate) const GEMINI_FILES_LIST_PLAN_KIND: &str = "gemini_files_list";
+pub(crate) const GEMINI_FILES_DELETE_PLAN_KIND: &str = "gemini_files_delete";
+pub(crate) const GEMINI_FILES_DOWNLOAD_PLAN_KIND: &str = "gemini_files_download";
+pub(crate) const OPENAI_VIDEO_CONTENT_PLAN_KIND: &str = "openai_video_content";
+pub(crate) const OPENAI_VIDEO_CANCEL_SYNC_PLAN_KIND: &str = "openai_video_cancel_sync";
+pub(crate) const OPENAI_VIDEO_REMIX_SYNC_PLAN_KIND: &str = "openai_video_remix_sync";
+pub(crate) const OPENAI_VIDEO_DELETE_SYNC_PLAN_KIND: &str = "openai_video_delete_sync";
+pub(crate) const GEMINI_VIDEO_CREATE_SYNC_PLAN_KIND: &str = "gemini_video_create_sync";
+pub(crate) const GEMINI_VIDEO_CANCEL_SYNC_PLAN_KIND: &str = "gemini_video_cancel_sync";
+pub(crate) const OPENAI_CHAT_STREAM_PLAN_KIND: &str = "openai_chat_stream";
+pub(crate) const CLAUDE_CHAT_STREAM_PLAN_KIND: &str = "claude_chat_stream";
+pub(crate) const GEMINI_CHAT_STREAM_PLAN_KIND: &str = "gemini_chat_stream";
+pub(crate) const OPENAI_CLI_STREAM_PLAN_KIND: &str = "openai_cli_stream";
+pub(crate) const OPENAI_COMPACT_STREAM_PLAN_KIND: &str = "openai_compact_stream";
+pub(crate) const CLAUDE_CLI_STREAM_PLAN_KIND: &str = "claude_cli_stream";
+pub(crate) const GEMINI_CLI_STREAM_PLAN_KIND: &str = "gemini_cli_stream";
+pub(crate) const OPENAI_VIDEO_CREATE_SYNC_PLAN_KIND: &str = "openai_video_create_sync";
+pub(crate) const OPENAI_CHAT_SYNC_PLAN_KIND: &str = "openai_chat_sync";
+pub(crate) const OPENAI_CLI_SYNC_PLAN_KIND: &str = "openai_cli_sync";
+pub(crate) const OPENAI_COMPACT_SYNC_PLAN_KIND: &str = "openai_compact_sync";
+pub(crate) const CLAUDE_CHAT_SYNC_PLAN_KIND: &str = "claude_chat_sync";
+pub(crate) const GEMINI_CHAT_SYNC_PLAN_KIND: &str = "gemini_chat_sync";
+pub(crate) const CLAUDE_CLI_SYNC_PLAN_KIND: &str = "claude_cli_sync";
+pub(crate) const GEMINI_CLI_SYNC_PLAN_KIND: &str = "gemini_cli_sync";
+pub(crate) const EXECUTION_RUNTIME_SYNC_ACTION: &str = "execution_runtime_sync";
+pub(crate) const EXECUTION_RUNTIME_SYNC_DECISION_ACTION: &str = "execution_runtime_sync_decision";
+pub(crate) const EXECUTION_RUNTIME_STREAM_ACTION: &str = "execution_runtime_stream";
+pub(crate) const EXECUTION_RUNTIME_STREAM_DECISION_ACTION: &str =
+    "execution_runtime_stream_decision";
+
+pub(crate) fn parse_direct_request_body(
+    parts: &http::request::Parts,
+    body_bytes: &Bytes,
+) -> Option<(serde_json::Value, Option<String>)> {
+    if is_json_request(&parts.headers) {
+        if body_bytes.is_empty() {
+            Some((serde_json::json!({}), None))
+        } else {
+            serde_json::from_slice::<serde_json::Value>(body_bytes)
+                .ok()
+                .map(|value| (value, None))
+        }
+    } else {
+        Some((
+            serde_json::json!({}),
+            (!body_bytes.is_empty())
+                .then(|| base64::engine::general_purpose::STANDARD.encode(body_bytes)),
+        ))
+    }
+}
--- a/apps/aether-gateway/src/ai_pipeline/planner/contracts.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/contracts.rs
@@ -0,0 +1,162 @@
+use std::collections::BTreeMap;
+
+use aether_contracts::{ExecutionPlan, ExecutionTimeouts, ProxySnapshot};
+use serde::{Deserialize, Serialize};
+use tracing::warn;
+
+use crate::gateway::headers::collect_control_headers;
+use crate::gateway::{AppState, GatewayControlAuthContext, GatewayControlDecision, GatewayError};
+
+#[derive(Debug, Serialize)]
+pub(crate) struct GatewayControlPlanRequest {
+    pub(crate) trace_id: String,
+    pub(crate) method: String,
+    pub(crate) path: String,
+    pub(crate) query_string: Option<String>,
+    pub(crate) headers: BTreeMap<String, String>,
+    pub(crate) body_json: serde_json::Value,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub(crate) body_base64: Option<String>,
+    pub(crate) auth_context: Option<GatewayControlAuthContext>,
+}
+
+#[derive(Debug, Deserialize, Serialize)]
+pub(crate) struct GatewayControlPlanResponse {
+    pub(crate) action: String,
+    #[serde(default)]
+    pub(crate) plan_kind: Option<String>,
+    #[serde(default)]
+    pub(crate) plan: Option<ExecutionPlan>,
+    #[serde(default)]
+    pub(crate) report_kind: Option<String>,
+    #[serde(default)]
+    pub(crate) report_context: Option<serde_json::Value>,
+    #[serde(default)]
+    pub(crate) auth_context: Option<GatewayControlAuthContext>,
+}
+
+#[derive(Debug, Deserialize, Serialize)]
+pub(crate) struct GatewayControlSyncDecisionResponse {
+    pub(crate) action: String,
+    #[serde(default)]
+    pub(crate) decision_kind: Option<String>,
+    #[serde(default)]
+    pub(crate) execution_strategy: Option<String>,
+    #[serde(default)]
+    pub(crate) conversion_mode: Option<String>,
+    #[serde(default)]
+    pub(crate) request_id: Option<String>,
+    #[serde(default)]
+    pub(crate) candidate_id: Option<String>,
+    #[serde(default)]
+    pub(crate) provider_name: Option<String>,
+    #[serde(default)]
+    pub(crate) provider_id: Option<String>,
+    #[serde(default)]
+    pub(crate) endpoint_id: Option<String>,
+    #[serde(default)]
+    pub(crate) key_id: Option<String>,
+    #[serde(default)]
+    pub(crate) upstream_base_url: Option<String>,
+    #[serde(default)]
+    pub(crate) upstream_url: Option<String>,
+    #[serde(default)]
+    pub(crate) provider_request_method: Option<String>,
+    #[serde(default)]
+    pub(crate) auth_header: Option<String>,
+    #[serde(default)]
+    pub(crate) auth_value: Option<String>,
+    #[serde(default)]
+    pub(crate) provider_api_format: Option<String>,
+    #[serde(default)]
+    pub(crate) client_api_format: Option<String>,
+    #[serde(default)]
+    pub(crate) provider_contract: Option<String>,
+    #[serde(default)]
+    pub(crate) client_contract: Option<String>,
+    #[serde(default)]
+    pub(crate) model_name: Option<String>,
+    #[serde(default)]
+    pub(crate) mapped_model: Option<String>,
+    #[serde(default)]
+    pub(crate) prompt_cache_key: Option<String>,
+    #[serde(default)]
+    pub(crate) extra_headers: BTreeMap<String, String>,
+    #[serde(default)]
+    pub(crate) provider_request_headers: BTreeMap<String, String>,
+    #[serde(default)]
+    pub(crate) provider_request_body: Option<serde_json::Value>,
+    #[serde(default)]
+    pub(crate) provider_request_body_base64: Option<String>,
+    #[serde(default)]
+    pub(crate) content_type: Option<String>,
+    #[serde(default)]
+    pub(crate) proxy: Option<ProxySnapshot>,
+    #[serde(default)]
+    pub(crate) tls_profile: Option<String>,
+    #[serde(default)]
+    pub(crate) timeouts: Option<ExecutionTimeouts>,
+    #[serde(default)]
+    pub(crate) upstream_is_stream: bool,
+    #[serde(default)]
+    pub(crate) report_kind: Option<String>,
+    #[serde(default)]
+    pub(crate) report_context: Option<serde_json::Value>,
+    #[serde(default)]
+    pub(crate) auth_context: Option<GatewayControlAuthContext>,
+}
+
+fn decision_has_exact_provider_request(payload: &GatewayControlSyncDecisionResponse) -> bool {
+    !payload.provider_request_headers.is_empty()
+        && (payload.provider_request_body.is_some()
+            || payload
+                .provider_request_body_base64
+                .as_ref()
+                .map(|value| !value.trim().is_empty())
+                .unwrap_or(false))
+}
+
+pub(crate) fn generic_decision_missing_exact_provider_request(
+    payload: &GatewayControlSyncDecisionResponse,
+) -> bool {
+    if decision_has_exact_provider_request(payload) {
+        return false;
+    }
+
+    warn!(
+        decision_kind = payload.decision_kind.as_deref().unwrap_or_default(),
+        provider_api_format = payload.provider_api_format.as_deref().unwrap_or_default(),
+        client_api_format = payload.client_api_format.as_deref().unwrap_or_default(),
+        "gateway generic decision missing exact provider request; falling back to plan"
+    );
+    true
+}
+
+pub(crate) async fn build_gateway_plan_request(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    body_json: serde_json::Value,
+    body_base64: Option<String>,
+) -> Result<GatewayControlPlanRequest, GatewayError> {
+    let auth_context = crate::gateway::resolve_execution_runtime_auth_context(
+        state,
+        decision,
+        &parts.headers,
+        &parts.uri,
+        trace_id,
+    )
+    .await?;
+
+    Ok(GatewayControlPlanRequest {
+        trace_id: trace_id.to_string(),
+        method: parts.method.to_string(),
+        path: parts.uri.path().to_string(),
+        query_string: parts.uri.query().map(ToOwned::to_owned),
+        headers: collect_control_headers(&parts.headers),
+        body_json,
+        body_base64,
+        auth_context,
+    })
+}
--- a/apps/aether-gateway/src/ai_pipeline/planner/decision/control_plan.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/decision/control_plan.rs
@@ -0,0 +1,175 @@
+use crate::gateway::ai_pipeline::planner::common::{
+    CLAUDE_CHAT_STREAM_PLAN_KIND, CLAUDE_CHAT_SYNC_PLAN_KIND, CLAUDE_CLI_STREAM_PLAN_KIND,
+    CLAUDE_CLI_SYNC_PLAN_KIND, EXECUTION_RUNTIME_STREAM_ACTION, EXECUTION_RUNTIME_SYNC_ACTION,
+    GEMINI_CHAT_STREAM_PLAN_KIND, GEMINI_CHAT_SYNC_PLAN_KIND, GEMINI_CLI_STREAM_PLAN_KIND,
+    GEMINI_CLI_SYNC_PLAN_KIND, GEMINI_FILES_DELETE_PLAN_KIND, GEMINI_FILES_DOWNLOAD_PLAN_KIND,
+    GEMINI_FILES_GET_PLAN_KIND, GEMINI_FILES_LIST_PLAN_KIND, GEMINI_VIDEO_CANCEL_SYNC_PLAN_KIND,
+    GEMINI_VIDEO_CREATE_SYNC_PLAN_KIND, OPENAI_CHAT_STREAM_PLAN_KIND, OPENAI_CHAT_SYNC_PLAN_KIND,
+    OPENAI_CLI_STREAM_PLAN_KIND, OPENAI_CLI_SYNC_PLAN_KIND, OPENAI_COMPACT_STREAM_PLAN_KIND,
+    OPENAI_COMPACT_SYNC_PLAN_KIND, OPENAI_VIDEO_CANCEL_SYNC_PLAN_KIND,
+    OPENAI_VIDEO_CONTENT_PLAN_KIND, OPENAI_VIDEO_CREATE_SYNC_PLAN_KIND,
+    OPENAI_VIDEO_DELETE_SYNC_PLAN_KIND, OPENAI_VIDEO_REMIX_SYNC_PLAN_KIND,
+};
+use crate::gateway::ai_pipeline::planner::plan_builders::{
+    build_gemini_stream_plan_from_decision, build_gemini_sync_plan_from_decision,
+    build_openai_chat_stream_plan_from_decision, build_openai_chat_sync_plan_from_decision,
+    build_openai_cli_stream_plan_from_decision, build_openai_cli_sync_plan_from_decision,
+    build_passthrough_stream_plan_from_decision, build_passthrough_sync_plan_from_decision,
+    build_standard_stream_plan_from_decision, build_standard_sync_plan_from_decision,
+    LocalStreamPlanAndReport, LocalSyncPlanAndReport,
+};
+use crate::gateway::{
+    AppState, GatewayControlAuthContext, GatewayControlDecision, GatewayControlPlanResponse,
+    GatewayControlSyncDecisionResponse, GatewayError,
+};
+
+pub(crate) async fn maybe_build_sync_plan_payload_impl(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    body_json: &serde_json::Value,
+    body_base64: Option<&str>,
+    body_is_empty: bool,
+) -> Result<Option<GatewayControlPlanResponse>, GatewayError> {
+    let Some(plan_kind) = super::resolve_sync_plan_kind(parts, decision) else {
+        return Ok(None);
+    };
+    let Some(payload) = super::maybe_build_sync_decision_payload(
+        state,
+        parts,
+        trace_id,
+        decision,
+        body_json,
+        body_base64,
+        body_is_empty,
+    )
+    .await?
+    else {
+        return Ok(None);
+    };
+
+    build_sync_plan_payload_from_decision(parts, body_json, plan_kind, payload)
+}
+
+pub(crate) async fn maybe_build_stream_plan_payload_impl(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    body_json: &serde_json::Value,
+) -> Result<Option<GatewayControlPlanResponse>, GatewayError> {
+    let Some(plan_kind) = super::resolve_stream_plan_kind(parts, decision) else {
+        return Ok(None);
+    };
+    let Some(payload) =
+        super::maybe_build_stream_decision_payload(state, parts, trace_id, decision, body_json)
+            .await?
+    else {
+        return Ok(None);
+    };
+
+    build_stream_plan_payload_from_decision(parts, body_json, plan_kind, payload)
+}
+
+fn build_sync_plan_payload_from_decision(
+    parts: &http::request::Parts,
+    body_json: &serde_json::Value,
+    plan_kind: &str,
+    payload: GatewayControlSyncDecisionResponse,
+) -> Result<Option<GatewayControlPlanResponse>, GatewayError> {
+    let auth_context = payload.auth_context.clone();
+    let plan_and_report = match plan_kind {
+        OPENAI_CHAT_SYNC_PLAN_KIND => {
+            build_openai_chat_sync_plan_from_decision(parts, body_json, payload)?
+        }
+        OPENAI_CLI_SYNC_PLAN_KIND => {
+            build_openai_cli_sync_plan_from_decision(parts, body_json, payload, false)?
+        }
+        OPENAI_COMPACT_SYNC_PLAN_KIND => {
+            build_openai_cli_sync_plan_from_decision(parts, body_json, payload, true)?
+        }
+        CLAUDE_CHAT_SYNC_PLAN_KIND | CLAUDE_CLI_SYNC_PLAN_KIND => {
+            build_standard_sync_plan_from_decision(parts, body_json, payload)?
+        }
+        GEMINI_CHAT_SYNC_PLAN_KIND | GEMINI_CLI_SYNC_PLAN_KIND => {
+            build_gemini_sync_plan_from_decision(parts, body_json, payload)?
+        }
+        OPENAI_VIDEO_CREATE_SYNC_PLAN_KIND
+        | OPENAI_VIDEO_REMIX_SYNC_PLAN_KIND
+        | OPENAI_VIDEO_CANCEL_SYNC_PLAN_KIND
+        | OPENAI_VIDEO_DELETE_SYNC_PLAN_KIND
+        | GEMINI_VIDEO_CREATE_SYNC_PLAN_KIND
+        | GEMINI_VIDEO_CANCEL_SYNC_PLAN_KIND
+        | GEMINI_FILES_LIST_PLAN_KIND
+        | GEMINI_FILES_GET_PLAN_KIND
+        | GEMINI_FILES_DELETE_PLAN_KIND => {
+            build_passthrough_sync_plan_from_decision(parts, payload)?
+        }
+        _ => None,
+    };
+
+    Ok(plan_and_report.map(|value| build_sync_plan_response(plan_kind, value, auth_context)))
+}
+
+fn build_stream_plan_payload_from_decision(
+    parts: &http::request::Parts,
+    body_json: &serde_json::Value,
+    plan_kind: &str,
+    payload: GatewayControlSyncDecisionResponse,
+) -> Result<Option<GatewayControlPlanResponse>, GatewayError> {
+    let auth_context = payload.auth_context.clone();
+    let plan_and_report = match plan_kind {
+        OPENAI_CHAT_STREAM_PLAN_KIND => {
+            build_openai_chat_stream_plan_from_decision(parts, body_json, payload)?
+        }
+        OPENAI_CLI_STREAM_PLAN_KIND => {
+            build_openai_cli_stream_plan_from_decision(parts, body_json, payload, false)?
+        }
+        OPENAI_COMPACT_STREAM_PLAN_KIND => {
+            build_openai_cli_stream_plan_from_decision(parts, body_json, payload, true)?
+        }
+        CLAUDE_CHAT_STREAM_PLAN_KIND | CLAUDE_CLI_STREAM_PLAN_KIND => {
+            build_standard_stream_plan_from_decision(parts, body_json, payload, true)?
+        }
+        GEMINI_CHAT_STREAM_PLAN_KIND | GEMINI_CLI_STREAM_PLAN_KIND => {
+            build_gemini_stream_plan_from_decision(parts, body_json, payload)?
+        }
+        OPENAI_VIDEO_CONTENT_PLAN_KIND | GEMINI_FILES_DOWNLOAD_PLAN_KIND => {
+            build_passthrough_stream_plan_from_decision(parts, payload)?
+        }
+        _ => None,
+    };
+
+    Ok(plan_and_report.map(|value| build_stream_plan_response(plan_kind, value, auth_context)))
+}
+
+fn build_sync_plan_response(
+    plan_kind: &str,
+    value: LocalSyncPlanAndReport,
+    auth_context: Option<GatewayControlAuthContext>,
+) -> GatewayControlPlanResponse {
+    GatewayControlPlanResponse {
+        action: EXECUTION_RUNTIME_SYNC_ACTION.to_string(),
+        plan_kind: Some(plan_kind.to_string()),
+        plan: Some(value.plan),
+        report_kind: value.report_kind,
+        report_context: value.report_context,
+        auth_context,
+    }
+}
+
+fn build_stream_plan_response(
+    plan_kind: &str,
+    value: LocalStreamPlanAndReport,
+    auth_context: Option<GatewayControlAuthContext>,
+) -> GatewayControlPlanResponse {
+    GatewayControlPlanResponse {
+        action: EXECUTION_RUNTIME_STREAM_ACTION.to_string(),
+        plan_kind: Some(plan_kind.to_string()),
+        plan: Some(value.plan),
+        report_kind: value.report_kind,
+        report_context: value.report_context,
+        auth_context,
+    }
+}
--- a/apps/aether-gateway/src/ai_pipeline/planner/decision/mod.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/decision/mod.rs
@@ -0,0 +1,22 @@
+mod control_plan;
+mod stream;
+mod sync;
+
+pub(crate) use self::control_plan::{
+    maybe_build_stream_plan_payload_impl, maybe_build_sync_plan_payload_impl,
+};
+pub(crate) use self::stream::maybe_build_stream_decision_payload_impl as maybe_build_stream_decision_payload;
+pub(crate) use self::sync::maybe_build_sync_decision_payload_impl as maybe_build_sync_decision_payload;
+pub(crate) use super::{
+    maybe_build_stream_local_decision_payload,
+    maybe_build_stream_local_gemini_files_decision_payload,
+    maybe_build_stream_local_openai_cli_decision_payload,
+    maybe_build_stream_local_same_format_provider_decision_payload,
+    maybe_build_stream_local_standard_decision_payload, maybe_build_sync_local_decision_payload,
+    maybe_build_sync_local_gemini_files_decision_payload,
+    maybe_build_sync_local_openai_cli_decision_payload,
+    maybe_build_sync_local_same_format_provider_decision_payload,
+    maybe_build_sync_local_standard_decision_payload,
+    maybe_build_sync_local_video_decision_payload, resolve_stream_plan_kind,
+    resolve_sync_plan_kind,
+};
--- a/apps/aether-gateway/src/ai_pipeline/planner/decision/stream.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/decision/stream.rs
@@ -0,0 +1,156 @@
+use std::collections::BTreeMap;
+
+use crate::gateway::ai_pipeline::planner::common::{
+    EXECUTION_RUNTIME_STREAM_DECISION_ACTION, OPENAI_VIDEO_CONTENT_PLAN_KIND,
+};
+use crate::gateway::scheduler::{
+    is_matching_stream_request, resolve_execution_runtime_stream_plan_kind,
+};
+use crate::gateway::{
+    AppState, GatewayControlDecision, GatewayControlSyncDecisionResponse, GatewayError,
+};
+
+pub(crate) async fn maybe_build_stream_decision_payload_impl(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    body_json: &serde_json::Value,
+) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
+    let Some(plan_kind) = resolve_execution_runtime_stream_plan_kind(parts, decision) else {
+        return Ok(None);
+    };
+
+    if !is_matching_stream_request(plan_kind, parts, body_json) {
+        return Ok(None);
+    }
+
+    if let Some(payload) = maybe_build_local_video_task_content_stream_decision_payload(
+        state, parts, trace_id, decision, plan_kind,
+    )
+    .await?
+    {
+        return Ok(Some(payload));
+    }
+
+    if let Some(payload) = super::maybe_build_stream_local_decision_payload(
+        state, parts, trace_id, decision, body_json, plan_kind,
+    )
+    .await?
+    {
+        return Ok(Some(payload));
+    }
+
+    if let Some(payload) = super::maybe_build_stream_local_openai_cli_decision_payload(
+        state, parts, trace_id, decision, body_json, plan_kind,
+    )
+    .await?
+    {
+        return Ok(Some(payload));
+    }
+
+    if let Some(payload) = super::maybe_build_stream_local_standard_decision_payload(
+        state, parts, trace_id, decision, body_json, plan_kind,
+    )
+    .await?
+    {
+        return Ok(Some(payload));
+    }
+
+    if let Some(payload) = super::maybe_build_stream_local_same_format_provider_decision_payload(
+        state, parts, trace_id, decision, body_json, plan_kind,
+    )
+    .await?
+    {
+        return Ok(Some(payload));
+    }
+
+    if let Some(payload) = super::maybe_build_stream_local_gemini_files_decision_payload(
+        state, parts, trace_id, decision, plan_kind,
+    )
+    .await?
+    {
+        return Ok(Some(payload));
+    }
+
+    Ok(None)
+}
+
+async fn maybe_build_local_video_task_content_stream_decision_payload(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    plan_kind: &str,
+) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
+    if plan_kind != OPENAI_VIDEO_CONTENT_PLAN_KIND
+        || decision.route_family.as_deref() != Some("openai")
+    {
+        return Ok(None);
+    }
+
+    let _ = state
+        .hydrate_video_task_for_route(decision.route_family.as_deref(), parts.uri.path())
+        .await?;
+
+    let Some(action) = state.video_tasks.prepare_openai_content_stream_action(
+        parts.uri.path(),
+        parts.uri.query(),
+        trace_id,
+    ) else {
+        return Ok(None);
+    };
+
+    let crate::gateway::video_tasks::LocalVideoTaskContentAction::StreamPlan(plan) = action else {
+        return Ok(None);
+    };
+    let provider_contract = plan.provider_api_format.clone();
+    let client_contract = plan.client_api_format.clone();
+    let execution_strategy = if plan.provider_api_format == plan.client_api_format {
+        crate::gateway::ExecutionStrategy::LocalSameFormat
+    } else {
+        crate::gateway::ExecutionStrategy::LocalCrossFormat
+    };
+    let conversion_mode = if plan.provider_api_format == plan.client_api_format {
+        crate::gateway::ConversionMode::None
+    } else {
+        crate::gateway::ConversionMode::Bidirectional
+    };
+
+    Ok(Some(GatewayControlSyncDecisionResponse {
+        action: EXECUTION_RUNTIME_STREAM_DECISION_ACTION.to_string(),
+        decision_kind: Some(plan_kind.to_string()),
+        execution_strategy: Some(execution_strategy.as_str().to_string()),
+        conversion_mode: Some(conversion_mode.as_str().to_string()),
+        request_id: Some(plan.request_id),
+        candidate_id: plan.candidate_id,
+        provider_name: plan.provider_name,
+        provider_id: Some(plan.provider_id),
+        endpoint_id: Some(plan.endpoint_id),
+        key_id: Some(plan.key_id),
+        upstream_base_url: None,
+        upstream_url: Some(plan.url),
+        provider_request_method: Some(plan.method),
+        auth_header: None,
+        auth_value: None,
+        provider_api_format: Some(plan.provider_api_format),
+        client_api_format: Some(plan.client_api_format),
+        provider_contract: Some(provider_contract),
+        client_contract: Some(client_contract),
+        model_name: plan.model_name,
+        mapped_model: None,
+        prompt_cache_key: None,
+        extra_headers: BTreeMap::new(),
+        provider_request_headers: plan.headers,
+        provider_request_body: None,
+        provider_request_body_base64: None,
+        content_type: plan.content_type,
+        proxy: plan.proxy,
+        tls_profile: plan.tls_profile,
+        timeouts: plan.timeouts,
+        upstream_is_stream: true,
+        report_kind: None,
+        report_context: None,
+        auth_context: decision.auth_context.clone(),
+    }))
+}
--- a/apps/aether-gateway/src/ai_pipeline/planner/decision/sync.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/decision/sync.rs
@@ -0,0 +1,222 @@
+use std::collections::BTreeMap;
+
+use url::Url;
+
+use crate::gateway::ai_pipeline::planner::common::{
+    EXECUTION_RUNTIME_SYNC_DECISION_ACTION, GEMINI_FILES_DELETE_PLAN_KIND,
+    GEMINI_FILES_GET_PLAN_KIND, GEMINI_FILES_LIST_PLAN_KIND, GEMINI_VIDEO_CANCEL_SYNC_PLAN_KIND,
+    OPENAI_VIDEO_CANCEL_SYNC_PLAN_KIND, OPENAI_VIDEO_DELETE_SYNC_PLAN_KIND,
+    OPENAI_VIDEO_REMIX_SYNC_PLAN_KIND,
+};
+use crate::gateway::scheduler::resolve_execution_runtime_sync_plan_kind;
+use crate::gateway::{
+    resolve_execution_runtime_auth_context, AppState, GatewayControlDecision,
+    GatewayControlSyncDecisionResponse, GatewayError,
+};
+
+pub(crate) async fn maybe_build_sync_decision_payload_impl(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    body_json: &serde_json::Value,
+    body_base64: Option<&str>,
+    body_is_empty: bool,
+) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
+    let Some(plan_kind) = resolve_execution_runtime_sync_plan_kind(parts, decision) else {
+        return Ok(None);
+    };
+
+    if let Some(payload) = maybe_build_local_video_task_follow_up_sync_decision_payload(
+        state, parts, body_json, trace_id, decision, plan_kind,
+    )
+    .await?
+    {
+        return Ok(Some(payload));
+    }
+
+    if let Some(payload) = super::maybe_build_sync_local_video_decision_payload(
+        state, parts, body_json, trace_id, decision, plan_kind,
+    )
+    .await?
+    {
+        return Ok(Some(payload));
+    }
+
+    if let Some(payload) = super::maybe_build_sync_local_decision_payload(
+        state, parts, trace_id, decision, body_json, plan_kind,
+    )
+    .await?
+    {
+        return Ok(Some(payload));
+    }
+
+    if let Some(payload) = super::maybe_build_sync_local_openai_cli_decision_payload(
+        state, parts, trace_id, decision, body_json, plan_kind,
+    )
+    .await?
+    {
+        return Ok(Some(payload));
+    }
+
+    if let Some(payload) = super::maybe_build_sync_local_standard_decision_payload(
+        state, parts, trace_id, decision, body_json, plan_kind,
+    )
+    .await?
+    {
+        return Ok(Some(payload));
+    }
+
+    if let Some(payload) = super::maybe_build_sync_local_same_format_provider_decision_payload(
+        state, parts, trace_id, decision, body_json, plan_kind,
+    )
+    .await?
+    {
+        return Ok(Some(payload));
+    }
+
+    if matches!(
+        plan_kind,
+        GEMINI_FILES_LIST_PLAN_KIND | GEMINI_FILES_GET_PLAN_KIND | GEMINI_FILES_DELETE_PLAN_KIND
+    ) {
+        if let Some(payload) = super::maybe_build_sync_local_gemini_files_decision_payload(
+            state,
+            parts,
+            body_json,
+            body_base64,
+            body_is_empty,
+            trace_id,
+            decision,
+            plan_kind,
+        )
+        .await?
+        {
+            return Ok(Some(payload));
+        }
+    }
+
+    Ok(None)
+}
+
+async fn maybe_build_local_video_task_follow_up_sync_decision_payload(
+    state: &AppState,
+    parts: &http::request::Parts,
+    body_json: &serde_json::Value,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    plan_kind: &str,
+) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
+    if !matches!(
+        plan_kind,
+        OPENAI_VIDEO_REMIX_SYNC_PLAN_KIND
+            | OPENAI_VIDEO_CANCEL_SYNC_PLAN_KIND
+            | OPENAI_VIDEO_DELETE_SYNC_PLAN_KIND
+            | GEMINI_VIDEO_CANCEL_SYNC_PLAN_KIND
+    ) {
+        return Ok(None);
+    }
+
+    let _ = state
+        .hydrate_video_task_for_route(decision.route_family.as_deref(), parts.uri.path())
+        .await?;
+
+    let auth_context = resolve_execution_runtime_auth_context(
+        state,
+        decision,
+        &parts.headers,
+        &parts.uri,
+        trace_id,
+    )
+    .await?;
+    let Some(auth_context) = auth_context else {
+        return Ok(None);
+    };
+    let Some(follow_up) = state.video_tasks.prepare_follow_up_sync_plan(
+        plan_kind,
+        parts.uri.path(),
+        Some(body_json),
+        Some(&auth_context),
+        trace_id,
+    ) else {
+        return Ok(None);
+    };
+
+    let auth_pair = extract_auth_header_pair(&follow_up.plan.headers);
+    let execution_strategy =
+        if follow_up.plan.provider_api_format == follow_up.plan.client_api_format {
+            crate::gateway::ExecutionStrategy::LocalSameFormat
+        } else {
+            crate::gateway::ExecutionStrategy::LocalCrossFormat
+        };
+    let conversion_mode = if follow_up.plan.provider_api_format == follow_up.plan.client_api_format
+    {
+        crate::gateway::ConversionMode::None
+    } else {
+        crate::gateway::ConversionMode::Bidirectional
+    };
+
+    Ok(Some(GatewayControlSyncDecisionResponse {
+        action: EXECUTION_RUNTIME_SYNC_DECISION_ACTION.to_string(),
+        decision_kind: Some(plan_kind.to_string()),
+        execution_strategy: Some(execution_strategy.as_str().to_string()),
+        conversion_mode: Some(conversion_mode.as_str().to_string()),
+        request_id: Some(trace_id.to_string()),
+        candidate_id: follow_up.plan.candidate_id.clone(),
+        provider_name: follow_up.plan.provider_name.clone(),
+        provider_id: Some(follow_up.plan.provider_id.clone()),
+        endpoint_id: Some(follow_up.plan.endpoint_id.clone()),
+        key_id: Some(follow_up.plan.key_id.clone()),
+        upstream_base_url: infer_upstream_base_url(&follow_up.plan.url),
+        upstream_url: Some(follow_up.plan.url.clone()),
+        provider_request_method: Some(follow_up.plan.method.clone()),
+        auth_header: auth_pair.as_ref().map(|(name, _)| name.clone()),
+        auth_value: auth_pair.as_ref().map(|(_, value)| value.clone()),
+        provider_api_format: Some(follow_up.plan.provider_api_format.clone()),
+        client_api_format: Some(follow_up.plan.client_api_format.clone()),
+        provider_contract: Some(follow_up.plan.provider_api_format.clone()),
+        client_contract: Some(follow_up.plan.client_api_format.clone()),
+        model_name: follow_up.plan.model_name.clone(),
+        mapped_model: None,
+        prompt_cache_key: None,
+        extra_headers: BTreeMap::new(),
+        provider_request_headers: follow_up.plan.headers.clone(),
+        provider_request_body: follow_up.plan.body.json_body.clone(),
+        provider_request_body_base64: follow_up.plan.body.body_bytes_b64.clone(),
+        content_type: follow_up.plan.content_type.clone(),
+        proxy: follow_up.plan.proxy.clone(),
+        tls_profile: follow_up.plan.tls_profile.clone(),
+        timeouts: follow_up.plan.timeouts.clone(),
+        upstream_is_stream: false,
+        report_kind: follow_up.report_kind,
+        report_context: follow_up.report_context,
+        auth_context: Some(auth_context),
+    }))
+}
+
+fn extract_auth_header_pair(headers: &BTreeMap<String, String>) -> Option<(String, String)> {
+    [
+        "authorization",
+        "x-api-key",
+        "api-key",
+        "x-goog-api-key",
+        "proxy-authorization",
+    ]
+    .into_iter()
+    .find_map(|name| {
+        headers
+            .iter()
+            .find(|(header_name, _)| header_name.eq_ignore_ascii_case(name))
+            .map(|(header_name, value)| (header_name.clone(), value.clone()))
+    })
+}
+
+fn infer_upstream_base_url(upstream_url: &str) -> Option<String> {
+    let parsed = Url::parse(upstream_url).ok()?;
+    let host = parsed.host_str()?;
+    let mut base = format!("{}://{}", parsed.scheme(), host);
+    if let Some(port) = parsed.port() {
+        base.push(':');
+        base.push_str(port.to_string().as_str());
+    }
+    Some(base)
+}
--- a/apps/aether-gateway/src/ai_pipeline/planner/family_core.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/family_core.rs
@@ -0,0 +1,132 @@
+use axum::body::Body;
+use axum::http::Response;
+
+use crate::gateway::ai_pipeline::planner::plan_builders::{
+    LocalStreamPlanAndReport, LocalSyncPlanAndReport,
+};
+use crate::gateway::request_candidates::record_local_request_candidate_status;
+use crate::gateway::{
+    execute_execution_runtime_stream, execute_execution_runtime_sync, AppState,
+    GatewayControlDecision, GatewayError,
+};
+
+pub(crate) trait LocalPlanAndReport {
+    fn plan(&self) -> &aether_contracts::ExecutionPlan;
+
+    fn report_kind(&self) -> Option<String>;
+
+    fn report_context(&self) -> Option<serde_json::Value>;
+}
+
+impl LocalPlanAndReport for LocalSyncPlanAndReport {
+    fn plan(&self) -> &aether_contracts::ExecutionPlan {
+        &self.plan
+    }
+
+    fn report_kind(&self) -> Option<String> {
+        self.report_kind.clone()
+    }
+
+    fn report_context(&self) -> Option<serde_json::Value> {
+        self.report_context.clone()
+    }
+}
+
+impl LocalPlanAndReport for LocalStreamPlanAndReport {
+    fn plan(&self) -> &aether_contracts::ExecutionPlan {
+        &self.plan
+    }
+
+    fn report_kind(&self) -> Option<String> {
+        self.report_kind.clone()
+    }
+
+    fn report_context(&self) -> Option<serde_json::Value> {
+        self.report_context.clone()
+    }
+}
+
+pub(crate) async fn execute_sync_plan_and_reports<T>(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    plan_kind: &str,
+    plan_and_reports: Vec<T>,
+) -> Result<Option<Response<Body>>, GatewayError>
+where
+    T: LocalPlanAndReport,
+{
+    let mut remaining = plan_and_reports.into_iter();
+    while let Some(plan_and_report) = remaining.next() {
+        if let Some(response) = execute_execution_runtime_sync(
+            state,
+            parts.uri.path(),
+            plan_and_report.plan().clone(),
+            trace_id,
+            decision,
+            plan_kind,
+            plan_and_report.report_kind(),
+            plan_and_report.report_context(),
+        )
+        .await?
+        {
+            mark_unused_local_candidates(state, remaining.collect()).await;
+            return Ok(Some(response));
+        }
+    }
+
+    Ok(None)
+}
+
+pub(crate) async fn execute_stream_plan_and_reports<T>(
+    state: &AppState,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    plan_kind: &str,
+    plan_and_reports: Vec<T>,
+) -> Result<Option<Response<Body>>, GatewayError>
+where
+    T: LocalPlanAndReport,
+{
+    let mut remaining = plan_and_reports.into_iter();
+    while let Some(plan_and_report) = remaining.next() {
+        if let Some(response) = execute_execution_runtime_stream(
+            state,
+            plan_and_report.plan().clone(),
+            trace_id,
+            decision,
+            plan_kind,
+            plan_and_report.report_kind(),
+            plan_and_report.report_context(),
+        )
+        .await?
+        {
+            mark_unused_local_candidates(state, remaining.collect()).await;
+            return Ok(Some(response));
+        }
+    }
+
+    Ok(None)
+}
+
+pub(crate) async fn mark_unused_local_candidates<T>(state: &AppState, remaining: Vec<T>)
+where
+    T: LocalPlanAndReport,
+{
+    for plan_and_report in remaining {
+        record_local_request_candidate_status(
+            state,
+            plan_and_report.plan(),
+            plan_and_report.report_context().as_ref(),
+            aether_data::repository::candidates::RequestCandidateStatus::Unused,
+            None,
+            None,
+            None,
+            None,
+            None,
+            None,
+        )
+        .await;
+    }
+}
--- a/apps/aether-gateway/src/ai_pipeline/planner/local_path.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/local_path.rs
@@ -0,0 +1,25 @@
+use axum::body::{Body, Bytes};
+use axum::http::Response;
+
+use crate::gateway::intent;
+use crate::gateway::{AppState, GatewayControlDecision, GatewayError};
+
+pub(crate) async fn maybe_execute_sync_local_path(
+    state: &AppState,
+    parts: &http::request::Parts,
+    body_bytes: &Bytes,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+) -> Result<Option<Response<Body>>, GatewayError> {
+    intent::maybe_execute_via_sync_intent_path(state, parts, body_bytes, trace_id, decision).await
+}
+
+pub(crate) async fn maybe_execute_stream_local_path(
+    state: &AppState,
+    parts: &http::request::Parts,
+    body_bytes: &Bytes,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+) -> Result<Option<Response<Body>>, GatewayError> {
+    intent::maybe_execute_via_stream_intent_path(state, parts, body_bytes, trace_id, decision).await
+}
--- a/apps/aether-gateway/src/ai_pipeline/planner/mod.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/mod.rs
@@ -0,0 +1,134 @@
+use crate::gateway::{AppState, GatewayControlDecision, GatewayError};
+
+pub(crate) mod candidate_affinity;
+pub(crate) mod common;
+pub(crate) mod contracts;
+mod decision;
+pub(crate) mod family_core;
+pub(crate) mod local_path;
+pub(crate) mod passthrough;
+pub(crate) mod plan_builders;
+pub(crate) mod specialized;
+pub(crate) mod standard;
+
+pub(crate) use self::candidate_affinity::prefer_local_tunnel_owner_candidates;
+pub(crate) use self::common::{
+    parse_direct_request_body, CLAUDE_CHAT_STREAM_PLAN_KIND, CLAUDE_CHAT_SYNC_PLAN_KIND,
+    CLAUDE_CLI_STREAM_PLAN_KIND, CLAUDE_CLI_SYNC_PLAN_KIND, EXECUTION_RUNTIME_STREAM_ACTION,
+    EXECUTION_RUNTIME_STREAM_DECISION_ACTION, EXECUTION_RUNTIME_SYNC_ACTION,
+    EXECUTION_RUNTIME_SYNC_DECISION_ACTION, GEMINI_CHAT_STREAM_PLAN_KIND,
+    GEMINI_CHAT_SYNC_PLAN_KIND, GEMINI_CLI_STREAM_PLAN_KIND, GEMINI_CLI_SYNC_PLAN_KIND,
+    GEMINI_FILES_DELETE_PLAN_KIND, GEMINI_FILES_DOWNLOAD_PLAN_KIND, GEMINI_FILES_GET_PLAN_KIND,
+    GEMINI_FILES_LIST_PLAN_KIND, GEMINI_FILES_UPLOAD_PLAN_KIND, GEMINI_VIDEO_CANCEL_SYNC_PLAN_KIND,
+    GEMINI_VIDEO_CREATE_SYNC_PLAN_KIND, OPENAI_CHAT_STREAM_PLAN_KIND, OPENAI_CHAT_SYNC_PLAN_KIND,
+    OPENAI_CLI_STREAM_PLAN_KIND, OPENAI_CLI_SYNC_PLAN_KIND, OPENAI_COMPACT_STREAM_PLAN_KIND,
+    OPENAI_COMPACT_SYNC_PLAN_KIND, OPENAI_VIDEO_CANCEL_SYNC_PLAN_KIND,
+    OPENAI_VIDEO_CONTENT_PLAN_KIND, OPENAI_VIDEO_CREATE_SYNC_PLAN_KIND,
+    OPENAI_VIDEO_DELETE_SYNC_PLAN_KIND, OPENAI_VIDEO_REMIX_SYNC_PLAN_KIND,
+};
+pub(crate) use self::contracts::{
+    build_gateway_plan_request, generic_decision_missing_exact_provider_request,
+    GatewayControlPlanRequest, GatewayControlPlanResponse, GatewayControlSyncDecisionResponse,
+};
+pub(crate) use crate::gateway::ai_pipeline::conversion::request::{
+    convert_openai_chat_request_to_claude_request, convert_openai_chat_request_to_gemini_request,
+    convert_openai_chat_request_to_openai_cli_request, extract_openai_text_content,
+    normalize_openai_cli_request_to_openai_chat_request, parse_openai_tool_result_content,
+};
+pub(crate) use crate::gateway::scheduler::{
+    is_matching_stream_request,
+    resolve_execution_runtime_stream_plan_kind as resolve_stream_plan_kind,
+    resolve_execution_runtime_sync_plan_kind as resolve_sync_plan_kind,
+};
+pub(crate) use passthrough::{
+    maybe_build_stream_local_same_format_provider_decision_payload,
+    maybe_build_sync_local_same_format_provider_decision_payload,
+    maybe_execute_stream_via_local_same_format_provider_decision,
+    maybe_execute_sync_via_local_same_format_provider_decision,
+};
+pub(crate) use specialized::{
+    maybe_build_stream_local_gemini_files_decision_payload,
+    maybe_build_sync_local_gemini_files_decision_payload,
+    maybe_build_sync_local_video_decision_payload,
+    maybe_execute_stream_via_local_gemini_files_decision,
+    maybe_execute_sync_via_local_gemini_files_decision,
+    maybe_execute_sync_via_local_video_decision,
+};
+pub(crate) use local_path::{maybe_execute_stream_local_path, maybe_execute_sync_local_path};
+pub(crate) use standard::{
+    copy_request_number_field, copy_request_number_field_as,
+    map_openai_reasoning_effort_to_claude_output, map_openai_reasoning_effort_to_gemini_budget,
+    maybe_build_stream_local_decision_payload,
+    maybe_build_stream_local_openai_cli_decision_payload,
+    maybe_build_stream_local_standard_decision_payload, maybe_build_sync_local_decision_payload,
+    maybe_build_sync_local_openai_cli_decision_payload,
+    maybe_build_sync_local_standard_decision_payload, maybe_execute_stream_via_local_decision,
+    maybe_execute_stream_via_local_openai_cli_decision,
+    maybe_execute_stream_via_local_standard_decision, maybe_execute_sync_via_local_decision,
+    maybe_execute_sync_via_local_openai_cli_decision,
+    maybe_execute_sync_via_local_standard_decision, parse_openai_stop_sequences,
+    resolve_openai_chat_max_tokens, value_as_u64,
+};
+
+pub(crate) async fn maybe_build_sync_decision_payload(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    body_json: &serde_json::Value,
+    body_base64: Option<&str>,
+    body_is_empty: bool,
+) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
+    decision::maybe_build_sync_decision_payload(
+        state,
+        parts,
+        trace_id,
+        decision,
+        body_json,
+        body_base64,
+        body_is_empty,
+    )
+    .await
+}
+
+pub(crate) async fn maybe_build_stream_decision_payload(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    body_json: &serde_json::Value,
+) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
+    decision::maybe_build_stream_decision_payload(state, parts, trace_id, decision, body_json).await
+}
+
+pub(crate) async fn maybe_build_sync_plan_payload(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    body_json: &serde_json::Value,
+    body_base64: Option<&str>,
+    body_is_empty: bool,
+) -> Result<Option<GatewayControlPlanResponse>, GatewayError> {
+    decision::maybe_build_sync_plan_payload_impl(
+        state,
+        parts,
+        trace_id,
+        decision,
+        body_json,
+        body_base64,
+        body_is_empty,
+    )
+    .await
+}
+
+pub(crate) async fn maybe_build_stream_plan_payload(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    body_json: &serde_json::Value,
+) -> Result<Option<GatewayControlPlanResponse>, GatewayError> {
+    decision::maybe_build_stream_plan_payload_impl(state, parts, trace_id, decision, body_json)
+        .await
+}
--- a/apps/aether-gateway/src/ai_pipeline/planner/passthrough/mod.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/passthrough/mod.rs
@@ -0,0 +1,11 @@
+//! Requests that can stay in the same public/provider contract family.
+
+mod provider;
+
+pub(crate) use self::provider::{
+    maybe_build_stream_local_same_format_provider_decision_payload,
+    maybe_build_sync_local_same_format_provider_decision_payload,
+    maybe_execute_stream_via_local_same_format_provider_decision,
+    maybe_execute_sync_via_local_same_format_provider_decision,
+};
+pub(crate) use crate::gateway::provider_transport::provider_type_supports_local_same_format_transport;
--- a/apps/aether-gateway/src/ai_pipeline/planner/passthrough/plan_builders.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/passthrough/plan_builders.rs
@@ -0,0 +1,233 @@
+use aether_contracts::{ExecutionPlan, RequestBody};
+
+use super::{augment_sync_report_context, LocalStreamPlanAndReport, LocalSyncPlanAndReport};
+use crate::gateway::{GatewayControlSyncDecisionResponse, GatewayError};
+
+pub(crate) fn build_passthrough_sync_plan_from_decision(
+    parts: &http::request::Parts,
+    payload: GatewayControlSyncDecisionResponse,
+) -> Result<Option<LocalSyncPlanAndReport>, GatewayError> {
+    let Some(request_id) = payload
+        .request_id
+        .clone()
+        .filter(|value| !value.trim().is_empty())
+    else {
+        return Ok(None);
+    };
+    let Some(provider_id) = payload
+        .provider_id
+        .clone()
+        .filter(|value| !value.trim().is_empty())
+    else {
+        return Ok(None);
+    };
+    let Some(endpoint_id) = payload
+        .endpoint_id
+        .clone()
+        .filter(|value| !value.trim().is_empty())
+    else {
+        return Ok(None);
+    };
+    let Some(key_id) = payload
+        .key_id
+        .clone()
+        .filter(|value| !value.trim().is_empty())
+    else {
+        return Ok(None);
+    };
+    let Some(provider_api_format) = payload
+        .provider_api_format
+        .clone()
+        .filter(|value| !value.trim().is_empty())
+    else {
+        return Ok(None);
+    };
+    let Some(client_api_format) = payload
+        .client_api_format
+        .clone()
+        .filter(|value| !value.trim().is_empty())
+    else {
+        return Ok(None);
+    };
+    let Some(upstream_url) = payload
+        .upstream_url
+        .clone()
+        .filter(|value| !value.trim().is_empty())
+    else {
+        return Ok(None);
+    };
+    let (request_body, provider_request_body_for_report) = resolve_passthrough_sync_request_body(
+        payload.provider_request_body.clone(),
+        payload.provider_request_body_base64.clone(),
+    );
+
+    let plan = ExecutionPlan {
+        request_id,
+        candidate_id: payload.candidate_id.clone(),
+        provider_name: payload.provider_name.clone(),
+        provider_id,
+        endpoint_id,
+        key_id,
+        method: payload
+            .provider_request_method
+            .clone()
+            .filter(|value| !value.trim().is_empty())
+            .unwrap_or_else(|| parts.method.to_string()),
+        url: upstream_url,
+        headers: payload.provider_request_headers.clone(),
+        content_type: payload.content_type.clone().or_else(|| {
+            payload
+                .provider_request_headers
+                .get("content-type")
+                .cloned()
+        }),
+        content_encoding: None,
+        body: request_body,
+        stream: false,
+        client_api_format,
+        provider_api_format,
+        model_name: payload.model_name.clone(),
+        proxy: payload.proxy.clone(),
+        tls_profile: payload.tls_profile.clone(),
+        timeouts: payload.timeouts.clone(),
+    };
+
+    let report_context = augment_sync_report_context(
+        payload.report_context,
+        &plan.headers,
+        &provider_request_body_for_report,
+    )?;
+
+    Ok(Some(LocalSyncPlanAndReport {
+        plan,
+        report_kind: payload.report_kind,
+        report_context,
+    }))
+}
+
+pub(crate) fn build_passthrough_stream_plan_from_decision(
+    parts: &http::request::Parts,
+    payload: GatewayControlSyncDecisionResponse,
+) -> Result<Option<LocalStreamPlanAndReport>, GatewayError> {
+    let Some(request_id) = payload
+        .request_id
+        .clone()
+        .filter(|value| !value.trim().is_empty())
+    else {
+        return Ok(None);
+    };
+    let Some(provider_id) = payload
+        .provider_id
+        .clone()
+        .filter(|value| !value.trim().is_empty())
+    else {
+        return Ok(None);
+    };
+    let Some(endpoint_id) = payload
+        .endpoint_id
+        .clone()
+        .filter(|value| !value.trim().is_empty())
+    else {
+        return Ok(None);
+    };
+    let Some(key_id) = payload
+        .key_id
+        .clone()
+        .filter(|value| !value.trim().is_empty())
+    else {
+        return Ok(None);
+    };
+    let Some(provider_api_format) = payload
+        .provider_api_format
+        .clone()
+        .filter(|value| !value.trim().is_empty())
+    else {
+        return Ok(None);
+    };
+    let Some(client_api_format) = payload
+        .client_api_format
+        .clone()
+        .filter(|value| !value.trim().is_empty())
+    else {
+        return Ok(None);
+    };
+    let Some(upstream_url) = payload
+        .upstream_url
+        .clone()
+        .filter(|value| !value.trim().is_empty())
+    else {
+        return Ok(None);
+    };
+    let plan = ExecutionPlan {
+        request_id,
+        candidate_id: payload.candidate_id.clone(),
+        provider_name: payload.provider_name.clone(),
+        provider_id,
+        endpoint_id,
+        key_id,
+        method: parts.method.to_string(),
+        url: upstream_url,
+        headers: payload.provider_request_headers.clone(),
+        content_type: payload.content_type.clone().or_else(|| {
+            payload
+                .provider_request_headers
+                .get("content-type")
+                .cloned()
+        }),
+        content_encoding: None,
+        body: RequestBody {
+            json_body: None,
+            body_bytes_b64: None,
+            body_ref: None,
+        },
+        stream: true,
+        client_api_format,
+        provider_api_format,
+        model_name: payload.model_name.clone(),
+        proxy: payload.proxy.clone(),
+        tls_profile: payload.tls_profile.clone(),
+        timeouts: payload.timeouts.clone(),
+    };
+
+    Ok(Some(LocalStreamPlanAndReport {
+        plan,
+        report_kind: payload.report_kind,
+        report_context: payload.report_context,
+    }))
+}
+
+fn resolve_passthrough_sync_request_body(
+    provider_request_body: Option<serde_json::Value>,
+    provider_request_body_base64: Option<String>,
+) -> (RequestBody, serde_json::Value) {
+    if let Some(body_bytes_b64) = provider_request_body_base64
+        .as_ref()
+        .map(|value| value.trim())
+        .filter(|value| !value.is_empty())
+        .map(ToOwned::to_owned)
+    {
+        return (
+            RequestBody {
+                json_body: None,
+                body_bytes_b64: Some(body_bytes_b64.clone()),
+                body_ref: None,
+            },
+            serde_json::json!({"body_bytes_b64": body_bytes_b64}),
+        );
+    }
+
+    match provider_request_body.unwrap_or(serde_json::Value::Null) {
+        serde_json::Value::Null => (
+            RequestBody {
+                json_body: None,
+                body_bytes_b64: None,
+                body_ref: None,
+            },
+            serde_json::Value::Null,
+        ),
+        other => {
+            let report_body = other.clone();
+            (RequestBody::from_json(other), report_body)
+        }
+    }
+}
--- a/apps/aether-gateway/src/ai_pipeline/planner/passthrough/provider.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/passthrough/provider.rs
@@ -0,0 +1,76 @@
+use axum::body::Body;
+use axum::http::Response;
+use std::collections::BTreeMap;
+use url::form_urlencoded;
+
+use aether_data::repository::candidates::{RequestCandidateStatus, UpsertRequestCandidateRecord};
+use serde_json::{json, Value};
+use tracing::warn;
+use uuid::Uuid;
+
+use crate::gateway::headers::collect_control_headers;
+use crate::gateway::provider_transport::{
+    apply_local_body_rules, apply_local_header_rules, build_antigravity_safe_v1internal_request,
+    build_antigravity_static_identity_headers, build_antigravity_v1internal_url,
+    build_claude_code_messages_url, build_claude_code_passthrough_headers,
+    build_claude_messages_url, build_gemini_content_url,
+    build_kiro_generate_assistant_response_url, build_kiro_provider_headers,
+    build_kiro_provider_request_body, build_openai_passthrough_headers, build_passthrough_headers,
+    build_passthrough_path_url, build_vertex_api_key_gemini_content_url,
+    classify_local_antigravity_request_support, ensure_upstream_auth_header,
+    resolve_local_gemini_auth, resolve_local_standard_auth,
+    resolve_local_vertex_api_key_query_auth, resolve_transport_execution_timeouts,
+    resolve_transport_proxy_snapshot_with_tunnel_affinity, resolve_transport_tls_profile,
+    sanitize_claude_code_request_body, supports_local_claude_code_transport_with_network,
+    supports_local_gemini_transport_with_network,
+    supports_local_kiro_request_transport_with_network,
+    supports_local_standard_transport_with_network,
+    supports_local_vertex_api_key_gemini_transport_with_network, AntigravityEnvelopeRequestType,
+    AntigravityRequestEnvelopeSupport, AntigravityRequestSideSupport, AntigravityRequestUrlAction,
+    LocalResolvedOAuthRequestAuth, KIRO_ENVELOPE_NAME,
+};
+use crate::gateway::request_candidates::{
+    current_unix_secs, record_local_request_candidate_status,
+};
+use crate::gateway::ai_pipeline::planner::plan_builders::{
+    LocalStreamPlanAndReport, LocalSyncPlanAndReport,
+};
+use crate::gateway::ai_pipeline::planner::prefer_local_tunnel_owner_candidates;
+use crate::gateway::scheduler::{
+    list_selectable_candidates, GatewayMinimalCandidateSelectionCandidate,
+};
+use crate::gateway::{
+    append_execution_contract_fields_to_value, execute_execution_runtime_stream,
+    execute_execution_runtime_sync, AppState, ConversionMode, ExecutionStrategy,
+    GatewayControlDecision, GatewayControlSyncDecisionResponse, GatewayError,
+};
+use crate::gateway::ai_pipeline::planner::{
+    EXECUTION_RUNTIME_STREAM_DECISION_ACTION, EXECUTION_RUNTIME_SYNC_DECISION_ACTION,
+};
+
+mod family;
+mod plans;
+mod request;
+
+pub(super) use self::family::{
+    materialize_local_same_format_provider_candidate_attempts,
+    maybe_build_local_same_format_provider_decision_payload_for_candidate,
+    resolve_local_same_format_provider_decision_input, LocalSameFormatProviderFamily,
+    LocalSameFormatProviderSpec,
+};
+pub(crate) use self::family::{
+    maybe_build_stream_local_same_format_provider_decision_payload,
+    maybe_build_sync_local_same_format_provider_decision_payload,
+    maybe_execute_stream_via_local_same_format_provider_decision,
+    maybe_execute_sync_via_local_same_format_provider_decision,
+};
+use self::plans::{
+    build_local_stream_plan_and_reports, build_local_sync_plan_and_reports, resolve_stream_spec,
+    resolve_sync_spec,
+};
+use self::request::{
+    build_same_format_provider_request_body, build_same_format_upstream_url,
+    extract_gemini_model_from_path,
+};
+
+const ANTIGRAVITY_ENVELOPE_NAME: &str = "antigravity:v1internal";
--- a/apps/aether-gateway/src/ai_pipeline/planner/passthrough/provider/family/candidates.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/passthrough/provider/family/candidates.rs
@@ -0,0 +1,163 @@
+use aether_data::repository::candidates::{RequestCandidateStatus, UpsertRequestCandidateRecord};
+use serde_json::json;
+use tracing::warn;
+use uuid::Uuid;
+
+use crate::gateway::request_candidates::current_unix_secs;
+use crate::gateway::scheduler::list_selectable_candidates;
+use crate::gateway::{
+    append_execution_contract_fields_to_value, AppState, ConversionMode, ExecutionStrategy,
+    GatewayControlDecision, GatewayError,
+};
+
+use super::types::{
+    LocalSameFormatProviderCandidateAttempt, LocalSameFormatProviderDecisionInput,
+    LocalSameFormatProviderFamily, LocalSameFormatProviderSpec,
+};
+
+pub(crate) async fn resolve_local_same_format_provider_decision_input(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    body_json: &serde_json::Value,
+    spec: LocalSameFormatProviderSpec,
+) -> Option<LocalSameFormatProviderDecisionInput> {
+    let Some(auth_context) = decision.auth_context.clone().filter(|auth_context| {
+        !auth_context.user_id.trim().is_empty() && !auth_context.api_key_id.trim().is_empty()
+    }) else {
+        return None;
+    };
+
+    let requested_model = match spec.family {
+        LocalSameFormatProviderFamily::Standard => body_json
+            .get("model")
+            .and_then(|value| value.as_str())
+            .map(str::trim)
+            .filter(|value| !value.is_empty())
+            .map(ToOwned::to_owned)?,
+        LocalSameFormatProviderFamily::Gemini => {
+            super::super::request::extract_gemini_model_from_path(parts.uri.path())?
+        }
+    };
+
+    let auth_snapshot = match state
+        .read_auth_api_key_snapshot(
+            &auth_context.user_id,
+            &auth_context.api_key_id,
+            current_unix_secs(),
+        )
+        .await
+    {
+        Ok(Some(snapshot)) => snapshot,
+        Ok(None) => return None,
+        Err(err) => {
+            warn!(
+                trace_id = %trace_id,
+                api_format = spec.api_format,
+                error = ?err,
+                "gateway local same-format decision auth snapshot read failed"
+            );
+            return None;
+        }
+    };
+
+    Some(LocalSameFormatProviderDecisionInput {
+        auth_context,
+        requested_model,
+        auth_snapshot,
+    })
+}
+
+pub(crate) async fn materialize_local_same_format_provider_candidate_attempts(
+    state: &AppState,
+    trace_id: &str,
+    input: &LocalSameFormatProviderDecisionInput,
+    spec: LocalSameFormatProviderSpec,
+) -> Result<Vec<LocalSameFormatProviderCandidateAttempt>, GatewayError> {
+    let candidates = list_selectable_candidates(
+        state,
+        spec.api_format,
+        &input.requested_model,
+        spec.require_streaming,
+        Some(&input.auth_snapshot),
+        current_unix_secs(),
+    )
+    .await?;
+    let candidates =
+        crate::gateway::ai_pipeline::planner::prefer_local_tunnel_owner_candidates(state, candidates)
+            .await;
+
+    let created_at_unix_secs = current_unix_secs();
+    let mut attempts = Vec::with_capacity(candidates.len());
+    for (candidate_index, candidate) in candidates.into_iter().enumerate() {
+        let generated_candidate_id = Uuid::new_v4().to_string();
+        let extra_data = append_execution_contract_fields_to_value(
+            json!({
+                "provider_api_format": spec.api_format,
+                "client_api_format": spec.api_format,
+                "global_model_id": candidate.global_model_id.clone(),
+                "global_model_name": candidate.global_model_name.clone(),
+                "model_id": candidate.model_id.clone(),
+                "selected_provider_model_name": candidate.selected_provider_model_name.clone(),
+                "mapping_matched_model": candidate.mapping_matched_model.clone(),
+                "provider_name": candidate.provider_name.clone(),
+                "key_name": candidate.key_name.clone(),
+            }),
+            ExecutionStrategy::LocalSameFormat,
+            ConversionMode::None,
+            spec.api_format,
+            spec.api_format,
+        );
+
+        let candidate_id = match state
+            .upsert_request_candidate(UpsertRequestCandidateRecord {
+                id: generated_candidate_id.clone(),
+                request_id: trace_id.to_string(),
+                user_id: Some(input.auth_context.user_id.clone()),
+                api_key_id: Some(input.auth_context.api_key_id.clone()),
+                username: None,
+                api_key_name: None,
+                candidate_index: candidate_index as u32,
+                retry_index: 0,
+                provider_id: Some(candidate.provider_id.clone()),
+                endpoint_id: Some(candidate.endpoint_id.clone()),
+                key_id: Some(candidate.key_id.clone()),
+                status: RequestCandidateStatus::Available,
+                skip_reason: None,
+                is_cached: Some(false),
+                status_code: None,
+                error_type: None,
+                error_message: None,
+                latency_ms: None,
+                concurrent_requests: None,
+                extra_data: Some(extra_data),
+                required_capabilities: candidate.key_capabilities.clone(),
+                created_at_unix_secs: Some(created_at_unix_secs),
+                started_at_unix_secs: None,
+                finished_at_unix_secs: None,
+            })
+            .await
+        {
+            Ok(Some(stored)) => stored.id,
+            Ok(None) => generated_candidate_id.clone(),
+            Err(err) => {
+                warn!(
+                    trace_id = %trace_id,
+                    api_format = spec.api_format,
+                    error = ?err,
+                    "gateway local same-format decision request candidate upsert failed"
+                );
+                generated_candidate_id.clone()
+            }
+        };
+
+        attempts.push(LocalSameFormatProviderCandidateAttempt {
+            candidate,
+            candidate_index: candidate_index as u32,
+            candidate_id,
+        });
+    }
+
+    Ok(attempts)
+}
--- a/apps/aether-gateway/src/ai_pipeline/planner/passthrough/provider/family/execute.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/passthrough/provider/family/execute.rs
@@ -0,0 +1,141 @@
+use axum::body::Body;
+use axum::http::Response;
+
+use crate::gateway::ai_pipeline::planner::family_core::{
+    execute_stream_plan_and_reports, execute_sync_plan_and_reports,
+};
+use crate::gateway::ai_pipeline::planner::plan_builders::{
+    LocalStreamPlanAndReport, LocalSyncPlanAndReport,
+};
+use crate::gateway::{AppState, GatewayControlDecision, GatewayControlSyncDecisionResponse, GatewayError};
+
+use super::super::plans::{
+    build_local_stream_plan_and_reports, build_local_sync_plan_and_reports, resolve_stream_spec,
+    resolve_sync_spec,
+};
+use super::candidates::{
+    materialize_local_same_format_provider_candidate_attempts,
+    resolve_local_same_format_provider_decision_input,
+};
+use super::payload::maybe_build_local_same_format_provider_decision_payload_for_candidate;
+
+pub(crate) async fn maybe_execute_sync_via_local_same_format_provider_decision(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    body_json: &serde_json::Value,
+    plan_kind: &str,
+) -> Result<Option<Response<Body>>, GatewayError> {
+    let Some(spec) = resolve_sync_spec(plan_kind) else {
+        return Ok(None);
+    };
+
+    let plan_and_reports =
+        build_local_sync_plan_and_reports(state, parts, trace_id, decision, body_json, spec)
+            .await?;
+    if plan_and_reports.is_empty() {
+        return Ok(None);
+    }
+
+    execute_sync_plan_and_reports(state, parts, trace_id, decision, plan_kind, plan_and_reports)
+        .await
+}
+
+pub(crate) async fn maybe_execute_stream_via_local_same_format_provider_decision(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    body_json: &serde_json::Value,
+    plan_kind: &str,
+) -> Result<Option<Response<Body>>, GatewayError> {
+    let Some(spec) = resolve_stream_spec(plan_kind) else {
+        return Ok(None);
+    };
+
+    let plan_and_reports =
+        build_local_stream_plan_and_reports(state, parts, trace_id, decision, body_json, spec)
+            .await?;
+    if plan_and_reports.is_empty() {
+        return Ok(None);
+    }
+
+    execute_stream_plan_and_reports(state, trace_id, decision, plan_kind, plan_and_reports).await
+}
+
+pub(crate) async fn maybe_build_sync_local_same_format_provider_decision_payload(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    body_json: &serde_json::Value,
+    plan_kind: &str,
+) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
+    let Some(spec) = resolve_sync_spec(plan_kind) else {
+        return Ok(None);
+    };
+
+    let Some(input) = resolve_local_same_format_provider_decision_input(
+        state, parts, trace_id, decision, body_json, spec,
+    )
+    .await
+    else {
+        return Ok(None);
+    };
+
+    let attempts =
+        materialize_local_same_format_provider_candidate_attempts(state, trace_id, &input, spec)
+            .await?;
+
+    for attempt in attempts {
+        if let Some(payload) =
+            maybe_build_local_same_format_provider_decision_payload_for_candidate(
+                state, parts, trace_id, body_json, &input, attempt, spec,
+            )
+            .await
+        {
+            return Ok(Some(payload));
+        }
+    }
+
+    Ok(None)
+}
+
+pub(crate) async fn maybe_build_stream_local_same_format_provider_decision_payload(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    body_json: &serde_json::Value,
+    plan_kind: &str,
+) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
+    let Some(spec) = resolve_stream_spec(plan_kind) else {
+        return Ok(None);
+    };
+
+    let Some(input) = resolve_local_same_format_provider_decision_input(
+        state, parts, trace_id, decision, body_json, spec,
+    )
+    .await
+    else {
+        return Ok(None);
+    };
+
+    let attempts =
+        materialize_local_same_format_provider_candidate_attempts(state, trace_id, &input, spec)
+            .await?;
+
+    for attempt in attempts {
+        if let Some(payload) =
+            maybe_build_local_same_format_provider_decision_payload_for_candidate(
+                state, parts, trace_id, body_json, &input, attempt, spec,
+            )
+            .await
+        {
+            return Ok(Some(payload));
+        }
+    }
+
+    Ok(None)
+}
--- a/apps/aether-gateway/src/ai_pipeline/planner/passthrough/provider/family/mod.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/passthrough/provider/family/mod.rs
@@ -0,0 +1,17 @@
+mod candidates;
+mod execute;
+mod payload;
+mod types;
+
+pub(crate) use self::candidates::{
+    materialize_local_same_format_provider_candidate_attempts,
+    resolve_local_same_format_provider_decision_input,
+};
+pub(crate) use self::execute::{
+    maybe_build_stream_local_same_format_provider_decision_payload,
+    maybe_build_sync_local_same_format_provider_decision_payload,
+    maybe_execute_stream_via_local_same_format_provider_decision,
+    maybe_execute_sync_via_local_same_format_provider_decision,
+};
+pub(crate) use self::payload::maybe_build_local_same_format_provider_decision_payload_for_candidate;
+pub(crate) use self::types::{LocalSameFormatProviderFamily, LocalSameFormatProviderSpec};
--- a/apps/aether-gateway/src/ai_pipeline/planner/passthrough/provider/family/payload.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/passthrough/provider/family/payload.rs
@@ -0,0 +1,562 @@
+use std::collections::BTreeMap;
+
+use aether_data::repository::candidates::{RequestCandidateStatus, UpsertRequestCandidateRecord};
+use serde_json::json;
+use tracing::warn;
+
+use crate::gateway::headers::collect_control_headers;
+use crate::gateway::provider_transport::{
+    apply_local_header_rules, build_antigravity_safe_v1internal_request,
+    build_antigravity_static_identity_headers, build_claude_code_passthrough_headers,
+    build_openai_passthrough_headers, build_passthrough_headers,
+    classify_local_antigravity_request_support, ensure_upstream_auth_header,
+    resolve_local_gemini_auth, resolve_local_standard_auth,
+    resolve_local_vertex_api_key_query_auth, resolve_transport_execution_timeouts,
+    resolve_transport_proxy_snapshot_with_tunnel_affinity, resolve_transport_tls_profile,
+    supports_local_claude_code_transport_with_network,
+    supports_local_gemini_transport_with_network,
+    supports_local_kiro_request_transport_with_network,
+    supports_local_standard_transport_with_network,
+    supports_local_vertex_api_key_gemini_transport_with_network, AntigravityEnvelopeRequestType,
+    AntigravityRequestEnvelopeSupport, AntigravityRequestSideSupport,
+    LocalResolvedOAuthRequestAuth, KIRO_ENVELOPE_NAME,
+};
+use crate::gateway::request_candidates::current_unix_secs;
+use crate::gateway::scheduler::GatewayMinimalCandidateSelectionCandidate;
+use crate::gateway::{
+    append_execution_contract_fields_to_value, AppState, ConversionMode, ExecutionStrategy,
+    GatewayControlSyncDecisionResponse, EXECUTION_RUNTIME_STREAM_DECISION_ACTION,
+    EXECUTION_RUNTIME_SYNC_DECISION_ACTION,
+};
+
+use super::types::{
+    LocalSameFormatProviderCandidateAttempt, LocalSameFormatProviderDecisionInput,
+    LocalSameFormatProviderFamily, LocalSameFormatProviderSpec,
+};
+
+pub(crate) async fn maybe_build_local_same_format_provider_decision_payload_for_candidate(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    body_json: &serde_json::Value,
+    input: &LocalSameFormatProviderDecisionInput,
+    attempt: LocalSameFormatProviderCandidateAttempt,
+    spec: LocalSameFormatProviderSpec,
+) -> Option<GatewayControlSyncDecisionResponse> {
+    let LocalSameFormatProviderCandidateAttempt {
+        candidate,
+        candidate_index,
+        candidate_id,
+    } = attempt;
+
+    let transport = match state
+        .read_provider_transport_snapshot(
+            &candidate.provider_id,
+            &candidate.endpoint_id,
+            &candidate.key_id,
+        )
+        .await
+    {
+        Ok(Some(snapshot)) => snapshot,
+        Ok(None) => {
+            mark_skipped_local_same_format_provider_candidate(
+                state,
+                input,
+                trace_id,
+                &candidate,
+                candidate_index,
+                &candidate_id,
+                "transport_snapshot_missing",
+            )
+            .await;
+            return None;
+        }
+        Err(err) => {
+            warn!(
+                trace_id = %trace_id,
+                api_format = spec.api_format,
+                error = ?err,
+                "gateway local same-format decision provider transport read failed"
+            );
+            mark_skipped_local_same_format_provider_candidate(
+                state,
+                input,
+                trace_id,
+                &candidate,
+                candidate_index,
+                &candidate_id,
+                "transport_snapshot_read_failed",
+            )
+            .await;
+            return None;
+        }
+    };
+
+    let is_antigravity = transport
+        .provider
+        .provider_type
+        .trim()
+        .eq_ignore_ascii_case("antigravity");
+    let is_claude_code = transport
+        .provider
+        .provider_type
+        .trim()
+        .eq_ignore_ascii_case("claude_code");
+    let is_vertex = transport
+        .provider
+        .provider_type
+        .trim()
+        .eq_ignore_ascii_case("vertex_ai");
+    let transport_supported = match spec.family {
+        _ if transport
+            .provider
+            .provider_type
+            .trim()
+            .eq_ignore_ascii_case("kiro") =>
+        {
+            supports_local_kiro_request_transport_with_network(&transport)
+        }
+        _ if is_antigravity => true,
+        _ if is_claude_code => {
+            supports_local_claude_code_transport_with_network(&transport, spec.api_format)
+        }
+        _ if is_vertex => supports_local_vertex_api_key_gemini_transport_with_network(&transport),
+        LocalSameFormatProviderFamily::Standard => {
+            supports_local_standard_transport_with_network(&transport, spec.api_format)
+        }
+        LocalSameFormatProviderFamily::Gemini => {
+            supports_local_gemini_transport_with_network(&transport, spec.api_format)
+        }
+    };
+    if !transport_supported {
+        mark_skipped_local_same_format_provider_candidate(
+            state,
+            input,
+            trace_id,
+            &candidate,
+            candidate_index,
+            &candidate_id,
+            "transport_unsupported",
+        )
+        .await;
+        return None;
+    }
+
+    let is_kiro = transport
+        .provider
+        .provider_type
+        .trim()
+        .eq_ignore_ascii_case("kiro");
+    let vertex_query_auth = if is_vertex {
+        resolve_local_vertex_api_key_query_auth(&transport)
+    } else {
+        None
+    };
+    let should_try_oauth_auth = is_kiro
+        || matches!(spec.family, LocalSameFormatProviderFamily::Standard)
+            && resolve_local_standard_auth(&transport).is_none()
+        || matches!(spec.family, LocalSameFormatProviderFamily::Gemini)
+            && !is_vertex
+            && resolve_local_gemini_auth(&transport).is_none();
+    let oauth_auth = if should_try_oauth_auth {
+        match state.resolve_local_oauth_request_auth(&transport).await {
+            Ok(Some(LocalResolvedOAuthRequestAuth::Kiro(auth))) => {
+                Some(LocalResolvedOAuthRequestAuth::Kiro(auth))
+            }
+            Ok(Some(LocalResolvedOAuthRequestAuth::Header { name, value })) => {
+                Some(LocalResolvedOAuthRequestAuth::Header { name, value })
+            }
+            Ok(None) => None,
+            Err(err) => {
+                warn!(
+                    trace_id = %trace_id,
+                    api_format = spec.api_format,
+                    provider_type = %transport.provider.provider_type,
+                    error = ?err,
+                    "gateway local same-format oauth auth resolution failed"
+                );
+                None
+            }
+        }
+    } else {
+        None
+    };
+    let kiro_auth = match oauth_auth.as_ref() {
+        Some(LocalResolvedOAuthRequestAuth::Kiro(auth)) => Some(auth),
+        _ => None,
+    };
+    let auth = if let Some(auth) = kiro_auth.as_ref() {
+        Some((auth.name.to_string(), auth.value.clone()))
+    } else if let Some(LocalResolvedOAuthRequestAuth::Header { name, value }) = oauth_auth.as_ref()
+    {
+        Some((name.clone(), value.clone()))
+    } else if is_vertex {
+        None
+    } else {
+        match spec.family {
+            LocalSameFormatProviderFamily::Standard => resolve_local_standard_auth(&transport),
+            LocalSameFormatProviderFamily::Gemini => resolve_local_gemini_auth(&transport),
+        }
+    };
+    let (auth_header, auth_value) = match auth {
+        Some((name, value)) => (Some(name), Some(value)),
+        None if is_vertex && vertex_query_auth.is_some() => (None, None),
+        None => {
+            mark_skipped_local_same_format_provider_candidate(
+                state,
+                input,
+                trace_id,
+                &candidate,
+                candidate_index,
+                &candidate_id,
+                "transport_auth_unavailable",
+            )
+            .await;
+            return None;
+        }
+    };
+    if is_vertex && vertex_query_auth.is_none() {
+        mark_skipped_local_same_format_provider_candidate(
+            state,
+            input,
+            trace_id,
+            &candidate,
+            candidate_index,
+            &candidate_id,
+            "transport_auth_unavailable",
+        )
+        .await;
+        return None;
+    }
+    let mapped_model = candidate.selected_provider_model_name.trim().to_string();
+    if mapped_model.is_empty() {
+        mark_skipped_local_same_format_provider_candidate(
+            state,
+            input,
+            trace_id,
+            &candidate,
+            candidate_index,
+            &candidate_id,
+            "mapped_model_missing",
+        )
+        .await;
+        return None;
+    }
+
+    let Some(base_provider_request_body) =
+        super::super::request::build_same_format_provider_request_body(
+            body_json,
+            &mapped_model,
+            spec,
+            transport.endpoint.body_rules.as_ref(),
+            is_kiro || is_antigravity || spec.require_streaming,
+            kiro_auth,
+            is_claude_code,
+        )
+    else {
+        mark_skipped_local_same_format_provider_candidate(
+            state,
+            input,
+            trace_id,
+            &candidate,
+            candidate_index,
+            &candidate_id,
+            "provider_request_body_missing",
+        )
+        .await;
+        return None;
+    };
+
+    let antigravity_auth = if is_antigravity {
+        match classify_local_antigravity_request_support(
+            &transport,
+            &base_provider_request_body,
+            AntigravityEnvelopeRequestType::Agent,
+        ) {
+            AntigravityRequestSideSupport::Supported(spec) => Some(spec.auth),
+            AntigravityRequestSideSupport::Unsupported(_) => {
+                mark_skipped_local_same_format_provider_candidate(
+                    state,
+                    input,
+                    trace_id,
+                    &candidate,
+                    candidate_index,
+                    &candidate_id,
+                    "transport_unsupported",
+                )
+                .await;
+                return None;
+            }
+        }
+    } else {
+        None
+    };
+    let provider_request_body = if let Some(antigravity_auth) = antigravity_auth.as_ref() {
+        match build_antigravity_safe_v1internal_request(
+            antigravity_auth,
+            trace_id,
+            &mapped_model,
+            &base_provider_request_body,
+            AntigravityEnvelopeRequestType::Agent,
+        ) {
+            AntigravityRequestEnvelopeSupport::Supported(envelope) => envelope,
+            AntigravityRequestEnvelopeSupport::Unsupported(_) => {
+                mark_skipped_local_same_format_provider_candidate(
+                    state,
+                    input,
+                    trace_id,
+                    &candidate,
+                    candidate_index,
+                    &candidate_id,
+                    "provider_request_body_missing",
+                )
+                .await;
+                return None;
+            }
+        }
+    } else {
+        base_provider_request_body
+    };
+    let upstream_is_stream = is_kiro || is_antigravity || spec.require_streaming;
+    let report_kind = if is_kiro && !spec.require_streaming {
+        "claude_cli_sync_finalize"
+    } else if is_antigravity && !spec.require_streaming {
+        match spec.api_format {
+            "gemini:chat" => "gemini_chat_sync_finalize",
+            "gemini:cli" => "gemini_cli_sync_finalize",
+            _ => spec.report_kind,
+        }
+    } else {
+        spec.report_kind
+    };
+
+    let Some(upstream_url) = super::super::request::build_same_format_upstream_url(
+        parts,
+        &transport,
+        &mapped_model,
+        spec,
+        upstream_is_stream,
+        kiro_auth,
+    ) else {
+        mark_skipped_local_same_format_provider_candidate(
+            state,
+            input,
+            trace_id,
+            &candidate,
+            candidate_index,
+            &candidate_id,
+            "upstream_url_missing",
+        )
+        .await;
+        return None;
+    };
+
+    let Some(provider_request_headers) = (if let Some(kiro_auth) = kiro_auth.as_ref() {
+        crate::gateway::provider_transport::build_kiro_provider_headers(
+            &parts.headers,
+            &provider_request_body,
+            body_json,
+            transport.endpoint.header_rules.as_ref(),
+            auth_header.as_deref().unwrap_or_default(),
+            auth_value.as_deref().unwrap_or_default(),
+            &kiro_auth.auth_config,
+            kiro_auth.machine_id.as_str(),
+        )
+    } else {
+        let extra_headers = antigravity_auth
+            .as_ref()
+            .map(build_antigravity_static_identity_headers)
+            .unwrap_or_default();
+        let mut provider_request_headers = if is_claude_code {
+            build_claude_code_passthrough_headers(
+                &parts.headers,
+                auth_header.as_deref().unwrap_or_default(),
+                auth_value.as_deref().unwrap_or_default(),
+                &extra_headers,
+                upstream_is_stream,
+                transport.key.fingerprint.as_ref(),
+            )
+        } else if is_vertex {
+            build_passthrough_headers(&parts.headers, &extra_headers, Some("application/json"))
+        } else {
+            build_openai_passthrough_headers(
+                &parts.headers,
+                auth_header.as_deref().unwrap_or_default(),
+                auth_value.as_deref().unwrap_or_default(),
+                &extra_headers,
+                Some("application/json"),
+            )
+        };
+        let protected_headers = auth_header
+            .as_deref()
+            .filter(|value| !value.trim().is_empty())
+            .map(|value| vec![value, "content-type"])
+            .unwrap_or_else(|| vec!["content-type"]);
+        if !apply_local_header_rules(
+            &mut provider_request_headers,
+            transport.endpoint.header_rules.as_ref(),
+            &protected_headers,
+            &provider_request_body,
+            Some(body_json),
+        ) {
+            None
+        } else {
+            if let (Some(auth_header), Some(auth_value)) =
+                (auth_header.as_deref(), auth_value.as_deref())
+            {
+                ensure_upstream_auth_header(&mut provider_request_headers, auth_header, auth_value);
+            }
+            if upstream_is_stream {
+                provider_request_headers
+                    .insert("accept".to_string(), "text/event-stream".to_string());
+            }
+            Some(provider_request_headers)
+        }
+    }) else {
+        mark_skipped_local_same_format_provider_candidate(
+            state,
+            input,
+            trace_id,
+            &candidate,
+            candidate_index,
+            &candidate_id,
+            "transport_header_rules_apply_failed",
+        )
+        .await;
+        return None;
+    };
+    let prompt_cache_key = provider_request_body
+        .get("prompt_cache_key")
+        .and_then(|value| value.as_str())
+        .map(str::trim)
+        .filter(|value| !value.is_empty())
+        .map(ToOwned::to_owned);
+    let proxy = resolve_transport_proxy_snapshot_with_tunnel_affinity(state, &transport).await;
+    let tls_profile = resolve_transport_tls_profile(&transport);
+    let report_context = append_execution_contract_fields_to_value(
+        json!({
+            "user_id": input.auth_context.user_id,
+            "api_key_id": input.auth_context.api_key_id,
+            "request_id": trace_id,
+            "candidate_id": candidate_id,
+            "candidate_index": candidate_index,
+            "retry_index": 0,
+            "model": input.requested_model,
+            "provider_name": transport.provider.name,
+            "provider_id": candidate.provider_id,
+            "endpoint_id": candidate.endpoint_id,
+            "key_id": candidate.key_id,
+            "provider_api_format": spec.api_format,
+            "client_api_format": spec.api_format,
+            "mapped_model": mapped_model,
+            "upstream_url": upstream_url,
+            "provider_request_method": serde_json::Value::Null,
+            "provider_request_headers": provider_request_headers,
+            "provider_request_body": provider_request_body,
+            "original_headers": collect_control_headers(&parts.headers),
+            "original_request_body": body_json,
+            "has_envelope": is_kiro || is_antigravity,
+            "envelope_name": if is_kiro {
+                Some(KIRO_ENVELOPE_NAME)
+            } else if is_antigravity {
+                Some(super::super::ANTIGRAVITY_ENVELOPE_NAME)
+            } else {
+                None
+            },
+            "needs_conversion": false,
+        }),
+        ExecutionStrategy::LocalSameFormat,
+        ConversionMode::None,
+        spec.api_format,
+        spec.api_format,
+    );
+
+    Some(GatewayControlSyncDecisionResponse {
+        action: if spec.require_streaming {
+            EXECUTION_RUNTIME_STREAM_DECISION_ACTION.to_string()
+        } else {
+            EXECUTION_RUNTIME_SYNC_DECISION_ACTION.to_string()
+        },
+        decision_kind: Some(spec.decision_kind.to_string()),
+        execution_strategy: Some(ExecutionStrategy::LocalSameFormat.as_str().to_string()),
+        conversion_mode: Some(ConversionMode::None.as_str().to_string()),
+        request_id: Some(trace_id.to_string()),
+        candidate_id: Some(candidate_id.clone()),
+        provider_name: Some(transport.provider.name.clone()),
+        provider_id: Some(candidate.provider_id.clone()),
+        endpoint_id: Some(candidate.endpoint_id.clone()),
+        key_id: Some(candidate.key_id.clone()),
+        upstream_base_url: Some(transport.endpoint.base_url.clone()),
+        upstream_url: Some(upstream_url.clone()),
+        provider_request_method: None,
+        auth_header,
+        auth_value,
+        provider_api_format: Some(spec.api_format.to_string()),
+        client_api_format: Some(spec.api_format.to_string()),
+        provider_contract: Some(spec.api_format.to_string()),
+        client_contract: Some(spec.api_format.to_string()),
+        model_name: Some(input.requested_model.clone()),
+        mapped_model: Some(mapped_model.clone()),
+        prompt_cache_key,
+        extra_headers: BTreeMap::new(),
+        provider_request_headers: provider_request_headers.clone(),
+        provider_request_body: Some(provider_request_body.clone()),
+        provider_request_body_base64: None,
+        content_type: Some("application/json".to_string()),
+        proxy,
+        tls_profile,
+        timeouts: resolve_transport_execution_timeouts(&transport),
+        upstream_is_stream,
+        report_kind: Some(report_kind.to_string()),
+        report_context: Some(report_context),
+        auth_context: Some(input.auth_context.clone()),
+    })
+}
+
+pub(super) async fn mark_skipped_local_same_format_provider_candidate(
+    state: &AppState,
+    input: &LocalSameFormatProviderDecisionInput,
+    trace_id: &str,
+    candidate: &GatewayMinimalCandidateSelectionCandidate,
+    candidate_index: u32,
+    candidate_id: &str,
+    skip_reason: &'static str,
+) {
+    if let Err(err) = state
+        .upsert_request_candidate(UpsertRequestCandidateRecord {
+            id: candidate_id.to_string(),
+            request_id: trace_id.to_string(),
+            user_id: Some(input.auth_context.user_id.clone()),
+            api_key_id: Some(input.auth_context.api_key_id.clone()),
+            username: None,
+            api_key_name: None,
+            candidate_index,
+            retry_index: 0,
+            provider_id: Some(candidate.provider_id.clone()),
+            endpoint_id: Some(candidate.endpoint_id.clone()),
+            key_id: Some(candidate.key_id.clone()),
+            status: RequestCandidateStatus::Skipped,
+            skip_reason: Some(skip_reason.to_string()),
+            is_cached: Some(false),
+            status_code: None,
+            error_type: None,
+            error_message: None,
+            latency_ms: None,
+            concurrent_requests: None,
+            extra_data: None,
+            required_capabilities: candidate.key_capabilities.clone(),
+            created_at_unix_secs: None,
+            started_at_unix_secs: None,
+            finished_at_unix_secs: Some(current_unix_secs()),
+        })
+        .await
+    {
+        warn!(
+            trace_id = %trace_id,
+            candidate_id = %candidate_id,
+            skip_reason,
+            error = ?err,
+            "gateway local same-format decision failed to persist skipped candidate"
+        );
+    }
+}
--- a/apps/aether-gateway/src/ai_pipeline/planner/passthrough/provider/family/types.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/passthrough/provider/family/types.rs
@@ -0,0 +1,28 @@
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub(crate) enum LocalSameFormatProviderFamily {
+    Standard,
+    Gemini,
+}
+
+#[derive(Debug, Clone, Copy)]
+pub(crate) struct LocalSameFormatProviderSpec {
+    pub(crate) api_format: &'static str,
+    pub(crate) decision_kind: &'static str,
+    pub(crate) report_kind: &'static str,
+    pub(crate) family: LocalSameFormatProviderFamily,
+    pub(crate) require_streaming: bool,
+}
+
+#[derive(Debug, Clone)]
+pub(crate) struct LocalSameFormatProviderDecisionInput {
+    pub(crate) auth_context: crate::gateway::GatewayControlAuthContext,
+    pub(crate) requested_model: String,
+    pub(crate) auth_snapshot: crate::gateway::gateway_data::StoredGatewayAuthApiKeySnapshot,
+}
+
+#[derive(Debug, Clone)]
+pub(crate) struct LocalSameFormatProviderCandidateAttempt {
+    pub(crate) candidate: crate::gateway::scheduler::GatewayMinimalCandidateSelectionCandidate,
+    pub(crate) candidate_index: u32,
+    pub(crate) candidate_id: String,
+}
--- a/apps/aether-gateway/src/ai_pipeline/planner/passthrough/provider/plans.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/passthrough/provider/plans.rs
@@ -0,0 +1,198 @@
+use tracing::warn;
+
+use super::{
+    materialize_local_same_format_provider_candidate_attempts,
+    maybe_build_local_same_format_provider_decision_payload_for_candidate,
+    resolve_local_same_format_provider_decision_input, AppState, GatewayControlDecision,
+    GatewayError, LocalSameFormatProviderFamily, LocalSameFormatProviderSpec,
+    LocalStreamPlanAndReport, LocalSyncPlanAndReport,
+};
+use crate::gateway::ai_pipeline::planner::plan_builders::{
+    build_gemini_stream_plan_from_decision, build_gemini_sync_plan_from_decision,
+    build_standard_stream_plan_from_decision, build_standard_sync_plan_from_decision,
+};
+use crate::gateway::ai_pipeline::planner::{
+    CLAUDE_CHAT_STREAM_PLAN_KIND, CLAUDE_CHAT_SYNC_PLAN_KIND, CLAUDE_CLI_STREAM_PLAN_KIND,
+    CLAUDE_CLI_SYNC_PLAN_KIND, GEMINI_CHAT_STREAM_PLAN_KIND, GEMINI_CHAT_SYNC_PLAN_KIND,
+    GEMINI_CLI_STREAM_PLAN_KIND, GEMINI_CLI_SYNC_PLAN_KIND,
+};
+
+pub(super) fn resolve_sync_spec(plan_kind: &str) -> Option<LocalSameFormatProviderSpec> {
+    match plan_kind {
+        CLAUDE_CHAT_SYNC_PLAN_KIND => Some(LocalSameFormatProviderSpec {
+            api_format: "claude:chat",
+            decision_kind: CLAUDE_CHAT_SYNC_PLAN_KIND,
+            report_kind: "claude_chat_sync_success",
+            family: LocalSameFormatProviderFamily::Standard,
+            require_streaming: false,
+        }),
+        CLAUDE_CLI_SYNC_PLAN_KIND => Some(LocalSameFormatProviderSpec {
+            api_format: "claude:cli",
+            decision_kind: CLAUDE_CLI_SYNC_PLAN_KIND,
+            report_kind: "claude_cli_sync_success",
+            family: LocalSameFormatProviderFamily::Standard,
+            require_streaming: false,
+        }),
+        GEMINI_CHAT_SYNC_PLAN_KIND => Some(LocalSameFormatProviderSpec {
+            api_format: "gemini:chat",
+            decision_kind: GEMINI_CHAT_SYNC_PLAN_KIND,
+            report_kind: "gemini_chat_sync_success",
+            family: LocalSameFormatProviderFamily::Gemini,
+            require_streaming: false,
+        }),
+        GEMINI_CLI_SYNC_PLAN_KIND => Some(LocalSameFormatProviderSpec {
+            api_format: "gemini:cli",
+            decision_kind: GEMINI_CLI_SYNC_PLAN_KIND,
+            report_kind: "gemini_cli_sync_success",
+            family: LocalSameFormatProviderFamily::Gemini,
+            require_streaming: false,
+        }),
+        _ => None,
+    }
+}
+
+pub(super) fn resolve_stream_spec(plan_kind: &str) -> Option<LocalSameFormatProviderSpec> {
+    match plan_kind {
+        CLAUDE_CHAT_STREAM_PLAN_KIND => Some(LocalSameFormatProviderSpec {
+            api_format: "claude:chat",
+            decision_kind: CLAUDE_CHAT_STREAM_PLAN_KIND,
+            report_kind: "claude_chat_stream_success",
+            family: LocalSameFormatProviderFamily::Standard,
+            require_streaming: true,
+        }),
+        CLAUDE_CLI_STREAM_PLAN_KIND => Some(LocalSameFormatProviderSpec {
+            api_format: "claude:cli",
+            decision_kind: CLAUDE_CLI_STREAM_PLAN_KIND,
+            report_kind: "claude_cli_stream_success",
+            family: LocalSameFormatProviderFamily::Standard,
+            require_streaming: true,
+        }),
+        GEMINI_CHAT_STREAM_PLAN_KIND => Some(LocalSameFormatProviderSpec {
+            api_format: "gemini:chat",
+            decision_kind: GEMINI_CHAT_STREAM_PLAN_KIND,
+            report_kind: "gemini_chat_stream_success",
+            family: LocalSameFormatProviderFamily::Gemini,
+            require_streaming: true,
+        }),
+        GEMINI_CLI_STREAM_PLAN_KIND => Some(LocalSameFormatProviderSpec {
+            api_format: "gemini:cli",
+            decision_kind: GEMINI_CLI_STREAM_PLAN_KIND,
+            report_kind: "gemini_cli_stream_success",
+            family: LocalSameFormatProviderFamily::Gemini,
+            require_streaming: true,
+        }),
+        _ => None,
+    }
+}
+
+pub(super) async fn build_local_sync_plan_and_reports(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    body_json: &serde_json::Value,
+    spec: LocalSameFormatProviderSpec,
+) -> Result<Vec<LocalSyncPlanAndReport>, GatewayError> {
+    let Some(input) = resolve_local_same_format_provider_decision_input(
+        state, parts, trace_id, decision, body_json, spec,
+    )
+    .await
+    else {
+        return Ok(Vec::new());
+    };
+
+    let attempts =
+        materialize_local_same_format_provider_candidate_attempts(state, trace_id, &input, spec)
+            .await?;
+
+    let mut plans = Vec::new();
+    for attempt in attempts {
+        let Some(payload) = maybe_build_local_same_format_provider_decision_payload_for_candidate(
+            state, parts, trace_id, body_json, &input, attempt, spec,
+        )
+        .await
+        else {
+            continue;
+        };
+
+        let built = match spec.family {
+            LocalSameFormatProviderFamily::Standard => {
+                build_standard_sync_plan_from_decision(parts, body_json, payload)
+            }
+            LocalSameFormatProviderFamily::Gemini => {
+                build_gemini_sync_plan_from_decision(parts, body_json, payload)
+            }
+        };
+
+        match built {
+            Ok(Some(value)) => plans.push(value),
+            Ok(None) => {}
+            Err(err) => {
+                warn!(
+                    trace_id = %trace_id,
+                    api_format = spec.api_format,
+                    error = ?err,
+                    "gateway local same-format sync decision plan build failed"
+                );
+            }
+        }
+    }
+
+    Ok(plans)
+}
+
+pub(super) async fn build_local_stream_plan_and_reports(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    body_json: &serde_json::Value,
+    spec: LocalSameFormatProviderSpec,
+) -> Result<Vec<LocalStreamPlanAndReport>, GatewayError> {
+    let Some(input) = resolve_local_same_format_provider_decision_input(
+        state, parts, trace_id, decision, body_json, spec,
+    )
+    .await
+    else {
+        return Ok(Vec::new());
+    };
+
+    let attempts =
+        materialize_local_same_format_provider_candidate_attempts(state, trace_id, &input, spec)
+            .await?;
+
+    let mut plans = Vec::new();
+    for attempt in attempts {
+        let Some(payload) = maybe_build_local_same_format_provider_decision_payload_for_candidate(
+            state, parts, trace_id, body_json, &input, attempt, spec,
+        )
+        .await
+        else {
+            continue;
+        };
+
+        let built = match spec.family {
+            LocalSameFormatProviderFamily::Standard => {
+                build_standard_stream_plan_from_decision(parts, body_json, payload, false)
+            }
+            LocalSameFormatProviderFamily::Gemini => {
+                build_gemini_stream_plan_from_decision(parts, body_json, payload)
+            }
+        };
+
+        match built {
+            Ok(Some(value)) => plans.push(value),
+            Ok(None) => {}
+            Err(err) => {
+                warn!(
+                    trace_id = %trace_id,
+                    api_format = spec.api_format,
+                    error = ?err,
+                    "gateway local same-format stream decision plan build failed"
+                );
+            }
+        }
+    }
+
+    Ok(plans)
+}
--- a/apps/aether-gateway/src/ai_pipeline/planner/passthrough/provider/request.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/passthrough/provider/request.rs
@@ -0,0 +1,186 @@
+use super::*;
+
+pub(super) fn build_same_format_provider_request_body(
+    body_json: &Value,
+    mapped_model: &str,
+    spec: LocalSameFormatProviderSpec,
+    body_rules: Option<&Value>,
+    upstream_is_stream: bool,
+    kiro_auth: Option<&crate::gateway::provider_transport::KiroRequestAuth>,
+    is_claude_code: bool,
+) -> Option<Value> {
+    if let Some(kiro_auth) = kiro_auth {
+        return build_kiro_provider_request_body(
+            body_json,
+            mapped_model,
+            &kiro_auth.auth_config,
+            body_rules,
+        );
+    }
+
+    let request_body_object = body_json.as_object()?;
+    let mut provider_request_body = serde_json::Map::from_iter(
+        request_body_object
+            .iter()
+            .map(|(key, value)| (key.clone(), value.clone())),
+    );
+    match spec.family {
+        LocalSameFormatProviderFamily::Standard => {
+            provider_request_body
+                .insert("model".to_string(), Value::String(mapped_model.to_string()));
+            if upstream_is_stream {
+                provider_request_body.insert("stream".to_string(), Value::Bool(true));
+            }
+        }
+        LocalSameFormatProviderFamily::Gemini => {
+            provider_request_body.remove("model");
+        }
+    }
+    let mut provider_request_body = Value::Object(provider_request_body);
+    if is_claude_code {
+        sanitize_claude_code_request_body(&mut provider_request_body);
+    }
+    if !apply_local_body_rules(&mut provider_request_body, body_rules, Some(body_json)) {
+        return None;
+    }
+    Some(provider_request_body)
+}
+
+pub(super) fn build_same_format_upstream_url(
+    parts: &http::request::Parts,
+    transport: &crate::gateway::provider_transport::GatewayProviderTransportSnapshot,
+    mapped_model: &str,
+    spec: LocalSameFormatProviderSpec,
+    upstream_is_stream: bool,
+    kiro_auth: Option<&crate::gateway::provider_transport::KiroRequestAuth>,
+) -> Option<String> {
+    if let Some(kiro_auth) = kiro_auth {
+        return build_kiro_generate_assistant_response_url(
+            &transport.endpoint.base_url,
+            parts.uri.query(),
+            Some(kiro_auth.auth_config.effective_api_region()),
+        );
+    }
+    if transport
+        .provider
+        .provider_type
+        .trim()
+        .eq_ignore_ascii_case("claude_code")
+    {
+        return Some(build_claude_code_messages_url(
+            &transport.endpoint.base_url,
+            parts.uri.query(),
+        ));
+    }
+    if transport
+        .provider
+        .provider_type
+        .trim()
+        .eq_ignore_ascii_case("vertex_ai")
+    {
+        let auth = resolve_local_vertex_api_key_query_auth(transport)?;
+        return build_vertex_api_key_gemini_content_url(
+            mapped_model,
+            upstream_is_stream,
+            &auth.value,
+            parts.uri.query(),
+        );
+    }
+    if transport
+        .provider
+        .provider_type
+        .trim()
+        .eq_ignore_ascii_case("antigravity")
+    {
+        let query = parts.uri.query().map(|query| {
+            form_urlencoded::parse(query.as_bytes())
+                .into_owned()
+                .collect::<BTreeMap<String, String>>()
+        });
+        return build_antigravity_v1internal_url(
+            &transport.endpoint.base_url,
+            if upstream_is_stream {
+                AntigravityRequestUrlAction::StreamGenerateContent
+            } else {
+                AntigravityRequestUrlAction::GenerateContent
+            },
+            query.as_ref(),
+        );
+    }
+
+    let custom_path = transport
+        .endpoint
+        .custom_path
+        .as_deref()
+        .map(str::trim)
+        .filter(|value| !value.is_empty());
+
+    if let Some(path) = custom_path {
+        let blocked_keys = match spec.family {
+            LocalSameFormatProviderFamily::Standard => &[][..],
+            LocalSameFormatProviderFamily::Gemini => &["key"][..],
+        };
+        let url = build_passthrough_path_url(
+            &transport.endpoint.base_url,
+            path,
+            parts.uri.query(),
+            blocked_keys,
+        )?;
+        return Some(maybe_add_gemini_stream_alt_sse(url, spec));
+    }
+
+    let url = match spec.family {
+        LocalSameFormatProviderFamily::Standard => Some(build_claude_messages_url(
+            &transport.endpoint.base_url,
+            parts.uri.query(),
+        )),
+        LocalSameFormatProviderFamily::Gemini => build_gemini_content_url(
+            &transport.endpoint.base_url,
+            mapped_model,
+            spec.require_streaming,
+            parts.uri.query(),
+        ),
+    }?;
+
+    Some(maybe_add_gemini_stream_alt_sse(url, spec))
+}
+
+pub(super) fn extract_gemini_model_from_path(path: &str) -> Option<String> {
+    let (_, suffix) = path.split_once("/models/")?;
+    let model = suffix
+        .split_once(':')
+        .map(|(value, _)| value)
+        .unwrap_or(suffix);
+    let model = model.trim();
+    if model.is_empty() {
+        None
+    } else {
+        Some(model.to_string())
+    }
+}
+
+fn maybe_add_gemini_stream_alt_sse(
+    upstream_url: String,
+    spec: LocalSameFormatProviderSpec,
+) -> String {
+    if spec.family != LocalSameFormatProviderFamily::Gemini || !spec.require_streaming {
+        return upstream_url;
+    }
+
+    let has_alt = upstream_url
+        .split_once('?')
+        .map(|(_, query)| {
+            form_urlencoded::parse(query.as_bytes())
+                .any(|(key, _)| key.as_ref().eq_ignore_ascii_case("alt"))
+        })
+        .unwrap_or(false);
+    if has_alt {
+        return upstream_url;
+    }
+
+    if upstream_url.contains('?') {
+        format!("{upstream_url}&alt=sse")
+    } else {
+        format!("{upstream_url}?alt=sse")
+    }
+}
--- a/apps/aether-gateway/src/ai_pipeline/planner/plan_builders.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/plan_builders.rs
@@ -0,0 +1,65 @@
+use std::collections::BTreeMap;
+
+use aether_contracts::ExecutionPlan;
+
+pub(crate) use crate::gateway::ai_pipeline::planner::generic_decision_missing_exact_provider_request;
+use crate::gateway::{GatewayControlSyncDecisionResponse, GatewayError};
+
+pub(crate) struct LocalSyncPlanAndReport {
+    pub(crate) plan: ExecutionPlan,
+    pub(crate) report_kind: Option<String>,
+    pub(crate) report_context: Option<serde_json::Value>,
+}
+
+pub(crate) struct LocalStreamPlanAndReport {
+    pub(crate) plan: ExecutionPlan,
+    pub(crate) report_kind: Option<String>,
+    pub(crate) report_context: Option<serde_json::Value>,
+}
+
+#[path = "standard/gemini/plan_builders.rs"]
+mod gemini_builders;
+#[path = "standard/openai/plan_builders.rs"]
+mod openai_builders;
+#[path = "passthrough/plan_builders.rs"]
+mod passthrough_builders;
+#[path = "standard/plan_builders.rs"]
+mod standard_builders;
+
+pub(crate) use gemini_builders::{
+    build_gemini_stream_plan_from_decision, build_gemini_sync_plan_from_decision,
+};
+pub(crate) use openai_builders::{
+    build_openai_chat_stream_plan_from_decision, build_openai_chat_sync_plan_from_decision,
+    build_openai_cli_stream_plan_from_decision, build_openai_cli_sync_plan_from_decision,
+};
+pub(crate) use passthrough_builders::{
+    build_passthrough_stream_plan_from_decision, build_passthrough_sync_plan_from_decision,
+};
+pub(crate) use standard_builders::{
+    build_standard_stream_plan_from_decision, build_standard_sync_plan_from_decision,
+};
+
+pub(super) fn augment_sync_report_context(
+    report_context: Option<serde_json::Value>,
+    provider_request_headers: &BTreeMap<String, String>,
+    provider_request_body: &serde_json::Value,
+) -> Result<Option<serde_json::Value>, GatewayError> {
+    let mut report_context = match report_context {
+        Some(serde_json::Value::Object(map)) => map,
+        Some(_) => serde_json::Map::new(),
+        None => serde_json::Map::new(),
+    };
+
+    report_context.insert(
+        "provider_request_headers".to_string(),
+        serde_json::to_value(provider_request_headers)
+            .map_err(|err| GatewayError::Internal(err.to_string()))?,
+    );
+    report_context.insert(
+        "provider_request_body".to_string(),
+        provider_request_body.clone(),
+    );
+
+    Ok(Some(serde_json::Value::Object(report_context)))
+}
--- a/apps/aether-gateway/src/ai_pipeline/planner/specialized/files.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/specialized/files.rs
@@ -7,32 +7,33 @@ use serde_json::json;
 use tracing::warn;
 use uuid::Uuid;

-use crate::gateway::executor::request_candidates::{
-    current_unix_secs, record_local_request_candidate_status,
-};
 use crate::gateway::headers::collect_control_headers;
 use crate::gateway::provider_transport::{
    apply_local_body_rules, apply_local_header_rules, build_gemini_files_passthrough_url,
    build_passthrough_headers_with_auth, resolve_local_gemini_auth,
-    resolve_transport_execution_timeouts, resolve_transport_proxy_snapshot,
+    resolve_transport_execution_timeouts, resolve_transport_proxy_snapshot_with_tunnel_affinity,
    resolve_transport_tls_profile, supports_local_gemini_transport_with_network,
 };
+use crate::gateway::request_candidates::{
+    current_unix_secs, record_local_request_candidate_status,
+};
+use crate::gateway::ai_pipeline::planner::plan_builders::{
+    build_passthrough_stream_plan_from_decision, build_passthrough_sync_plan_from_decision,
+    LocalStreamPlanAndReport, LocalSyncPlanAndReport,
+};
+use crate::gateway::ai_pipeline::planner::prefer_local_tunnel_owner_candidates;
 use crate::gateway::scheduler::{
    list_selectable_candidates_for_required_capability_without_requested_model,
    GatewayMinimalCandidateSelectionCandidate,
 };
-use crate::gateway::{AppState, GatewayControlDecision, GatewayError};
-
-use super::super::plan_builders::{
-    build_passthrough_stream_plan_from_decision, build_passthrough_sync_plan_from_decision,
-    LocalStreamPlanAndReport, LocalSyncPlanAndReport,
+use crate::gateway::{
+    execute_execution_runtime_stream, execute_execution_runtime_sync, AppState,
+    GatewayControlDecision, GatewayControlSyncDecisionResponse, GatewayError,
 };
-use super::super::stream::execute_executor_stream;
-use super::super::sync::execute_executor_sync;
-use super::super::{
-    GatewayControlSyncDecisionResponse, EXECUTOR_STREAM_DECISION_ACTION,
-    EXECUTOR_SYNC_DECISION_ACTION, GEMINI_FILES_DELETE_PLAN_KIND, GEMINI_FILES_DOWNLOAD_PLAN_KIND,
-    GEMINI_FILES_GET_PLAN_KIND, GEMINI_FILES_LIST_PLAN_KIND, GEMINI_FILES_UPLOAD_PLAN_KIND,
+use crate::gateway::ai_pipeline::planner::{
+    EXECUTION_RUNTIME_STREAM_DECISION_ACTION, EXECUTION_RUNTIME_SYNC_DECISION_ACTION,
+    GEMINI_FILES_DELETE_PLAN_KIND, GEMINI_FILES_DOWNLOAD_PLAN_KIND, GEMINI_FILES_GET_PLAN_KIND,
+    GEMINI_FILES_LIST_PLAN_KIND, GEMINI_FILES_UPLOAD_PLAN_KIND,
 };

 const GEMINI_FILES_CANDIDATE_API_FORMAT: &str = "gemini:chat";
@@ -49,7 +50,7 @@ struct LocalGeminiFilesSpec {
 #[derive(Debug, Clone)]
 struct LocalGeminiFilesDecisionInput {
    auth_context: crate::gateway::GatewayControlAuthContext,
-    auth_snapshot: crate::gateway::data::StoredGatewayAuthApiKeySnapshot,
+    auth_snapshot: crate::gateway::gateway_data::StoredGatewayAuthApiKeySnapshot,
 }

 #[derive(Debug, Clone)]
@@ -59,10 +60,8 @@ struct LocalGeminiFilesCandidateAttempt {
    candidate_id: String,
 }

-pub(in crate::gateway::executor) async fn maybe_execute_sync_via_local_gemini_files_decision(
+pub(crate) async fn maybe_execute_sync_via_local_gemini_files_decision(
    state: &AppState,
-    control_base_url: &str,
-    executor_base_url: &str,
    parts: &http::request::Parts,
    body_json: &serde_json::Value,
    body_base64: Option<&str>,
@@ -92,10 +91,8 @@ pub(in crate::gateway::executor) async fn maybe_execute_sync_via_local_gemini_fi

    let mut remaining = plan_and_reports.into_iter();
    while let Some(plan_and_report) = remaining.next() {
-        if let Some(response) = execute_executor_sync(
+        if let Some(response) = execute_execution_runtime_sync(
            state,
-            control_base_url,
-            executor_base_url,
            parts.uri.path(),
            plan_and_report.plan,
            trace_id,
@@ -114,10 +111,8 @@ pub(in crate::gateway::executor) async fn maybe_execute_sync_via_local_gemini_fi
    Ok(None)
 }

-pub(in crate::gateway::executor) async fn maybe_execute_stream_via_local_gemini_files_decision(
+pub(crate) async fn maybe_execute_stream_via_local_gemini_files_decision(
    state: &AppState,
-    control_base_url: &str,
-    executor_base_url: &str,
    parts: &http::request::Parts,
    trace_id: &str,
    decision: &GatewayControlDecision,
@@ -135,10 +130,8 @@ pub(in crate::gateway::executor) async fn maybe_execute_stream_via_local_gemini_

    let mut remaining = plan_and_reports.into_iter();
    while let Some(plan_and_report) = remaining.next() {
-        if let Some(response) = execute_executor_stream(
+        if let Some(response) = execute_execution_runtime_stream(
            state,
-            control_base_url,
-            executor_base_url,
            plan_and_report.plan,
            trace_id,
            decision,
@@ -156,7 +149,7 @@ pub(in crate::gateway::executor) async fn maybe_execute_stream_via_local_gemini_
    Ok(None)
 }

-pub(in crate::gateway::executor) async fn maybe_build_sync_local_gemini_files_decision_payload(
+pub(crate) async fn maybe_build_sync_local_gemini_files_decision_payload(
    state: &AppState,
    parts: &http::request::Parts,
    body_json: &serde_json::Value,
@@ -199,7 +192,7 @@ pub(in crate::gateway::executor) async fn maybe_build_sync_local_gemini_files_de
    Ok(None)
 }

-pub(in crate::gateway::executor) async fn maybe_build_stream_local_gemini_files_decision_payload(
+pub(crate) async fn maybe_build_stream_local_gemini_files_decision_payload(
    state: &AppState,
    parts: &http::request::Parts,
    trace_id: &str,
@@ -432,6 +425,7 @@ async fn materialize_local_gemini_files_candidate_attempts(
        current_unix_secs(),
    )
    .await?;
+    let candidates = prefer_local_tunnel_owner_candidates(state, candidates).await;

    let created_at_unix_secs = current_unix_secs();
    let mut attempts = Vec::with_capacity(candidates.len());
@@ -712,16 +706,22 @@ async fn maybe_build_local_gemini_files_decision_payload_for_candidate(
        .trim_start_matches("/v1beta/")
        .trim()
        .to_string();
-    let proxy = resolve_transport_proxy_snapshot(&transport);
+    let proxy = resolve_transport_proxy_snapshot_with_tunnel_affinity(state, &transport).await;
    let tls_profile = resolve_transport_tls_profile(&transport);

    Some(GatewayControlSyncDecisionResponse {
        action: if spec.require_streaming {
-            EXECUTOR_STREAM_DECISION_ACTION.to_string()
+            EXECUTION_RUNTIME_STREAM_DECISION_ACTION.to_string()
        } else {
-            EXECUTOR_SYNC_DECISION_ACTION.to_string()
+            EXECUTION_RUNTIME_SYNC_DECISION_ACTION.to_string()
        },
        decision_kind: Some(spec.decision_kind.to_string()),
+        execution_strategy: Some(
+            crate::gateway::ExecutionStrategy::LocalSameFormat
+                .as_str()
+                .to_string(),
+        ),
+        conversion_mode: Some(crate::gateway::ConversionMode::None.as_str().to_string()),
        request_id: Some(trace_id.to_string()),
        candidate_id: Some(candidate_id.clone()),
        provider_name: Some(transport.provider.name.clone()),
@@ -735,6 +735,8 @@ async fn maybe_build_local_gemini_files_decision_payload_for_candidate(
        auth_value: Some(auth_value),
        provider_api_format: Some(GEMINI_FILES_CLIENT_API_FORMAT.to_string()),
        client_api_format: Some(GEMINI_FILES_CLIENT_API_FORMAT.to_string()),
+        provider_contract: Some(GEMINI_FILES_CLIENT_API_FORMAT.to_string()),
+        client_contract: Some(GEMINI_FILES_CLIENT_API_FORMAT.to_string()),
        model_name: Some("gemini-files".to_string()),
        mapped_model: Some(candidate.selected_provider_model_name.clone()),
        prompt_cache_key: None,
--- a/apps/aether-gateway/src/ai_pipeline/planner/specialized/mod.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/specialized/mod.rs
@@ -0,0 +1,14 @@
+//! Non-matrix AI surfaces such as files and video.
+
+mod files;
+mod video;
+
+pub(crate) use self::files::{
+    maybe_build_stream_local_gemini_files_decision_payload,
+    maybe_build_sync_local_gemini_files_decision_payload,
+    maybe_execute_stream_via_local_gemini_files_decision,
+    maybe_execute_sync_via_local_gemini_files_decision,
+};
+pub(crate) use self::video::{
+    maybe_build_sync_local_video_decision_payload, maybe_execute_sync_via_local_video_decision,
+};
--- a/apps/aether-gateway/src/ai_pipeline/planner/specialized/video.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/specialized/video.rs
@@ -7,29 +7,31 @@ use serde_json::{json, Value};
 use tracing::warn;
 use uuid::Uuid;

-use crate::gateway::executor::request_candidates::{
-    current_unix_secs, record_local_request_candidate_status,
-};
 use crate::gateway::headers::collect_control_headers;
 use crate::gateway::provider_transport::{
    apply_local_body_rules, apply_local_header_rules, build_gemini_video_predict_long_running_url,
    build_passthrough_headers_with_auth, build_passthrough_path_url, resolve_local_gemini_auth,
    resolve_local_openai_chat_auth, resolve_transport_execution_timeouts,
-    resolve_transport_proxy_snapshot, resolve_transport_tls_profile,
+    resolve_transport_proxy_snapshot_with_tunnel_affinity, resolve_transport_tls_profile,
    supports_local_gemini_transport_with_network, supports_local_standard_transport_with_network,
 };
+use crate::gateway::request_candidates::{
+    current_unix_secs, record_local_request_candidate_status,
+};
+use crate::gateway::ai_pipeline::planner::plan_builders::{
+    build_passthrough_sync_plan_from_decision, LocalSyncPlanAndReport,
+};
+use crate::gateway::ai_pipeline::planner::prefer_local_tunnel_owner_candidates;
 use crate::gateway::scheduler::{
    list_selectable_candidates, GatewayMinimalCandidateSelectionCandidate,
 };
-use crate::gateway::{AppState, GatewayControlDecision, GatewayError};
-
-use super::super::plan_builders::{
-    build_passthrough_sync_plan_from_decision, LocalSyncPlanAndReport,
+use crate::gateway::{
+    execute_execution_runtime_sync, AppState, GatewayControlDecision,
+    GatewayControlSyncDecisionResponse, GatewayError,
 };
-use super::super::sync::execute_executor_sync;
-use super::super::{
-    GatewayControlSyncDecisionResponse, EXECUTOR_SYNC_DECISION_ACTION,
-    GEMINI_VIDEO_CREATE_SYNC_PLAN_KIND, OPENAI_VIDEO_CREATE_SYNC_PLAN_KIND,
+use crate::gateway::ai_pipeline::planner::{
+    EXECUTION_RUNTIME_SYNC_DECISION_ACTION, GEMINI_VIDEO_CREATE_SYNC_PLAN_KIND,
+    OPENAI_VIDEO_CREATE_SYNC_PLAN_KIND,
 };

 #[derive(Debug, Clone, Copy, PartialEq, Eq)]
@@ -50,7 +52,7 @@ struct LocalVideoCreateSpec {
 struct LocalVideoCreateDecisionInput {
    auth_context: crate::gateway::GatewayControlAuthContext,
    requested_model: String,
-    auth_snapshot: crate::gateway::data::StoredGatewayAuthApiKeySnapshot,
+    auth_snapshot: crate::gateway::gateway_data::StoredGatewayAuthApiKeySnapshot,
 }

 #[derive(Debug, Clone)]
@@ -60,10 +62,8 @@ struct LocalVideoCreateCandidateAttempt {
    candidate_id: String,
 }

-pub(in crate::gateway::executor) async fn maybe_execute_sync_via_local_video_decision(
+pub(crate) async fn maybe_execute_sync_via_local_video_decision(
    state: &AppState,
-    control_base_url: &str,
-    executor_base_url: &str,
    parts: &http::request::Parts,
    body_json: &serde_json::Value,
    trace_id: &str,
@@ -83,10 +83,8 @@ pub(in crate::gateway::executor) async fn maybe_execute_sync_via_local_video_dec

    let mut remaining = plan_and_reports.into_iter();
    while let Some(plan_and_report) = remaining.next() {
-        if let Some(response) = execute_executor_sync(
+        if let Some(response) = execute_execution_runtime_sync(
            state,
-            control_base_url,
-            executor_base_url,
            parts.uri.path(),
            plan_and_report.plan,
            trace_id,
@@ -105,7 +103,7 @@ pub(in crate::gateway::executor) async fn maybe_execute_sync_via_local_video_dec
    Ok(None)
 }

-pub(in crate::gateway::executor) async fn maybe_build_sync_local_video_decision_payload(
+pub(crate) async fn maybe_build_sync_local_video_decision_payload(
    state: &AppState,
    parts: &http::request::Parts,
    body_json: &serde_json::Value,
@@ -483,12 +481,18 @@ async fn maybe_build_local_video_create_decision_payload_for_candidate(
        .await;
        return None;
    }
-    let proxy = resolve_transport_proxy_snapshot(&transport);
+    let proxy = resolve_transport_proxy_snapshot_with_tunnel_affinity(state, &transport).await;
    let tls_profile = resolve_transport_tls_profile(&transport);

    Some(GatewayControlSyncDecisionResponse {
-        action: EXECUTOR_SYNC_DECISION_ACTION.to_string(),
+        action: EXECUTION_RUNTIME_SYNC_DECISION_ACTION.to_string(),
        decision_kind: Some(spec.decision_kind.to_string()),
+        execution_strategy: Some(
+            crate::gateway::ExecutionStrategy::LocalSameFormat
+                .as_str()
+                .to_string(),
+        ),
+        conversion_mode: Some(crate::gateway::ConversionMode::None.as_str().to_string()),
        request_id: Some(trace_id.to_string()),
        candidate_id: Some(candidate_id.clone()),
        provider_name: Some(transport.provider.name.clone()),
@@ -502,6 +506,8 @@ async fn maybe_build_local_video_create_decision_payload_for_candidate(
        auth_value: Some(auth_value),
        provider_api_format: Some(spec.api_format.to_string()),
        client_api_format: Some(spec.api_format.to_string()),
+        provider_contract: Some(spec.api_format.to_string()),
+        client_contract: Some(spec.api_format.to_string()),
        model_name: Some(input.requested_model.clone()),
        mapped_model: Some(mapped_model.clone()),
        prompt_cache_key: None,
@@ -614,6 +620,7 @@ async fn materialize_local_video_create_candidate_attempts(
    candidates: Vec<GatewayMinimalCandidateSelectionCandidate>,
    api_format: &str,
 ) -> Vec<LocalVideoCreateCandidateAttempt> {
+    let candidates = prefer_local_tunnel_owner_candidates(state, candidates).await;
    let created_at_unix_secs = current_unix_secs();
    let mut attempts = Vec::with_capacity(candidates.len());

--- a/apps/aether-gateway/src/ai_pipeline/planner/standard/claude/chat.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/standard/claude/chat.rs
@@ -0,0 +1,33 @@
+use crate::gateway::ai_pipeline::planner::{
+    CLAUDE_CHAT_STREAM_PLAN_KIND, CLAUDE_CHAT_SYNC_PLAN_KIND,
+};
+
+use super::super::family::{LocalStandardSourceFamily, LocalStandardSourceMode, LocalStandardSpec};
+
+pub(super) fn resolve_sync_spec(plan_kind: &str) -> Option<LocalStandardSpec> {
+    match plan_kind {
+        CLAUDE_CHAT_SYNC_PLAN_KIND => Some(LocalStandardSpec {
+            api_format: "claude:chat",
+            decision_kind: CLAUDE_CHAT_SYNC_PLAN_KIND,
+            report_kind: "claude_chat_sync_finalize",
+            family: LocalStandardSourceFamily::Standard,
+            mode: LocalStandardSourceMode::Chat,
+            require_streaming: false,
+        }),
+        _ => None,
+    }
+}
+
+pub(super) fn resolve_stream_spec(plan_kind: &str) -> Option<LocalStandardSpec> {
+    match plan_kind {
+        CLAUDE_CHAT_STREAM_PLAN_KIND => Some(LocalStandardSpec {
+            api_format: "claude:chat",
+            decision_kind: CLAUDE_CHAT_STREAM_PLAN_KIND,
+            report_kind: "claude_chat_stream_success",
+            family: LocalStandardSourceFamily::Standard,
+            mode: LocalStandardSourceMode::Chat,
+            require_streaming: true,
+        }),
+        _ => None,
+    }
+}
--- a/apps/aether-gateway/src/ai_pipeline/planner/standard/claude/cli.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/standard/claude/cli.rs
@@ -0,0 +1,33 @@
+use crate::gateway::ai_pipeline::planner::{
+    CLAUDE_CLI_STREAM_PLAN_KIND, CLAUDE_CLI_SYNC_PLAN_KIND,
+};
+
+use super::super::family::{LocalStandardSourceFamily, LocalStandardSourceMode, LocalStandardSpec};
+
+pub(super) fn resolve_sync_spec(plan_kind: &str) -> Option<LocalStandardSpec> {
+    match plan_kind {
+        CLAUDE_CLI_SYNC_PLAN_KIND => Some(LocalStandardSpec {
+            api_format: "claude:cli",
+            decision_kind: CLAUDE_CLI_SYNC_PLAN_KIND,
+            report_kind: "claude_cli_sync_finalize",
+            family: LocalStandardSourceFamily::Standard,
+            mode: LocalStandardSourceMode::Cli,
+            require_streaming: false,
+        }),
+        _ => None,
+    }
+}
+
+pub(super) fn resolve_stream_spec(plan_kind: &str) -> Option<LocalStandardSpec> {
+    match plan_kind {
+        CLAUDE_CLI_STREAM_PLAN_KIND => Some(LocalStandardSpec {
+            api_format: "claude:cli",
+            decision_kind: CLAUDE_CLI_STREAM_PLAN_KIND,
+            report_kind: "claude_cli_stream_success",
+            family: LocalStandardSourceFamily::Standard,
+            mode: LocalStandardSourceMode::Cli,
+            require_streaming: true,
+        }),
+        _ => None,
+    }
+}
--- a/apps/aether-gateway/src/ai_pipeline/planner/standard/claude/mod.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/standard/claude/mod.rs
@@ -0,0 +1,104 @@
+use axum::body::Body;
+use axum::http::Response;
+
+use crate::gateway::{
+    AppState, GatewayControlDecision, GatewayControlSyncDecisionResponse, GatewayError,
+};
+
+use super::family::{
+    maybe_build_stream_via_standard_family_payload, maybe_build_sync_via_standard_family_payload,
+    maybe_execute_stream_via_standard_family_decision,
+    maybe_execute_sync_via_standard_family_decision,
+};
+pub(crate) use crate::gateway::ai_pipeline::conversion::request::normalize_claude_request_to_openai_chat_request;
+
+mod chat;
+mod cli;
+
+pub(crate) async fn maybe_execute_sync_via_local_claude_decision(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    body_json: &serde_json::Value,
+    plan_kind: &str,
+) -> Result<Option<Response<Body>>, GatewayError> {
+    maybe_execute_sync_via_standard_family_decision(
+        state,
+        parts,
+        trace_id,
+        decision,
+        body_json,
+        plan_kind,
+        |plan_kind| {
+            chat::resolve_sync_spec(plan_kind).or_else(|| cli::resolve_sync_spec(plan_kind))
+        },
+    )
+    .await
+}
+
+pub(crate) async fn maybe_execute_stream_via_local_claude_decision(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    body_json: &serde_json::Value,
+    plan_kind: &str,
+) -> Result<Option<Response<Body>>, GatewayError> {
+    maybe_execute_stream_via_standard_family_decision(
+        state,
+        parts,
+        trace_id,
+        decision,
+        body_json,
+        plan_kind,
+        |plan_kind| {
+            chat::resolve_stream_spec(plan_kind).or_else(|| cli::resolve_stream_spec(plan_kind))
+        },
+    )
+    .await
+}
+
+pub(crate) async fn maybe_build_sync_local_claude_decision_payload(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    body_json: &serde_json::Value,
+    plan_kind: &str,
+) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
+    maybe_build_sync_via_standard_family_payload(
+        state,
+        parts,
+        trace_id,
+        decision,
+        body_json,
+        plan_kind,
+        |plan_kind| {
+            chat::resolve_sync_spec(plan_kind).or_else(|| cli::resolve_sync_spec(plan_kind))
+        },
+    )
+    .await
+}
+
+pub(crate) async fn maybe_build_stream_local_claude_decision_payload(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    body_json: &serde_json::Value,
+    plan_kind: &str,
+) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
+    maybe_build_stream_via_standard_family_payload(
+        state,
+        parts,
+        trace_id,
+        decision,
+        body_json,
+        plan_kind,
+        |plan_kind| {
+            chat::resolve_stream_spec(plan_kind).or_else(|| cli::resolve_stream_spec(plan_kind))
+        },
+    )
+    .await
+}
--- a/apps/aether-gateway/src/ai_pipeline/planner/standard/family/candidates.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/standard/family/candidates.rs
@@ -0,0 +1,271 @@
+use std::collections::BTreeSet;
+
+use aether_data::repository::candidates::{RequestCandidateStatus, UpsertRequestCandidateRecord};
+use serde_json::json;
+use tracing::warn;
+use uuid::Uuid;
+
+use crate::gateway::request_candidates::current_unix_secs;
+use crate::gateway::scheduler::{
+    list_selectable_candidates, GatewayMinimalCandidateSelectionCandidate,
+};
+use crate::gateway::{
+    append_execution_contract_fields_to_value, AppState, ConversionMode, ExecutionStrategy,
+    GatewayControlDecision, GatewayError,
+};
+
+use super::types::{
+    LocalStandardCandidateAttempt, LocalStandardDecisionInput, LocalStandardSourceFamily,
+    LocalStandardSourceMode, LocalStandardSpec,
+};
+
+pub(super) async fn resolve_local_standard_decision_input(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    body_json: &serde_json::Value,
+    spec: LocalStandardSpec,
+) -> Option<LocalStandardDecisionInput> {
+    let Some(auth_context) = decision.auth_context.clone().filter(|auth_context| {
+        !auth_context.user_id.trim().is_empty() && !auth_context.api_key_id.trim().is_empty()
+    }) else {
+        return None;
+    };
+
+    let requested_model = match spec.family {
+        LocalStandardSourceFamily::Standard => body_json
+            .get("model")
+            .and_then(serde_json::Value::as_str)
+            .map(str::trim)
+            .filter(|value| !value.is_empty())
+            .map(ToOwned::to_owned)?,
+        LocalStandardSourceFamily::Gemini => extract_gemini_model_from_path(parts.uri.path())?,
+    };
+
+    let auth_snapshot = match state
+        .read_auth_api_key_snapshot(
+            &auth_context.user_id,
+            &auth_context.api_key_id,
+            current_unix_secs(),
+        )
+        .await
+    {
+        Ok(Some(snapshot)) => snapshot,
+        Ok(None) => return None,
+        Err(err) => {
+            warn!(
+                trace_id = %trace_id,
+                api_format = spec.api_format,
+                error = ?err,
+                "gateway local standard decision auth snapshot read failed"
+            );
+            return None;
+        }
+    };
+
+    Some(LocalStandardDecisionInput {
+        auth_context,
+        requested_model,
+        auth_snapshot,
+    })
+}
+
+pub(super) async fn materialize_local_standard_candidate_attempts(
+    state: &AppState,
+    trace_id: &str,
+    input: &LocalStandardDecisionInput,
+    spec: LocalStandardSpec,
+) -> Result<Vec<LocalStandardCandidateAttempt>, GatewayError> {
+    let mut seen_candidates = BTreeSet::new();
+    let mut candidates = Vec::new();
+    for candidate_api_format in candidate_api_formats_for_spec(spec) {
+        let auth_snapshot = if *candidate_api_format == spec.api_format {
+            Some(&input.auth_snapshot)
+        } else {
+            None
+        };
+        let mut selected_candidates = list_selectable_candidates(
+            state,
+            candidate_api_format,
+            &input.requested_model,
+            spec.require_streaming,
+            auth_snapshot,
+            current_unix_secs(),
+        )
+        .await?;
+        if auth_snapshot.is_none() {
+            selected_candidates.retain(|candidate| {
+                auth_snapshot_allows_cross_format_candidate(
+                    &input.auth_snapshot,
+                    &input.requested_model,
+                    candidate,
+                )
+            });
+        }
+        for candidate in selected_candidates {
+            let candidate_key = format!(
+                "{}:{}:{}:{}:{}:{}",
+                candidate.provider_id,
+                candidate.endpoint_id,
+                candidate.key_id,
+                candidate.model_id,
+                candidate.selected_provider_model_name,
+                candidate.endpoint_api_format,
+            );
+            if seen_candidates.insert(candidate_key) {
+                candidates.push(candidate);
+            }
+        }
+    }
+    let candidates =
+        crate::gateway::ai_pipeline::planner::prefer_local_tunnel_owner_candidates(state, candidates)
+            .await;
+
+    let created_at_unix_secs = current_unix_secs();
+    let mut attempts = Vec::with_capacity(candidates.len());
+    for (candidate_index, candidate) in candidates.into_iter().enumerate() {
+        let candidate_id = Uuid::new_v4().to_string();
+        let provider_api_format = candidate.endpoint_api_format.trim().to_ascii_lowercase();
+        let execution_strategy = if provider_api_format == spec.api_format {
+            ExecutionStrategy::LocalSameFormat
+        } else {
+            ExecutionStrategy::LocalCrossFormat
+        };
+        let conversion_mode = if crate::gateway::ai_pipeline::conversion::request_conversion_kind(
+            spec.api_format,
+            provider_api_format.as_str(),
+        )
+        .is_some()
+        {
+            ConversionMode::Bidirectional
+        } else {
+            ConversionMode::None
+        };
+        let extra_data = append_execution_contract_fields_to_value(
+            json!({
+                "provider_api_format": provider_api_format,
+                "client_api_format": spec.api_format,
+                "global_model_id": candidate.global_model_id.clone(),
+                "global_model_name": candidate.global_model_name.clone(),
+                "model_id": candidate.model_id.clone(),
+                "selected_provider_model_name": candidate.selected_provider_model_name.clone(),
+                "mapping_matched_model": candidate.mapping_matched_model.clone(),
+                "provider_name": candidate.provider_name.clone(),
+                "key_name": candidate.key_name.clone(),
+            }),
+            execution_strategy,
+            conversion_mode,
+            spec.api_format,
+            candidate.endpoint_api_format.as_str(),
+        );
+
+        let stored_candidate_id = match state
+            .upsert_request_candidate(UpsertRequestCandidateRecord {
+                id: candidate_id.clone(),
+                request_id: trace_id.to_string(),
+                user_id: Some(input.auth_context.user_id.clone()),
+                api_key_id: Some(input.auth_context.api_key_id.clone()),
+                username: None,
+                api_key_name: None,
+                candidate_index: candidate_index as u32,
+                retry_index: 0,
+                provider_id: Some(candidate.provider_id.clone()),
+                endpoint_id: Some(candidate.endpoint_id.clone()),
+                key_id: Some(candidate.key_id.clone()),
+                status: RequestCandidateStatus::Available,
+                skip_reason: None,
+                is_cached: Some(false),
+                status_code: None,
+                error_type: None,
+                error_message: None,
+                latency_ms: None,
+                concurrent_requests: None,
+                extra_data: Some(extra_data),
+                required_capabilities: candidate.key_capabilities.clone(),
+                created_at_unix_secs: Some(created_at_unix_secs),
+                started_at_unix_secs: None,
+                finished_at_unix_secs: None,
+            })
+            .await
+        {
+            Ok(Some(stored)) => stored.id,
+            Ok(None) => candidate_id.clone(),
+            Err(err) => {
+                warn!(
+                    trace_id = %trace_id,
+                    api_format = spec.api_format,
+                    error = ?err,
+                    "gateway local standard decision request candidate upsert failed"
+                );
+                candidate_id.clone()
+            }
+        };
+
+        attempts.push(LocalStandardCandidateAttempt {
+            candidate,
+            candidate_index: candidate_index as u32,
+            candidate_id: stored_candidate_id,
+        });
+    }
+
+    Ok(attempts)
+}
+
+fn auth_snapshot_allows_cross_format_candidate(
+    auth_snapshot: &crate::gateway::gateway_data::StoredGatewayAuthApiKeySnapshot,
+    requested_model: &str,
+    candidate: &GatewayMinimalCandidateSelectionCandidate,
+) -> bool {
+    if let Some(allowed_providers) = auth_snapshot.effective_allowed_providers() {
+        let provider_allowed = allowed_providers.iter().any(|value| {
+            value
+                .trim()
+                .eq_ignore_ascii_case(candidate.provider_id.trim())
+                || value
+                    .trim()
+                    .eq_ignore_ascii_case(candidate.provider_name.trim())
+        });
+        if !provider_allowed {
+            return false;
+        }
+    }
+
+    if let Some(allowed_models) = auth_snapshot.effective_allowed_models() {
+        let model_allowed = allowed_models
+            .iter()
+            .any(|value| value == requested_model || value == &candidate.global_model_name);
+        if !model_allowed {
+            return false;
+        }
+    }
+
+    true
+}
+
+fn candidate_api_formats_for_spec(spec: LocalStandardSpec) -> &'static [&'static str] {
+    match spec.mode {
+        LocalStandardSourceMode::Chat | LocalStandardSourceMode::Cli => &[
+            "openai:chat",
+            "openai:cli",
+            "openai:compact",
+            "claude:chat",
+            "claude:cli",
+            "gemini:chat",
+            "gemini:cli",
+        ],
+    }
+}
+
+fn extract_gemini_model_from_path(path: &str) -> Option<String> {
+    let marker = "/models/";
+    let start = path.find(marker)? + marker.len();
+    let tail = &path[start..];
+    let end = tail.find(':').unwrap_or(tail.len());
+    let model = tail[..end].trim();
+    if model.is_empty() {
+        None
+    } else {
+        Some(model.to_string())
+    }
+}
--- a/apps/aether-gateway/src/ai_pipeline/planner/standard/family/execute.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/standard/family/execute.rs
@@ -0,0 +1,236 @@
+use axum::body::Body;
+use axum::http::Response;
+use tracing::warn;
+
+use crate::gateway::ai_pipeline::planner::family_core::{
+    execute_stream_plan_and_reports, execute_sync_plan_and_reports,
+};
+use crate::gateway::ai_pipeline::planner::plan_builders::{
+    build_gemini_stream_plan_from_decision, build_gemini_sync_plan_from_decision,
+    build_standard_stream_plan_from_decision, build_standard_sync_plan_from_decision,
+    LocalStreamPlanAndReport, LocalSyncPlanAndReport,
+};
+use crate::gateway::{AppState, GatewayControlDecision, GatewayControlSyncDecisionResponse, GatewayError};
+
+use super::candidates::{
+    materialize_local_standard_candidate_attempts, resolve_local_standard_decision_input,
+};
+use super::payload::maybe_build_local_standard_decision_payload_for_candidate;
+use super::types::{LocalStandardSourceFamily, LocalStandardSpec};
+
+pub(crate) async fn maybe_execute_sync_via_standard_family_decision(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    body_json: &serde_json::Value,
+    plan_kind: &str,
+    resolve_sync_spec: fn(&str) -> Option<LocalStandardSpec>,
+) -> Result<Option<Response<Body>>, GatewayError> {
+    let Some(spec) = resolve_sync_spec(plan_kind) else {
+        return Ok(None);
+    };
+
+    let plan_and_reports =
+        build_local_sync_plan_and_reports(state, parts, trace_id, decision, body_json, spec)
+            .await?;
+    if plan_and_reports.is_empty() {
+        return Ok(None);
+    }
+
+    execute_sync_plan_and_reports(state, parts, trace_id, decision, plan_kind, plan_and_reports)
+        .await
+}
+
+pub(crate) async fn maybe_execute_stream_via_standard_family_decision(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    body_json: &serde_json::Value,
+    plan_kind: &str,
+    resolve_stream_spec: fn(&str) -> Option<LocalStandardSpec>,
+) -> Result<Option<Response<Body>>, GatewayError> {
+    let Some(spec) = resolve_stream_spec(plan_kind) else {
+        return Ok(None);
+    };
+
+    let plan_and_reports =
+        build_local_stream_plan_and_reports(state, parts, trace_id, decision, body_json, spec)
+            .await?;
+    if plan_and_reports.is_empty() {
+        return Ok(None);
+    }
+
+    execute_stream_plan_and_reports(state, trace_id, decision, plan_kind, plan_and_reports).await
+}
+
+pub(crate) async fn maybe_build_sync_via_standard_family_payload(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    body_json: &serde_json::Value,
+    plan_kind: &str,
+    resolve_sync_spec: fn(&str) -> Option<LocalStandardSpec>,
+) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
+    let Some(spec) = resolve_sync_spec(plan_kind) else {
+        return Ok(None);
+    };
+
+    let Some(input) =
+        resolve_local_standard_decision_input(state, parts, trace_id, decision, body_json, spec)
+            .await
+    else {
+        return Ok(None);
+    };
+
+    let attempts =
+        materialize_local_standard_candidate_attempts(state, trace_id, &input, spec).await?;
+
+    for attempt in attempts {
+        if let Some(payload) = maybe_build_local_standard_decision_payload_for_candidate(
+            state, parts, trace_id, body_json, &input, attempt, spec,
+        )
+        .await
+        {
+            return Ok(Some(payload));
+        }
+    }
+
+    Ok(None)
+}
+
+pub(crate) async fn maybe_build_stream_via_standard_family_payload(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    body_json: &serde_json::Value,
+    plan_kind: &str,
+    resolve_stream_spec: fn(&str) -> Option<LocalStandardSpec>,
+) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
+    let Some(spec) = resolve_stream_spec(plan_kind) else {
+        return Ok(None);
+    };
+
+    let Some(input) =
+        resolve_local_standard_decision_input(state, parts, trace_id, decision, body_json, spec)
+            .await
+    else {
+        return Ok(None);
+    };
+
+    let attempts =
+        materialize_local_standard_candidate_attempts(state, trace_id, &input, spec).await?;
+
+    for attempt in attempts {
+        if let Some(payload) = maybe_build_local_standard_decision_payload_for_candidate(
+            state, parts, trace_id, body_json, &input, attempt, spec,
+        )
+        .await
+        {
+            return Ok(Some(payload));
+        }
+    }
+
+    Ok(None)
+}
+
+async fn build_local_sync_plan_and_reports(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    body_json: &serde_json::Value,
+    spec: LocalStandardSpec,
+) -> Result<Vec<LocalSyncPlanAndReport>, GatewayError> {
+    let Some(input) =
+        resolve_local_standard_decision_input(state, parts, trace_id, decision, body_json, spec)
+            .await
+    else {
+        return Ok(Vec::new());
+    };
+    let attempts =
+        materialize_local_standard_candidate_attempts(state, trace_id, &input, spec).await?;
+    let mut plans = Vec::new();
+    for attempt in attempts {
+        let Some(payload) = maybe_build_local_standard_decision_payload_for_candidate(
+            state, parts, trace_id, body_json, &input, attempt, spec,
+        )
+        .await
+        else {
+            continue;
+        };
+        let built = match spec.family {
+            LocalStandardSourceFamily::Standard => {
+                build_standard_sync_plan_from_decision(parts, body_json, payload)
+            }
+            LocalStandardSourceFamily::Gemini => {
+                build_gemini_sync_plan_from_decision(parts, body_json, payload)
+            }
+        };
+        match built {
+            Ok(Some(value)) => plans.push(value),
+            Ok(None) => {}
+            Err(err) => {
+                warn!(
+                    trace_id = %trace_id,
+                    api_format = spec.api_format,
+                    error = ?err,
+                    "gateway local standard sync plan build failed"
+                );
+            }
+        }
+    }
+    Ok(plans)
+}
+
+async fn build_local_stream_plan_and_reports(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    body_json: &serde_json::Value,
+    spec: LocalStandardSpec,
+) -> Result<Vec<LocalStreamPlanAndReport>, GatewayError> {
+    let Some(input) =
+        resolve_local_standard_decision_input(state, parts, trace_id, decision, body_json, spec)
+            .await
+    else {
+        return Ok(Vec::new());
+    };
+    let attempts =
+        materialize_local_standard_candidate_attempts(state, trace_id, &input, spec).await?;
+    let mut plans = Vec::new();
+    for attempt in attempts {
+        let Some(payload) = maybe_build_local_standard_decision_payload_for_candidate(
+            state, parts, trace_id, body_json, &input, attempt, spec,
+        )
+        .await
+        else {
+            continue;
+        };
+        let built = match spec.family {
+            LocalStandardSourceFamily::Standard => {
+                build_standard_stream_plan_from_decision(parts, body_json, payload, false)
+            }
+            LocalStandardSourceFamily::Gemini => {
+                build_gemini_stream_plan_from_decision(parts, body_json, payload)
+            }
+        };
+        match built {
+            Ok(Some(value)) => plans.push(value),
+            Ok(None) => {}
+            Err(err) => {
+                warn!(
+                    trace_id = %trace_id,
+                    api_format = spec.api_format,
+                    error = ?err,
+                    "gateway local standard stream plan build failed"
+                );
+            }
+        }
+    }
+    Ok(plans)
+}
--- a/apps/aether-gateway/src/ai_pipeline/planner/standard/family/mod.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/standard/family/mod.rs
@@ -0,0 +1,13 @@
+mod candidates;
+mod execute;
+mod payload;
+mod types;
+
+pub(crate) use self::execute::{
+    maybe_build_stream_via_standard_family_payload, maybe_build_sync_via_standard_family_payload,
+    maybe_execute_stream_via_standard_family_decision,
+    maybe_execute_sync_via_standard_family_decision,
+};
+pub(crate) use self::types::{
+    LocalStandardSourceFamily, LocalStandardSourceMode, LocalStandardSpec,
+};
--- a/apps/aether-gateway/src/ai_pipeline/planner/standard/family/payload.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/standard/family/payload.rs
@@ -0,0 +1,360 @@
+use std::collections::BTreeMap;
+
+use aether_data::repository::candidates::{RequestCandidateStatus, UpsertRequestCandidateRecord};
+use serde_json::json;
+use tracing::warn;
+
+use crate::gateway::headers::collect_control_headers;
+use crate::gateway::provider_transport::{
+    apply_local_header_rules, build_openai_passthrough_headers, ensure_upstream_auth_header,
+    resolve_transport_execution_timeouts, resolve_transport_proxy_snapshot_with_tunnel_affinity,
+    resolve_transport_tls_profile, LocalResolvedOAuthRequestAuth,
+};
+use crate::gateway::request_candidates::current_unix_secs;
+use crate::gateway::{
+    append_execution_contract_fields_to_value, AppState, ConversionMode, ExecutionStrategy,
+    GatewayControlSyncDecisionResponse, EXECUTION_RUNTIME_STREAM_DECISION_ACTION,
+    EXECUTION_RUNTIME_SYNC_DECISION_ACTION,
+};
+
+use super::types::{LocalStandardCandidateAttempt, LocalStandardDecisionInput, LocalStandardSpec};
+
+pub(super) async fn maybe_build_local_standard_decision_payload_for_candidate(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    body_json: &serde_json::Value,
+    input: &LocalStandardDecisionInput,
+    attempt: LocalStandardCandidateAttempt,
+    spec: LocalStandardSpec,
+) -> Option<GatewayControlSyncDecisionResponse> {
+    let LocalStandardCandidateAttempt {
+        candidate,
+        candidate_index,
+        candidate_id,
+    } = attempt;
+    let provider_api_format = candidate.endpoint_api_format.trim().to_ascii_lowercase();
+    let Some(conversion_kind) = crate::gateway::ai_pipeline::conversion::request_conversion_kind(
+        spec.api_format,
+        provider_api_format.as_str(),
+    ) else {
+        if provider_api_format == spec.api_format {
+            return None;
+        }
+        return None;
+    };
+
+    let transport = match state
+        .read_provider_transport_snapshot(
+            &candidate.provider_id,
+            &candidate.endpoint_id,
+            &candidate.key_id,
+        )
+        .await
+    {
+        Ok(Some(snapshot)) => snapshot,
+        Ok(None) => {
+            mark_skipped_local_standard_candidate(
+                state,
+                input,
+                trace_id,
+                &candidate,
+                candidate_index,
+                &candidate_id,
+                "transport_snapshot_missing",
+            )
+            .await;
+            return None;
+        }
+        Err(err) => {
+            warn!(
+                trace_id = %trace_id,
+                api_format = spec.api_format,
+                error = ?err,
+                "gateway local standard decision provider transport read failed"
+            );
+            mark_skipped_local_standard_candidate(
+                state,
+                input,
+                trace_id,
+                &candidate,
+                candidate_index,
+                &candidate_id,
+                "transport_snapshot_read_failed",
+            )
+            .await;
+            return None;
+        }
+    };
+
+    if !crate::gateway::ai_pipeline::conversion::request_conversion_transport_supported(
+        &transport,
+        conversion_kind,
+    ) {
+        mark_skipped_local_standard_candidate(
+            state,
+            input,
+            trace_id,
+            &candidate,
+            candidate_index,
+            &candidate_id,
+            "transport_unsupported",
+        )
+        .await;
+        return None;
+    }
+
+    let resolved_auth = crate::gateway::ai_pipeline::conversion::request_conversion_direct_auth(
+        &transport,
+        conversion_kind,
+    );
+    let oauth_auth = if resolved_auth.is_none() {
+        match state.resolve_local_oauth_request_auth(&transport).await {
+            Ok(Some(LocalResolvedOAuthRequestAuth::Header { name, value })) => Some((name, value)),
+            Ok(Some(LocalResolvedOAuthRequestAuth::Kiro(_))) => None,
+            Ok(None) => None,
+            Err(err) => {
+                warn!(
+                    trace_id = %trace_id,
+                    api_format = spec.api_format,
+                    provider_type = %transport.provider.provider_type,
+                    error = ?err,
+                    "gateway local standard oauth auth resolution failed"
+                );
+                None
+            }
+        }
+    } else {
+        None
+    };
+
+    let Some((auth_header, auth_value)) = resolved_auth.or(oauth_auth) else {
+        mark_skipped_local_standard_candidate(
+            state,
+            input,
+            trace_id,
+            &candidate,
+            candidate_index,
+            &candidate_id,
+            "transport_auth_unavailable",
+        )
+        .await;
+        return None;
+    };
+
+    let mapped_model = candidate.selected_provider_model_name.trim().to_string();
+    if mapped_model.is_empty() {
+        mark_skipped_local_standard_candidate(
+            state,
+            input,
+            trace_id,
+            &candidate,
+            candidate_index,
+            &candidate_id,
+            "mapped_model_missing",
+        )
+        .await;
+        return None;
+    }
+
+    let provider_request_body =
+        match crate::gateway::ai_pipeline::planner::standard::build_standard_request_body(
+            body_json,
+            spec.api_format,
+            &mapped_model,
+            provider_api_format.as_str(),
+            parts.uri.path(),
+            spec.require_streaming,
+            transport.endpoint.body_rules.as_ref(),
+        ) {
+            Some(body) => body,
+            None => {
+                mark_skipped_local_standard_candidate(
+                    state,
+                    input,
+                    trace_id,
+                    &candidate,
+                    candidate_index,
+                    &candidate_id,
+                    "provider_request_body_missing",
+                )
+                .await;
+                return None;
+            }
+        };
+
+    let upstream_url =
+        match crate::gateway::ai_pipeline::planner::standard::build_standard_upstream_url(
+            parts,
+            &transport,
+            &mapped_model,
+            provider_api_format.as_str(),
+            spec.require_streaming,
+        ) {
+            Some(url) => url,
+            None => {
+                mark_skipped_local_standard_candidate(
+                    state,
+                    input,
+                    trace_id,
+                    &candidate,
+                    candidate_index,
+                    &candidate_id,
+                    "upstream_url_missing",
+                )
+                .await;
+                return None;
+            }
+        };
+
+    let mut provider_request_headers = build_openai_passthrough_headers(
+        &parts.headers,
+        &auth_header,
+        &auth_value,
+        &BTreeMap::new(),
+        Some("application/json"),
+    );
+    if !apply_local_header_rules(
+        &mut provider_request_headers,
+        transport.endpoint.header_rules.as_ref(),
+        &[&auth_header, "content-type"],
+        &provider_request_body,
+        Some(body_json),
+    ) {
+        mark_skipped_local_standard_candidate(
+            state,
+            input,
+            trace_id,
+            &candidate,
+            candidate_index,
+            &candidate_id,
+            "transport_header_rules_apply_failed",
+        )
+        .await;
+        return None;
+    }
+    ensure_upstream_auth_header(&mut provider_request_headers, &auth_header, &auth_value);
+    if spec.require_streaming {
+        provider_request_headers
+            .entry("accept".to_string())
+            .or_insert_with(|| "text/event-stream".to_string());
+    }
+
+    Some(GatewayControlSyncDecisionResponse {
+        action: if spec.require_streaming {
+            EXECUTION_RUNTIME_STREAM_DECISION_ACTION.to_string()
+        } else {
+            EXECUTION_RUNTIME_SYNC_DECISION_ACTION.to_string()
+        },
+        decision_kind: Some(spec.decision_kind.to_string()),
+        execution_strategy: Some(ExecutionStrategy::LocalCrossFormat.as_str().to_string()),
+        conversion_mode: Some(ConversionMode::Bidirectional.as_str().to_string()),
+        request_id: Some(trace_id.to_string()),
+        candidate_id: Some(candidate_id.clone()),
+        provider_name: Some(candidate.provider_name.clone()),
+        provider_id: Some(candidate.provider_id.clone()),
+        endpoint_id: Some(candidate.endpoint_id.clone()),
+        key_id: Some(candidate.key_id.clone()),
+        upstream_base_url: Some(transport.endpoint.base_url.clone()),
+        upstream_url: Some(upstream_url.clone()),
+        provider_request_method: None,
+        auth_header: Some(auth_header),
+        auth_value: Some(auth_value),
+        provider_api_format: Some(provider_api_format.clone()),
+        client_api_format: Some(spec.api_format.to_string()),
+        provider_contract: Some(provider_api_format.clone()),
+        client_contract: Some(spec.api_format.to_string()),
+        model_name: Some(input.requested_model.clone()),
+        mapped_model: Some(mapped_model.clone()),
+        prompt_cache_key: None,
+        extra_headers: BTreeMap::new(),
+        provider_request_headers: provider_request_headers.clone(),
+        provider_request_body: Some(provider_request_body.clone()),
+        provider_request_body_base64: None,
+        content_type: Some("application/json".to_string()),
+        proxy: resolve_transport_proxy_snapshot_with_tunnel_affinity(state, &transport).await,
+        tls_profile: resolve_transport_tls_profile(&transport),
+        timeouts: resolve_transport_execution_timeouts(&transport),
+        upstream_is_stream: spec.require_streaming,
+        report_kind: Some(spec.report_kind.to_string()),
+        report_context: Some(append_execution_contract_fields_to_value(
+            json!({
+                "user_id": input.auth_context.user_id,
+                "api_key_id": input.auth_context.api_key_id,
+                "request_id": trace_id,
+                "candidate_id": candidate_id,
+                "candidate_index": candidate_index,
+                "retry_index": 0,
+                "model": input.requested_model,
+                "provider_name": candidate.provider_name,
+                "provider_id": candidate.provider_id,
+                "endpoint_id": candidate.endpoint_id,
+                "key_id": candidate.key_id,
+                "provider_api_format": provider_api_format,
+                "client_api_format": spec.api_format,
+                "mapped_model": mapped_model,
+                "upstream_url": upstream_url,
+                "provider_request_method": serde_json::Value::Null,
+                "provider_request_headers": provider_request_headers,
+                "provider_request_body": provider_request_body,
+                "original_headers": collect_control_headers(&parts.headers),
+                "original_request_body": body_json,
+                "has_envelope": false,
+                "needs_conversion": true,
+            }),
+            ExecutionStrategy::LocalCrossFormat,
+            ConversionMode::Bidirectional,
+            spec.api_format,
+            candidate.endpoint_api_format.as_str(),
+        )),
+        auth_context: Some(input.auth_context.clone()),
+    })
+}
+
+pub(super) async fn mark_skipped_local_standard_candidate(
+    state: &AppState,
+    input: &LocalStandardDecisionInput,
+    trace_id: &str,
+    candidate: &crate::gateway::scheduler::GatewayMinimalCandidateSelectionCandidate,
+    candidate_index: u32,
+    candidate_id: &str,
+    skip_reason: &'static str,
+) {
+    if let Err(err) = state
+        .upsert_request_candidate(UpsertRequestCandidateRecord {
+            id: candidate_id.to_string(),
+            request_id: trace_id.to_string(),
+            user_id: Some(input.auth_context.user_id.clone()),
+            api_key_id: Some(input.auth_context.api_key_id.clone()),
+            username: None,
+            api_key_name: None,
+            candidate_index,
+            retry_index: 0,
+            provider_id: Some(candidate.provider_id.clone()),
+            endpoint_id: Some(candidate.endpoint_id.clone()),
+            key_id: Some(candidate.key_id.clone()),
+            status: RequestCandidateStatus::Skipped,
+            skip_reason: Some(skip_reason.to_string()),
+            is_cached: Some(false),
+            status_code: None,
+            error_type: None,
+            error_message: None,
+            latency_ms: None,
+            concurrent_requests: None,
+            extra_data: None,
+            required_capabilities: candidate.key_capabilities.clone(),
+            created_at_unix_secs: None,
+            started_at_unix_secs: None,
+            finished_at_unix_secs: Some(current_unix_secs()),
+        })
+        .await
+    {
+        warn!(
+            trace_id = %trace_id,
+            candidate_id = %candidate_id,
+            skip_reason,
+            error = ?err,
+            "gateway local standard decision failed to persist skipped candidate"
+        );
+    }
+}
--- a/apps/aether-gateway/src/ai_pipeline/planner/standard/family/types.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/standard/family/types.rs
@@ -0,0 +1,35 @@
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub(crate) enum LocalStandardSourceFamily {
+    Standard,
+    Gemini,
+}
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub(crate) enum LocalStandardSourceMode {
+    Chat,
+    Cli,
+}
+
+#[derive(Debug, Clone, Copy)]
+pub(crate) struct LocalStandardSpec {
+    pub(crate) api_format: &'static str,
+    pub(crate) decision_kind: &'static str,
+    pub(crate) report_kind: &'static str,
+    pub(crate) family: LocalStandardSourceFamily,
+    pub(crate) mode: LocalStandardSourceMode,
+    pub(crate) require_streaming: bool,
+}
+
+#[derive(Debug, Clone)]
+pub(super) struct LocalStandardDecisionInput {
+    pub(super) auth_context: crate::gateway::GatewayControlAuthContext,
+    pub(super) requested_model: String,
+    pub(super) auth_snapshot: crate::gateway::gateway_data::StoredGatewayAuthApiKeySnapshot,
+}
+
+#[derive(Debug, Clone)]
+pub(super) struct LocalStandardCandidateAttempt {
+    pub(super) candidate: crate::gateway::scheduler::GatewayMinimalCandidateSelectionCandidate,
+    pub(super) candidate_index: u32,
+    pub(super) candidate_id: String,
+}
--- a/apps/aether-gateway/src/ai_pipeline/planner/standard/gemini/chat.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/standard/gemini/chat.rs
@@ -0,0 +1,33 @@
+use crate::gateway::ai_pipeline::planner::{
+    GEMINI_CHAT_STREAM_PLAN_KIND, GEMINI_CHAT_SYNC_PLAN_KIND,
+};
+
+use super::super::family::{LocalStandardSourceFamily, LocalStandardSourceMode, LocalStandardSpec};
+
+pub(super) fn resolve_sync_spec(plan_kind: &str) -> Option<LocalStandardSpec> {
+    match plan_kind {
+        GEMINI_CHAT_SYNC_PLAN_KIND => Some(LocalStandardSpec {
+            api_format: "gemini:chat",
+            decision_kind: GEMINI_CHAT_SYNC_PLAN_KIND,
+            report_kind: "gemini_chat_sync_finalize",
+            family: LocalStandardSourceFamily::Gemini,
+            mode: LocalStandardSourceMode::Chat,
+            require_streaming: false,
+        }),
+        _ => None,
+    }
+}
+
+pub(super) fn resolve_stream_spec(plan_kind: &str) -> Option<LocalStandardSpec> {
+    match plan_kind {
+        GEMINI_CHAT_STREAM_PLAN_KIND => Some(LocalStandardSpec {
+            api_format: "gemini:chat",
+            decision_kind: GEMINI_CHAT_STREAM_PLAN_KIND,
+            report_kind: "gemini_chat_stream_success",
+            family: LocalStandardSourceFamily::Gemini,
+            mode: LocalStandardSourceMode::Chat,
+            require_streaming: true,
+        }),
+        _ => None,
+    }
+}
--- a/apps/aether-gateway/src/ai_pipeline/planner/standard/gemini/cli.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/standard/gemini/cli.rs
@@ -0,0 +1,33 @@
+use crate::gateway::ai_pipeline::planner::{
+    GEMINI_CLI_STREAM_PLAN_KIND, GEMINI_CLI_SYNC_PLAN_KIND,
+};
+
+use super::super::family::{LocalStandardSourceFamily, LocalStandardSourceMode, LocalStandardSpec};
+
+pub(super) fn resolve_sync_spec(plan_kind: &str) -> Option<LocalStandardSpec> {
+    match plan_kind {
+        GEMINI_CLI_SYNC_PLAN_KIND => Some(LocalStandardSpec {
+            api_format: "gemini:cli",
+            decision_kind: GEMINI_CLI_SYNC_PLAN_KIND,
+            report_kind: "gemini_cli_sync_finalize",
+            family: LocalStandardSourceFamily::Gemini,
+            mode: LocalStandardSourceMode::Cli,
+            require_streaming: false,
+        }),
+        _ => None,
+    }
+}
+
+pub(super) fn resolve_stream_spec(plan_kind: &str) -> Option<LocalStandardSpec> {
+    match plan_kind {
+        GEMINI_CLI_STREAM_PLAN_KIND => Some(LocalStandardSpec {
+            api_format: "gemini:cli",
+            decision_kind: GEMINI_CLI_STREAM_PLAN_KIND,
+            report_kind: "gemini_cli_stream_success",
+            family: LocalStandardSourceFamily::Gemini,
+            mode: LocalStandardSourceMode::Cli,
+            require_streaming: true,
+        }),
+        _ => None,
+    }
+}
--- a/apps/aether-gateway/src/ai_pipeline/planner/standard/gemini/mod.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/standard/gemini/mod.rs
@@ -0,0 +1,104 @@
+use axum::body::Body;
+use axum::http::Response;
+
+use crate::gateway::{
+    AppState, GatewayControlDecision, GatewayControlSyncDecisionResponse, GatewayError,
+};
+
+use super::family::{
+    maybe_build_stream_via_standard_family_payload, maybe_build_sync_via_standard_family_payload,
+    maybe_execute_stream_via_standard_family_decision,
+    maybe_execute_sync_via_standard_family_decision,
+};
+pub(crate) use crate::gateway::ai_pipeline::conversion::request::normalize_gemini_request_to_openai_chat_request;
+
+mod chat;
+mod cli;
+
+pub(crate) async fn maybe_execute_sync_via_local_gemini_decision(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    body_json: &serde_json::Value,
+    plan_kind: &str,
+) -> Result<Option<Response<Body>>, GatewayError> {
+    maybe_execute_sync_via_standard_family_decision(
+        state,
+        parts,
+        trace_id,
+        decision,
+        body_json,
+        plan_kind,
+        |plan_kind| {
+            chat::resolve_sync_spec(plan_kind).or_else(|| cli::resolve_sync_spec(plan_kind))
+        },
+    )
+    .await
+}
+
+pub(crate) async fn maybe_execute_stream_via_local_gemini_decision(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    body_json: &serde_json::Value,
+    plan_kind: &str,
+) -> Result<Option<Response<Body>>, GatewayError> {
+    maybe_execute_stream_via_standard_family_decision(
+        state,
+        parts,
+        trace_id,
+        decision,
+        body_json,
+        plan_kind,
+        |plan_kind| {
+            chat::resolve_stream_spec(plan_kind).or_else(|| cli::resolve_stream_spec(plan_kind))
+        },
+    )
+    .await
+}
+
+pub(crate) async fn maybe_build_sync_local_gemini_decision_payload(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    body_json: &serde_json::Value,
+    plan_kind: &str,
+) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
+    maybe_build_sync_via_standard_family_payload(
+        state,
+        parts,
+        trace_id,
+        decision,
+        body_json,
+        plan_kind,
+        |plan_kind| {
+            chat::resolve_sync_spec(plan_kind).or_else(|| cli::resolve_sync_spec(plan_kind))
+        },
+    )
+    .await
+}
+
+pub(crate) async fn maybe_build_stream_local_gemini_decision_payload(
+    state: &AppState,
+    parts: &http::request::Parts,
+    trace_id: &str,
+    decision: &GatewayControlDecision,
+    body_json: &serde_json::Value,
+    plan_kind: &str,
+) -> Result<Option<GatewayControlSyncDecisionResponse>, GatewayError> {
+    maybe_build_stream_via_standard_family_payload(
+        state,
+        parts,
+        trace_id,
+        decision,
+        body_json,
+        plan_kind,
+        |plan_kind| {
+            chat::resolve_stream_spec(plan_kind).or_else(|| cli::resolve_stream_spec(plan_kind))
+        },
+    )
+    .await
+}
--- a/apps/aether-gateway/src/ai_pipeline/planner/standard/gemini/plan_builders.rs
+++ b/apps/aether-gateway/src/ai_pipeline/planner/standard/gemini/plan_builders.rs
@@ -0,0 +1,242 @@
+use aether_contracts::{ExecutionPlan, RequestBody};
+
+use super::{
+    augment_sync_report_context, generic_decision_missing_exact_provider_request,
+    LocalStreamPlanAndReport, LocalSyncPlanAndReport,
+};
+use crate::gateway::provider_transport::ensure_upstream_auth_header;
+use crate::gateway::{GatewayControlSyncDecisionResponse, GatewayError};
+
+pub(crate) fn build_gemini_sync_plan_from_decision(
+    _parts: &http::request::Parts,
+    _body_json: &serde_json::Value,
+    payload: GatewayControlSyncDecisionResponse,
+) -> Result<Option<LocalSyncPlanAndReport>, GatewayError> {
+    if generic_decision_missing_exact_provider_request(&payload) {
+        return Ok(None);
+    }
+    let Some(request_id) = payload
+        .request_id
+        .clone()
+        .filter(|value| !value.trim().is_empty())
+    else {
+        return Ok(None);
+    };
+    let Some(provider_id) = payload
+        .provider_id
+        .clone()
+        .filter(|value| !value.trim().is_empty())
+    else {
+        return Ok(None);
+    };
+    let Some(endpoint_id) = payload
+        .endpoint_id
+        .clone()
+        .filter(|value| !value.trim().is_empty())
+    else {
+        return Ok(None);
+    };
+    let Some(key_id) = payload
+        .key_id
+        .clone()
+        .filter(|value| !value.trim().is_empty())
+    else {
+        return Ok(None);
+    };
+    let Some(url) = payload
+        .upstream_url
+        .clone()
+        .filter(|value| !value.trim().is_empty())
+    else {
+        return Ok(None);
+    };
+    let auth_header = payload
+        .auth_header
+        .clone()
+        .filter(|value| !value.trim().is_empty());
+    let auth_value = payload
+        .auth_value
+        .clone()
+        .filter(|value| !value.trim().is_empty());
+    if auth_header.is_some() != auth_value.is_some() {
+        return Ok(None);
+    }
+    let Some(provider_api_format) = payload
+        .provider_api_format
+        .clone()
+        .filter(|value| !value.trim().is_empty())
+    else {
+        return Ok(None);
+    };
+    let Some(client_api_format) = payload
+        .client_api_format
+        .clone()
+        .filter(|value| !value.trim().is_empty())
+    else {
+        return Ok(None);
+    };
+    let Some(provider_request_body_value) = payload.provider_request_body.clone() else {
+        return Ok(None);
+    };
+
+    let mut provider_request_headers = payload.provider_request_headers.clone();
+    if let (Some(auth_header), Some(auth_value)) = (auth_header.as_deref(), auth_value.as_deref()) {
+        ensure_upstream_auth_header(&mut provider_request_headers, auth_header, auth_value);
+    }
+    if payload.upstream_is_stream {
+        provider_request_headers
+            .entry("accept".to_string())
+            .or_insert_with(|| "text/event-stream".to_string());
+    }
+    let plan = ExecutionPlan {
+        request_id,
+        candidate_id: payload.candidate_id.clone(),
+        provider_name: payload.provider_name.clone(),
+        provider_id,
+        endpoint_id,
+        key_id,
+        method: "POST".to_string(),
+        url,
+        headers: std::mem::take(&mut provider_request_headers),
+        content_type: payload
+            .content_type
+            .clone()
+            .or_else(|| Some("application/json".to_string())),
+        content_encoding: None,
+        body: RequestBody::from_json(provider_request_body_value.clone()),
+        stream: payload.upstream_is_stream,
+        client_api_format,
+        provider_api_format,
+        model_name: payload.model_name.clone(),
+        proxy: payload.proxy.clone(),
+        tls_profile: payload.tls_profile.clone(),
+        timeouts: payload.timeouts.clone(),
+    };
+
+    let report_context = augment_sync_report_context(
+        payload.report_context,
+        &plan.headers,
+        &provider_request_body_value,
+    )?;
+
+    Ok(Some(LocalSyncPlanAndReport {
+        plan,
+        report_kind: payload.report_kind,
+        report_context,
+    }))
+}
+
+pub(crate) fn build_gemini_stream_plan_from_decision(
+    _parts: &http::request::Parts,
+    _body_json: &serde_json::Value,
+    payload: GatewayControlSyncDecisionResponse,
+) -> Result<Option<LocalStreamPlanAndReport>, GatewayError> {
+    if generic_decision_missing_exact_provider_request(&payload) {
+        return Ok(None);
+    }
+    let Some(request_id) = payload
+        .request_id
+        .clone()
+        .filter(|value| !value.trim().is_empty())
+    else {
+        return Ok(None);
+    };
+    let Some(provider_id) = payload
+        .provider_id
+        .clone()
+        .filter(|value| !value.trim().is_empty())
+    else {
+        return Ok(None);
+    };
+    let Some(endpoint_id) = payload
+        .endpoint_id
+        .clone()
+        .filter(|value| !value.trim().is_empty())
+    else {
+        return Ok(None);
+    };
+    let Some(key_id) = payload
+        .key_id
+        .clone()
+        .filter(|value| !value.trim().is_empty())
+    else {
+        return Ok(None);
+    };
+    let Some(url) = payload
+        .upstream_url
+        .clone()
+        .filter(|value| !value.trim().is_empty())
+    else {
+        return Ok(None);
+    };
+    let auth_header = payload
+        .auth_header
+        .clone()
+        .filter(|value| !value.trim().is_empty());
+    let auth_value = payload
+        .auth_value
+        .clone()
+        .filter(|value| !value.trim().is_empty());
+    if auth_header.is_some() != auth_value.is_some() {
+        return Ok(None);
+    }
+    let Some(provider_api_format) = payload
+        .provider_api_format
+        .clone()
+        .filter(|value| !value.trim().is_empty())
+    else {
+        return Ok(None);
+    };
+    let Some(client_api_format) = payload
+        .client_api_format
+        .clone()
+        .filter(|value| !value.trim().is_empty())
+    else {
+        return Ok(None);
+    };
+    let Some(provider_request_body_value) = payload.provider_request_body.clone() else {
+        return Ok(None);
+    };
+
+    let mut provider_request_headers = payload.provider_request_headers.clone();
+    if let (Some(auth_header), Some(auth_value)) = (auth_header.as_deref(), auth_value.as_deref()) {
+        ensure_upstream_auth_header(&mut provider_request_headers, auth_header, auth_value);
+    }
+    provider_request_headers.insert("accept".to_string(), "text/event-stream".to_string());
+    let plan = ExecutionPlan {
+        request_id,
+        candidate_id: payload.candidate_id.clone(),
+        provider_name: payload.provider_name.clone(),
+        provider_id,
+        endpoint_id,
+        key_id,
+        method: "POST".to_string(),
+        url,
+        headers: std::mem::take(&mut provider_request_headers),
+        content_type: payload
+            .content_type
+            .clone()
+            .or_else(|| Some("application/json".to_string())),
+        content_encoding: None,
+        body: RequestBody::from_json(provider_request_body_value.clone()),
+        stream: true,
+        client_api_format,
+        provider_api_format,
+        model_name: payload.model_name.clone(),
+        proxy: payload.proxy.clone(),
+        tls_profile: payload.tls_profile.clone(),
+        timeouts: payload.timeouts.clone(),
+    };
+
+    let report_context = augment_sync_report_context(
+        payload.report_context,
+        &plan.headers,
+        &provider_request_body_value,
+    )?;
+
+    Ok(Some(LocalStreamPlanAndReport {
+        plan,
+        report_kind: payload.report_kind,
+        report_context,
+    }))
+}
--- a/Show More
+++ b/Show More