feat: add reset-password CLI command for any user

GORM sqlite AutoMigrate 对已有表判定不收敛（实测已迁移库再跑
仍触发 12 次表重建），SQLite 无在线 ALTER，每次重建走
CREATE __temp + INSERT SELECT + DROP 全表拷贝；usage_logs 含
raw_request/raw_response 大文本，一轮 40~80s，生产每次重启
重演导致启动数分钟。

- 新增 schemaVersion 常量，PRAGMA user_version 记录已迁移版本
- 版本达标跳过 AutoMigrate 秒开；旧库首次升级迁移一次写版本
- postgres 不受影响（支持 ALTER，保持原行为）
- 维护约定：改表结构须递增 schemaVersion

.dockerignore

@@ -0,0 +1,20 @@
+# VCS
+.git
+.gitignore
+
+# 依赖与构建产物（镜像内重新安装/构建）
+**/node_modules
+**/dist
+server/bin
+
+# 本地数据与密钥
+data/
+*.db
+*.log
+.env
+.env.*
+
+# 编辑器/系统
+.DS_Store
+.idea/
+.vscode/

.env.example

@@ -27,12 +27,15 @@ OT_MASTER_KEY=change-me-master-key
 OT_PROXY_UPSTREAM_KEY=
 OT_PROXY_UPSTREAM_BASE_URL=https://api.openai.com
 OT_PROXY_DEFAULT_MODEL=gpt-4o-mini
-OT_PROXY_TIMEOUT=120s
+OT_PROXY_TIMEOUT=300s
 
 # 渠道健康检查
 OT_PROXY_HEALTH_INTERVAL=60s
 OT_PROXY_HEALTH_FAIL_THRESHOLD=2
 
+# 调试：记录管理员的原始请求与响应到请求明细（默认关闭；流式记录全部事件）
+OT_PROXY_LOG_RAW=false
+
 # 限流（内存计数，Redis 后置）：用户级每秒请求数上限（0=不限制）
 OT_RATELIMIT_USER_RPS=20
 
@@ -45,3 +48,6 @@ OT_WEBAUTHN_RP_NAME=openteam
 OT_ADMIN_USERNAME=admin
 OT_ADMIN_EMAIL=admin@localhost
 OT_ADMIN_PASSWORD=admin123
+
+# 记录原始请求日志
+OT_PROXY_LOG_RAW=false

.gitignore

@@ -7,6 +7,8 @@ dist/
 # Go
 server/bin/
 server/data/
+# 本地静态托管软链（指向 ../web/dist），不入库
+server/web/
 scripts/mockupstream/bin/
 
 # TypeScript 增量构建产物

Dockerfile

@@ -0,0 +1,53 @@
+# openteam 多架构镜像（linux/amd64 + linux/arm64）
+#
+# 构建并推送：
+#   docker buildx build --platform linux/amd64,linux/arm64 -t openteam:latest --push .
+# 本地单架构验证：
+#   docker build -t openteam:test .
+#
+# 前端产物 web/dist 是平台无关的纯静态文件，只编译一次。
+# 关键在 web-builder 阶段用 --platform=$BUILDPLATFORM 固定为本机构建平台：
+# buildx 做多架构时该阶段只在本机跑一次，两个架构的 Go 阶段分别 --from 拷贝同一份 dist。
+
+# ---------- 阶段 1：前端构建（仅一次，产物平台无关） ----------
+FROM --platform=$BUILDPLATFORM node:24-alpine AS web-builder
+WORKDIR /src
+# corepack 固定 pnpm 版本，配合 pnpm-lock.yaml 可复现安装。
+# --ignore-scripts：依赖 postinstall（esbuild/vue-demi）非必需且被 pnpm 拦截，
+# esbuild 走 optionalDependencies 自带平台二进制；跳过脚本更安全、可复现。
+RUN corepack enable && corepack prepare pnpm@11.21.0 --activate
+COPY web/package.json web/pnpm-lock.yaml ./
+RUN pnpm install --frozen-lockfile --ignore-scripts
+COPY web/ ./
+RUN pnpm build
+
+# ---------- 阶段 2：Go 后端编译（每个目标架构各执行一次） ----------
+FROM golang:1.26-alpine AS go-builder
+ARG TARGETOS
+ARG TARGETARCH
+WORKDIR /src
+# 先拷依赖清单再装依赖，利用层缓存
+COPY server/go.mod server/go.sum ./
+RUN go mod download
+COPY server/ ./
+RUN CGO_ENABLED=0 GOOS=$TARGETOS GOARCH=$TARGETARCH \
+    go build -trimpath -ldflags="-s -w" -o /out/openteam ./cmd/server
+
+# ---------- 阶段 3：运行镜像 ----------
+FROM alpine:3.21
+RUN apk add --no-cache ca-certificates tzdata su-exec
+WORKDIR /app
+COPY --from=go-builder /out/openteam ./openteam
+# 与二进制同目录放置，符合服务 CWD 约定（router 按相对路径 web/dist 找静态资源）
+COPY --from=web-builder /src/dist ./web/dist
+# 入口脚本以 root 启动，chown 数据目录后 su-exec 降权为 OT_UID:OT_GID（默认 1000:1000）
+COPY docker/entrypoint.sh /entrypoint.sh
+RUN chmod +x /entrypoint.sh && mkdir -p /app/data
+ENV OT_ENV=production \
+    OT_PORT=8080 \
+    OT_DB_DRIVER=sqlite \
+    OT_DB_DSN=data/openteam.db
+EXPOSE 8080
+VOLUME ["/app/data"]
+ENTRYPOINT ["/entrypoint.sh"]
+CMD ["./openteam"]

Makefile

@@ -1,8 +1,14 @@
-.PHONY: run build test tidy mock-upstream web-dev web-build
+.PHONY: run build test tidy mock-upstream web-dev web-build reset-admin-password reset-password
 
 run:
 	cd server && go run ./cmd/server
 
+reset-admin-password:
+	cd server && go run ./cmd/server reset-admin-password
+
+reset-password:
+	cd server && go run ./cmd/server reset-password
+
 build:
 	cd server && go build -o bin/openteam ./cmd/server
 

PLANNING.md

@@ -275,8 +275,8 @@ Anthropic /v1/messages    ──┘
 
 #### 5.3.3 API Key
 
-- 格式：`sk-` + 48 位随机 base62，**创建时仅展示一次**。
-- 存储：仅 SHA-256 哈希 + 展示前缀（如 `sk-aB3c…`）；请求时哈希后查表。
+- 格式：`sk-ot-` + 48 位随机 base62，**创建时仅展示一次**。
+- 存储：仅 SHA-256 哈希 + 展示前缀（如 `sk-ot-aB3c…`）；请求时哈希后查表。
 - 附加能力：密钥级配额（每日 token / 请求数）、模型白名单、过期时间、启停。
 - 限额检查用 Redis 计数，与用户级限流叠加。
 

README.md

@@ -17,7 +17,7 @@
 - **限流/配额**（内存计数）：密钥级每日请求数 / Token 数配额、用户级每秒速率（`OT_RATELIMIT_USER_RPS`）；超限返回 429
 - **前端**：Dashboard / 管理总览骨架屏加载态
 - **用户体系**：注册（开放/邀请码可切换，管理后台可改）、登录（JWT access + HttpOnly refresh cookie）、argon2id 密码
-- **API Key**：`sk-` 48 位 base62，仅存 SHA-256 哈希，明文一次性展示；支持限额/过期/白名单字段
+- **API Key**：`sk-ot-` 48 位 base62，仅存 SHA-256 哈希，明文一次性展示；支持限额/过期/白名单字段
 - **用量计费**：请求级 `usage_logs` 异步批量落库，按模型价格扣减余额，日粒度预聚合（`usage_daily`）
 - **管理 API**：渠道 CRUD + 连通测试 + 模型导入、模型管理 + 定价 + 渠道绑定、用户管理、全局用量/统计、系统配置
 - **前端**（Vue3 + Tailwind，taste-skill 设计，深色优先）

docker-compose.yml

@@ -0,0 +1,40 @@
+# openteam 自托管部署
+#
+# 启动：docker compose up -d
+# 首次会本地构建镜像（见 Dockerfile）；若已用 build-image.sh 推到仓库，
+# 把 image 改为仓库地址并删掉 build 段即可拉取多架构镜像。
+#
+# 密钥从仓库根 .env 读取（与本地开发共用），缺失时 compose 直接报错。
+name: openteam
+
+services:
+  openteam:
+    build:
+      context: .
+      dockerfile: Dockerfile
+    image: openteam:latest
+    container_name: openteam
+    restart: unless-stopped
+    ports:
+      - "${HOST_PORT:-8080}:8080"
+    environment:
+      OT_ENV: production
+      OT_PORT: 8080
+      # sqlite 落盘到当前目录 ./data（容器 WORKDIR=/app，DSN 用相对路径）
+      OT_DB_DRIVER: sqlite
+      OT_DB_DSN: data/openteam.db
+      # 必填密钥：渠道加密主密钥 / 管理员密码 / JWT 签名
+      OT_MASTER_KEY: ${OT_MASTER_KEY:?请在 .env 中设置 OT_MASTER_KEY}
+      OT_ADMIN_PASSWORD: ${OT_ADMIN_PASSWORD:?请在 .env 中设置 OT_ADMIN_PASSWORD}
+      OT_JWT_SECRET: ${OT_JWT_SECRET:?请在 .env 中设置 OT_JWT_SECRET}
+      # 入口脚本据此 chown 数据目录并降权运行（默认 1000:1000）
+      OT_UID: "${OT_UID:-1000}"
+      OT_GID: "${OT_GID:-1000}"
+    volumes:
+      - ./data:/app/data
+    healthcheck:
+      test: ["CMD", "wget", "-qO-", "http://127.0.0.1:8080/healthz"]
+      interval: 30s
+      timeout: 5s
+      retries: 3
+      start_period: 10s

docker/entrypoint.sh

@@ -0,0 +1,12 @@
+#!/bin/sh
+# openteam 容器入口：以 root 短暂启动，修正数据目录属主后降权执行服务。
+# 解决绑定挂载/命名卷被 docker 以 root 自动创建、应用非 root 用户写不进 db 的问题。
+set -e
+
+PUID="${OT_UID:-1000}"
+PGID="${OT_GID:-1000}"
+
+mkdir -p /app/data
+chown -R "${PUID}:${PGID}" /app/data
+
+exec su-exec "${PUID}:${PGID}" "$@"

scripts/build-image.sh

@@ -0,0 +1,72 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+DOCKER_USER="${DOCKER_USER:-openteam}"
+VERSION="${VERSION:-$(git describe --tags --always 2>/dev/null || echo latest)}"
+PLATFORMS="linux/amd64,linux/arm64"
+
+# --local：本地构建不推送（默认 docker driver，单架构，tag 不带仓库前缀）。
+# 不用 buildx docker-container builder——嵌套创建容器在受限环境会报
+# "can't mask dir /proc/acpi"，普通 docker build 无此问题。
+if [[ "${1:-}" == "--local" ]]; then
+  LOCAL=1
+else
+  LOCAL=0
+fi
+if [[ $# -gt 1 || ( $# -eq 1 && "$1" != "--local" ) ]]; then
+  echo "用法: $0 [--local]" >&2
+  exit 1
+fi
+
+native_arch() {
+  case "$(uname -m)" in
+    x86_64|amd64) echo "linux/amd64" ;;
+    aarch64|arm64) echo "linux/arm64" ;;
+    *) echo "linux/$(uname -m)" ;;
+  esac
+}
+
+# ---------- 本地模式：单架构，加载到本地 docker ----------
+if [[ $LOCAL -eq 1 ]]; then
+  PLATFORM="$(native_arch)"
+  echo ""
+  echo "========================================"
+  echo "🚀 本地构建 openteam  (${PLATFORM})"
+  echo "   tag: openteam:${VERSION}"
+  echo "========================================"
+  docker build \
+    -t "openteam:${VERSION}" \
+    -t "openteam:latest" \
+    .
+  echo ""
+  echo "✅ 完成！openteam:${VERSION}（已加载到本地 docker）"
+  exit 0
+fi
+
+# ---------- 多架构推送模式：需要 buildx docker-container builder ----------
+if ! docker buildx version &>/dev/null; then
+  echo "❌ docker buildx 不可用"
+  exit 1
+fi
+
+if ! docker buildx inspect multiarch &>/dev/null 2>&1; then
+  echo "🔧 创建 multiarch builder..."
+  docker buildx create --name multiarch --driver docker-container --use
+else
+  docker buildx use multiarch
+fi
+docker buildx inspect --bootstrap
+
+echo ""
+echo "========================================"
+echo "🚀 构建 openteam  (${PLATFORMS})"
+echo "   tag: ${DOCKER_USER}/openteam:${VERSION}"
+echo "========================================"
+docker buildx build \
+  --platform "${PLATFORMS}" \
+  -t "${DOCKER_USER}/openteam:${VERSION}" \
+  -t "${DOCKER_USER}/openteam:latest" \
+  --push \
+  .
+echo ""
+echo "✅ 完成！${DOCKER_USER}/openteam:${VERSION}"

server/cmd/server/main.go

@@ -2,60 +2,8 @@
 // 启动：OT_PROXY_UPSTREAM_KEY=sk-xxx go run ./cmd/server
 package main
 
-import (
-	"context"
-	"errors"
-	"log"
-	"net/http"
-	"os"
-	"os/signal"
-	"strconv"
-	"syscall"
-	"time"
-
-	"github.com/openteam/server/internal/api"
-	"github.com/openteam/server/internal/app"
-	"github.com/openteam/server/internal/config"
-	"github.com/openteam/server/internal/proxy"
-)
+import "github.com/openteam/server/internal/cli"
 
 func main() {
-	cfg, err := config.Load()
-	if err != nil {
-		log.Fatalf("config: %v", err)
-	}
-
-	a, err := app.New(cfg)
-	if err != nil {
-		log.Fatalf("app init: %v", err)
-	}
-	defer a.Shutdown(context.Background())
-
-	gw := proxy.NewGateway(a.DB, a.Enc, a.Usage, a.Limit, cfg.RateLimit.UserRPS)
-	router := api.NewRouter(a, gw)
-
-	srv := &http.Server{
-		Addr:              ":" + strconv.Itoa(cfg.Port),
-		Handler:           router,
-		ReadHeaderTimeout: 10 * time.Second,
-	}
-
-	go func() {
-		log.Printf("openteam listening on %s (env=%s)", srv.Addr, cfg.Env)
-		if err := srv.ListenAndServe(); err != nil && !errors.Is(err, http.ErrServerClosed) {
-			log.Fatalf("server: %v", err)
-		}
-	}()
-
-	quit := make(chan os.Signal, 1)
-	signal.Notify(quit, syscall.SIGINT, syscall.SIGTERM)
-	<-quit
-
-	log.Println("shutting down...")
-	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
-	defer cancel()
-	if err := srv.Shutdown(ctx); err != nil {
-		log.Printf("server shutdown: %v", err)
-	}
-	a.Usage.Close()
+	cli.Execute()
 }

server/go.mod

@@ -7,9 +7,11 @@ require (
 	github.com/glebarez/sqlite v1.11.0
 	github.com/go-webauthn/webauthn v0.17.4
 	github.com/golang-jwt/jwt/v5 v5.3.1
+	github.com/spf13/cobra v1.10.2
 	github.com/spf13/viper v1.21.0
 	github.com/tiktoken-go/tokenizer v0.8.1
 	golang.org/x/crypto v0.55.0
+	golang.org/x/term v0.45.0
 	gorm.io/driver/postgres v1.6.2
 	gorm.io/gorm v1.31.2
 )
@@ -35,6 +37,7 @@ require (
 	github.com/goccy/go-yaml v1.19.2 // indirect
 	github.com/google/go-tpm v0.9.8 // indirect
 	github.com/google/uuid v1.6.0 // indirect
+	github.com/inconshreveable/mousetrap v1.1.0 // indirect
 	github.com/jackc/pgpassfile v1.0.0 // indirect
 	github.com/jackc/pgservicefile v0.0.0-20240606120523-5a60cdf6a761 // indirect
 	github.com/jackc/pgx/v5 v5.10.0 // indirect

server/go.sum

@@ -6,6 +6,7 @@ github.com/bytedance/sonic/loader v0.5.0 h1:gXH3KVnatgY7loH5/TkeVyXPfESoqSBSBEiD
 github.com/bytedance/sonic/loader v0.5.0/go.mod h1:AR4NYCk5DdzZizZ5djGqQ92eEhCCcdf5x77udYiSJRo=
 github.com/cloudwego/base64x v0.1.6 h1:t11wG9AECkCDk5fMSoxmufanudBtJ+/HemLstXDLI2M=
 github.com/cloudwego/base64x v0.1.6/go.mod h1:OFcloc187FXDaYHvrNIjxSe8ncn0OOM8gEHfghB2IPU=
+github.com/cpuguy83/go-md2man/v2 v2.0.6/go.mod h1:oOW0eioCTA6cOiMLiUPZOpcVxMig6NIQQ7OS05n1F4g=
 github.com/davecgh/go-spew v1.1.0/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
 github.com/davecgh/go-spew v1.1.1 h1:vj9j/u1bqnvCEfJOwUhtlOARqs3+rkHYY13jYWTU97c=
 github.com/davecgh/go-spew v1.1.1/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
@@ -60,6 +61,8 @@ github.com/google/pprof v0.0.0-20221118152302-e6195bd50e26 h1:Xim43kblpZXfIBQsbu
 github.com/google/pprof v0.0.0-20221118152302-e6195bd50e26/go.mod h1:dDKJzRmX4S37WGHujM7tX//fmj1uioxKzKxz3lo4HJo=
 github.com/google/uuid v1.6.0 h1:NIvaJDMOsjHA8n1jAhLSgzrAzy1Hgr+hNrb57e+94F0=
 github.com/google/uuid v1.6.0/go.mod h1:TIyPZe4MgqvfeYDBFedMoGGpEw/LqOeaOT+nhxU+yHo=
+github.com/inconshreveable/mousetrap v1.1.0 h1:wN+x4NVGpMsO7ErUn/mUI3vEoE6Jt13X2s0bqwp9tc8=
+github.com/inconshreveable/mousetrap v1.1.0/go.mod h1:vpF70FUmC8bwa3OWnCshd2FqLfsEA9PFc4w1p2J65bw=
 github.com/jackc/pgpassfile v1.0.0 h1:/6Hmqy13Ss2zCq62VdNG8tM1wchn8zjSGOBJ6icpsIM=
 github.com/jackc/pgpassfile v1.0.0/go.mod h1:CEx0iS5ambNFdcRtxPj5JhEz+xB6uRky5eyVu/W2HEg=
 github.com/jackc/pgservicefile v0.0.0-20240606120523-5a60cdf6a761 h1:iCEnooe7UlwOQYpKFhBabPMi4aNAfoODPEFNiAnClxo=
@@ -106,6 +109,7 @@ github.com/remyoudompheng/bigfft v0.0.0-20230129092748-24d4a6f8daec h1:W09IVJc94
 github.com/remyoudompheng/bigfft v0.0.0-20230129092748-24d4a6f8daec/go.mod h1:qqbHyh8v60DhA7CoWK5oRCqLrMHRGoxYCSS9EjAz6Eo=
 github.com/rogpeppe/go-internal v1.10.0 h1:TMyTOH3F/DB16zRVcYyreMH6GnZZrwQVAoYjRBZyWFQ=
 github.com/rogpeppe/go-internal v1.10.0/go.mod h1:UQnix2H7Ngw/k4C5ijL5+65zddjncjaFoBhdsK/akog=
+github.com/russross/blackfriday/v2 v2.1.0/go.mod h1:+Rmxgy9KzJVeS9/2gXHxylqXiyQDYRxCVz55jmeOWTM=
 github.com/sagikazarmark/locafero v0.11.0 h1:1iurJgmM9G3PA/I+wWYIOw/5SyBtxapeHDcg+AAIFXc=
 github.com/sagikazarmark/locafero v0.11.0/go.mod h1:nVIGvgyzw595SUSUE6tvCp3YYTeHs15MvlmU87WwIik=
 github.com/sourcegraph/conc v0.3.1-0.20240121214520-5f936abd7ae8 h1:+jumHNA0Wrelhe64i8F6HNlS8pkoyMv5sreGx2Ry5Rw=
@@ -114,6 +118,9 @@ github.com/spf13/afero v1.15.0 h1:b/YBCLWAJdFWJTN9cLhiXXcD7mzKn9Dm86dNnfyQw1I=
 github.com/spf13/afero v1.15.0/go.mod h1:NC2ByUVxtQs4b3sIUphxK0NioZnmxgyCrfzeuq8lxMg=
 github.com/spf13/cast v1.10.0 h1:h2x0u2shc1QuLHfxi+cTJvs30+ZAHOGRic8uyGTDWxY=
 github.com/spf13/cast v1.10.0/go.mod h1:jNfB8QC9IA6ZuY2ZjDp0KtFO2LZZlg4S/7bzP6qqeHo=
+github.com/spf13/cobra v1.10.2 h1:DMTTonx5m65Ic0GOoRY2c16WCbHxOOw6xxezuLaBpcU=
+github.com/spf13/cobra v1.10.2/go.mod h1:7C1pvHqHw5A4vrJfjNwvOdzYu0Gml16OCs2GRiTUUS4=
+github.com/spf13/pflag v1.0.9/go.mod h1:McXfInJRrz4CZXVZOBLb0bTZqETkiAhM9Iw0y3An2Bg=
 github.com/spf13/pflag v1.0.10 h1:4EBh2KAYBwaONj6b2Ye1GiHfwjqyROoF4RwYO+vPwFk=
 github.com/spf13/pflag v1.0.10/go.mod h1:McXfInJRrz4CZXVZOBLb0bTZqETkiAhM9Iw0y3An2Bg=
 github.com/spf13/viper v1.21.0 h1:x5S+0EU27Lbphp4UKm1C+1oQO+rKx36vfCoaVebLFSU=
@@ -159,6 +166,8 @@ golang.org/x/sync v0.22.0/go.mod h1:9xrNwdLfx4jkKbNva9FpL6vEN7evnE43NNNJQ2LF3+0=
 golang.org/x/sys v0.6.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.47.0 h1:o7XGOvZQCADBQQ4Y7VNq2dRWQR7JmOUW8Kxx4ZsNgWs=
 golang.org/x/sys v0.47.0/go.mod h1:4GL1E5IUh+htKOUEOaiffhrAeqysfVGipDYzABqnCmw=
+golang.org/x/term v0.45.0 h1:NwWyBmoJCbfTHpxrWoZ9C6/VxOf7ic219I8xZZFdrf0=
+golang.org/x/term v0.45.0/go.mod h1:9aqxs0blBcrm/n0L9QW0aRVD+ktan8ssZromtqJC43w=
 golang.org/x/text v0.41.0 h1:vz/seA0lnX87Othu2f/0L24RcgrXD9/YFTSuGjj3rH8=
 golang.org/x/text v0.41.0/go.mod h1:jvf1O8ajNzZqhSrQBPbutR/EB83Cc0CFrezNQIwbb5M=
 google.golang.org/protobuf v1.36.10 h1:AYd7cD/uASjIL6Q9LiTjz8JLcrh/88q5UObnmY3aOOE=

server/internal/api/admin_channel_models.go

@@ -118,7 +118,7 @@ func (h *Handler) AdminChannelAddModel(c *gin.Context) {
 	// 解析或创建全局模型（客户端名）
 	var m store.Model
 	if err := h.a.DB.Where("name = ?", globalName).First(&m).Error; err != nil {
-		m = store.Model{Name: globalName, DisplayName: globalName, Enabled: true}
+		m = store.Model{Name: globalName, Enabled: true}
 		if err := h.a.DB.Create(&m).Error; err != nil {
 			resp.Fail(c, http.StatusInternalServerError, "failed to create model")
 			return

server/internal/api/admin_channels.go

@@ -184,7 +184,7 @@ func (h *Handler) AdminCreateChannel(c *gin.Context) {
 		Name: req.Name, Provider: req.Provider, Formats: formats, BaseURL: baseURL,
 		BaseURLs: normalizeBaseURLs(req.BaseURLs),
 		APIKeyEnc: enc, Weight: intOr(req.Weight, 1), Priority: intOr(req.Priority, 0),
-		TimeoutMS: intOr(req.TimeoutMS, 120000), MaxConcurrency: intOr(req.MaxConcurrency, 16),
+		TimeoutMS: intOr(req.TimeoutMS, 300000), MaxConcurrency: intOr(req.MaxConcurrency, 16),
 		HealthStatus: store.ChannelHealthHealthy, Enabled: boolOr(req.Enabled, true),
 	}
 	if err := h.a.DB.Create(&ch).Error; err != nil {

server/internal/api/admin_models.go

@@ -29,16 +29,19 @@ func (h *Handler) AdminModels(c *gin.Context) {
 		h.a.DB.Preload("Channel").Where("model_id = ?", m.ID).Find(&bindings)
 		chs := make([]gin.H, 0, len(bindings))
 		for _, b := range bindings {
+			if !b.Channel.Enabled {
+				continue
+			}
 			chs = append(chs, gin.H{
 				"id": b.ID, "channel_id": b.ChannelID, "channel_name": b.Channel.Name,
 				"upstream_model": b.UpstreamModel, "weight": b.Weight,
 			})
 		}
-		used := len(bindings) > 0
+		used := len(chs) > 0
 		needsPricing := used && m.InputPrice == 0 && m.OutputPrice == 0 && m.CacheReadPrice == 0
 		denied := containsStr(deny, m.Name) || (len(allow) > 0 && !containsStr(allow, m.Name))
 		out = append(out, gin.H{
-			"id": m.ID, "name": m.Name, "display_name": m.DisplayName,
+			"id": m.ID, "name": m.Name,
 			"input_price": m.InputPrice, "output_price": m.OutputPrice, "cache_read_price": m.CacheReadPrice,
 			"enabled": m.Enabled, "sort": m.Sort, "channels": chs,
 			"used": used, "needs_pricing": needsPricing, "denied": denied,
@@ -116,7 +119,6 @@ func containsStr(list []string, s string) bool {
 func (h *Handler) AdminCreateModel(c *gin.Context) {
 	var req struct {
 		Name           string  `json:"name" binding:"required,min=1,max=128"`
-		DisplayName    string  `json:"display_name"`
 		InputPrice     float64 `json:"input_price"`
 		OutputPrice    float64 `json:"output_price"`
 		CacheReadPrice float64 `json:"cache_read_price"`
@@ -127,13 +129,10 @@ func (h *Handler) AdminCreateModel(c *gin.Context) {
 		return
 	}
 	m := store.Model{
-		Name: req.Name, DisplayName: req.DisplayName,
+		Name: req.Name,
 		InputPrice: req.InputPrice, OutputPrice: req.OutputPrice, CacheReadPrice: req.CacheReadPrice,
 		Enabled: boolOr(req.Enabled, true),
 	}
-	if m.DisplayName == "" {
-		m.DisplayName = m.Name
-	}
 	if err := h.a.DB.Create(&m).Error; err != nil {
 		resp.Fail(c, http.StatusConflict, "failed to create model (name may already exist)")
 		return
@@ -141,7 +140,7 @@ func (h *Handler) AdminCreateModel(c *gin.Context) {
 	resp.Created(c, gin.H{"id": m.ID, "name": m.Name})
 }
 
-// AdminUpdateModel PUT /api/v1/admin/models/:id — 价格/展示名/启停/排序。
+// AdminUpdateModel PUT /api/v1/admin/models/:id — 价格/启停/排序。
 func (h *Handler) AdminUpdateModel(c *gin.Context) {
 	id, err := strconv.ParseUint(c.Param("id"), 10, 64)
 	if err != nil {
@@ -149,7 +148,6 @@ func (h *Handler) AdminUpdateModel(c *gin.Context) {
 		return
 	}
 	var req struct {
-		DisplayName    *string  `json:"display_name"`
 		InputPrice     *float64 `json:"input_price"`
 		OutputPrice    *float64 `json:"output_price"`
 		CacheReadPrice *float64 `json:"cache_read_price"`
@@ -166,9 +164,6 @@ func (h *Handler) AdminUpdateModel(c *gin.Context) {
 		return
 	}
 	updates := map[string]any{}
-	if req.DisplayName != nil {
-		updates["display_name"] = *req.DisplayName
-	}
 	if req.InputPrice != nil {
 		updates["input_price"] = *req.InputPrice
 	}

server/internal/api/admin_stats.go

@@ -90,11 +90,17 @@ func (h *Handler) AdminUsage(c *gin.Context) {
 	for _, l := range logs {
 		var uname string
 		h.a.DB.Model(&store.User{}).Where("id = ?", l.UserID).Pluck("username", &uname)
+		var chName string
+		if l.ChannelID > 0 {
+			h.a.DB.Model(&store.Channel{}).Where("id = ?", l.ChannelID).Pluck("name", &chName)
+		}
 		out = append(out, gin.H{
 			"id": l.ID, "user": uname, "user_id": l.UserID, "model": l.ModelName, "protocol": l.Protocol,
+			"channel_id": l.ChannelID, "channel": chName,
 			"input_tokens": l.InputTokens, "output_tokens": l.OutputTokens,
 			"cache_read_tokens": l.CacheReadTokens, "cost": l.Cost,
 			"latency_ms": l.LatencyMS, "status": l.Status, "error_code": l.ErrorCode,
+			"raw_request": l.RawRequest, "raw_response": l.RawResponse,
 			"created_at": l.CreatedAt,
 		})
 	}

server/internal/api/router.go

@@ -34,7 +34,13 @@ func NewRouter(a *app.App, gw *proxy.Gateway) *gin.Engine {
 	// 静态资源（前端构建产物，存在时托管）
 	const dist = "web/dist"
 	if _, err := os.Stat(dist); err == nil {
-		r.Static("/assets", dist+"/assets")
+		// /assets 文件名含内容 hash，不可变：长缓存 + immutable，避免每次回源
+		assets := r.Group("/assets")
+		assets.Use(func(c *gin.Context) {
+			c.Header("Cache-Control", "public, max-age=31536000, immutable")
+			c.Next()
+		})
+		assets.Static("", dist+"/assets")
 		r.StaticFile("/favicon.svg", dist+"/favicon.svg")
 	}
 
@@ -48,6 +54,8 @@ func NewRouter(a *app.App, gw *proxy.Gateway) *gin.Engine {
 			return
 		}
 		if _, err := os.Stat(dist); err == nil {
+			// SPA 入口：必须每次回源校验，否则浏览器启发式缓存会保留引用已删除 chunk 的旧 HTML → 黑屏
+			c.Header("Cache-Control", "no-cache")
 			c.File(dist + "/index.html")
 			return
 		}

server/internal/api/user.go

@@ -49,15 +49,22 @@ func (h *Handler) todayUsage(c *gin.Context, userID uint64) gin.H {
 }
 
 // UserModels GET /api/v1/user/models — 控制台可用模型列表（无需 API Key）。
+// 仅返回启用的模型且至少绑定到一个启用且健康的渠道，与 /v1/models 口径一致。
 func (h *Handler) UserModels(c *gin.Context) {
-	var ms []store.Model
-	if err := h.a.DB.Where("enabled = ?", true).Order("sort ASC, id ASC").Find(&ms).Error; err != nil {
+	var names []string
+	if err := h.a.DB.Table("models").
+		Joins("JOIN channel_model_bindings ON channel_model_bindings.model_id = models.id").
+		Joins("JOIN channels ON channels.id = channel_model_bindings.channel_id").
+		Where("models.enabled = ? AND channels.enabled = ? AND channels.health_status = ?",
+			true, true, store.ChannelHealthHealthy).
+		Distinct("models.name").
+		Order("models.sort ASC, models.id ASC").
+		Pluck("models.name", &names).Error; err != nil {
 		resp.Fail(c, http.StatusInternalServerError, "failed to load models")
 		return
 	}
-	out := make([]string, 0, len(ms))
-	for _, m := range ms {
-		out = append(out, m.Name)
+	if names == nil {
+		names = []string{}
 	}
-	resp.OK(c, gin.H{"items": out})
+	resp.OK(c, gin.H{"items": names})
 }

server/internal/app/app.go

@@ -125,7 +125,6 @@ func (a *App) Seed() error {
 			// 默认模型 + 绑定
 			m := store.Model{
 				Name:       a.Cfg.Proxy.DefaultModel,
-				DisplayName: a.Cfg.Proxy.DefaultModel,
 				InputPrice: 0.15, // 每百万 token，示例价
 				OutputPrice: 0.60,
 				Enabled:    true,

server/internal/channel/channel.go

@@ -87,6 +87,20 @@ func (s *Service) loadBound(bindings []store.ChannelModelBinding) []Candidate {
 	return out
 }
 
+// AvailableModelIDs 返回对外可见的模型 ID：启用的模型且至少绑定到一个启用且健康的渠道。
+// 与 Candidates 的过滤口径一致（enabled + health_status=healthy），避免暴露绑定到已停用渠道的模型。
+func (s *Service) AvailableModelIDs() []uint64 {
+	var ids []uint64
+	s.db.Model(&store.ChannelModelBinding{}).
+		Joins("JOIN channels ON channels.id = channel_model_bindings.channel_id").
+		Joins("JOIN models ON models.id = channel_model_bindings.model_id").
+		Where("channels.enabled = ? AND channels.health_status = ?", true, store.ChannelHealthHealthy).
+		Where("models.enabled = ?", true).
+		Distinct("channel_model_bindings.model_id").
+		Pluck("channel_model_bindings.model_id", &ids)
+	return ids
+}
+
 // Pick 按权重加权随机选一个候选渠道（负载均衡）。
 func (s *Service) Pick(cands []Candidate) *store.Channel {
 	if len(cands) == 0 {

server/internal/cli/reset_admin.go

@@ -0,0 +1,178 @@
+// Package cli 命令行入口。
+package cli
+
+import (
+	"crypto/rand"
+	"fmt"
+	"log"
+	"math/big"
+	"os"
+	"strings"
+
+	"github.com/openteam/server/internal/config"
+	"github.com/openteam/server/internal/pkg/crypto"
+	"github.com/openteam/server/internal/store"
+	"github.com/spf13/cobra"
+	"golang.org/x/term"
+	"gorm.io/gorm"
+)
+
+// pwAlphabet 密码随机串字符集（去掉 0/O/1/l/I 等易混淆字符）。
+const pwAlphabet = "23456789ABCDEFGHJKLMNPQRSTUVWXYZabcdefghijkmnpqrstuvwxyz"
+
+var (
+	resetUser     string
+	resetPassword string
+)
+
+var resetAdminPasswordCmd = &cobra.Command{
+	Use:   "reset-admin-password",
+	Short: "重置管理员密码",
+	Long: `重置管理员密码（忘记密码时用，需在可加载配置的目录下执行）。
+
+交互式引导（默认）：
+  go run ./cmd/server reset-admin-password
+
+非交互式（用于脚本）：
+  go run ./cmd/server reset-admin-password --user admin --password "新密码"`,
+	Run: func(cmd *cobra.Command, args []string) {
+		runResetAdminPassword()
+	},
+}
+
+func init() {
+	rootCmd.AddCommand(resetAdminPasswordCmd)
+	resetAdminPasswordCmd.Flags().StringVar(&resetUser, "user", "", "目标管理员用户名（留空则交互选择）")
+	resetAdminPasswordCmd.Flags().StringVar(&resetPassword, "password", "", "新密码，至少 8 位（留空则交互输入）")
+}
+
+func runResetAdminPassword() {
+	// --- 阶段 1：加载配置与数据库 ---
+	cfg, err := config.Load()
+	if err != nil {
+		log.Fatalf("加载配置失败: %v", err)
+	}
+	db, err := store.Open(cfg.DB.Driver, cfg.DB.DSN)
+	if err != nil {
+		log.Fatalf("连接数据库失败: %v", err)
+	}
+
+	// --- 阶段 2：列出管理员 ---
+	var admins []struct {
+		ID       uint64
+		Username string
+	}
+	db.Model(&store.User{}).Select("id, username").Where("role = ?", store.RoleAdmin).Find(&admins)
+	if len(admins) == 0 {
+		log.Fatalf("系统中没有管理员用户")
+	}
+
+	// --- 阶段 3：选择目标管理员 ---
+	target := targetAdmin(db, admins)
+
+	// --- 阶段 4：获取新密码 ---
+	var pw string
+	if resetPassword != "" {
+		pw = resetPassword
+	} else {
+		pw = promptPassword()
+	}
+	if pw == "" {
+		log.Fatalf("密码不能为空")
+	}
+	if len(pw) < 8 || len(pw) > 72 {
+		log.Fatalf("密码长度必须在 8 到 72 位之间")
+	}
+
+	// --- 阶段 5：加密并更新 ---
+	hasher := crypto.NewPasswordHasher(cfg.Auth.Argon2Time, cfg.Auth.Argon2Memory, cfg.Auth.Argon2Threads, cfg.Auth.Argon2KeyLen, cfg.Auth.SaltLen)
+	hash, err := hasher.HashPassword(pw)
+	if err != nil {
+		log.Fatalf("加密密码失败: %v", err)
+	}
+	if err := db.Model(&store.User{}).Where("id = ?", target.ID).Update("password_hash", hash).Error; err != nil {
+		log.Fatalf("更新密码失败: %v", err)
+	}
+
+	fmt.Printf("✅ 已重置管理员 %q 的密码！\n", target.Username)
+}
+
+// targetAdmin 解析 --user 或交互选择目标管理员。
+func targetAdmin(db *gorm.DB, admins []struct {
+	ID       uint64
+	Username string
+}) *store.User {
+	if resetUser != "" {
+		for i := range admins {
+			if admins[i].Username == resetUser {
+				return &store.User{ID: admins[i].ID, Username: admins[i].Username}
+			}
+		}
+		log.Fatalf("管理员 %q 不存在", resetUser)
+	}
+	if len(admins) == 1 {
+		return &store.User{ID: admins[0].ID, Username: admins[0].Username}
+	}
+	return selectAdmin(admins)
+}
+
+// selectAdmin 交互式选择管理员用户。
+func selectAdmin(admins []struct {
+	ID       uint64
+	Username string
+}) *store.User {
+	fmt.Println("检测到多个管理员用户，请选择要重置密码的目标：")
+	for i, a := range admins {
+		fmt.Printf("  [%d] %s\n", i+1, a.Username)
+	}
+	fmt.Printf("请输入编号 (1-%d): ", len(admins))
+	var n int
+	if _, err := fmt.Scanf("%d", &n); err != nil || n < 1 || n > len(admins) {
+		log.Fatalf("无效输入")
+	}
+	return &store.User{ID: admins[n-1].ID, Username: admins[n-1].Username}
+}
+
+// promptPassword 交互式输入密码，并提供两种方式选择。
+func promptPassword() string {
+	fmt.Println("请选择密码设置方式：")
+	fmt.Println("  [1] 手动输入密码")
+	fmt.Println("  [2] 自动生成随机密码（16 位，不含易混淆字符）")
+	fmt.Printf("请输入编号 (1-2): ")
+
+	var n int
+	if _, err := fmt.Scanf("%d", &n); err != nil || n < 1 || n > 2 {
+		log.Fatalf("无效输入，请输入 1 或 2")
+	}
+	if n == 2 {
+		pw, err := randomPassword(16)
+		if err != nil {
+			log.Fatalf("生成随机密码失败: %v", err)
+		}
+		fmt.Printf("🔑 生成的随机密码：%s\n", pw)
+		fmt.Println("（请立即保存，不会再次显示）")
+		return pw
+	}
+
+	// 手动输入：静默输入（不回显）
+	fmt.Print("请输入新密码（至少 8 位）：")
+	raw, err := term.ReadPassword(int(os.Stdin.Fd()))
+	fmt.Println()
+	if err != nil {
+		log.Fatalf("读取密码失败: %v", err)
+	}
+	return strings.TrimSpace(string(raw))
+}
+
+// randomPassword 用 crypto/rand 从 pwAlphabet 生成 n 位随机密码。
+func randomPassword(n int) (string, error) {
+	buf := make([]byte, n)
+	for i := range buf {
+		k, err := rand.Int(rand.Reader, big.NewInt(int64(len(pwAlphabet))))
+		if err != nil {
+			return "", err
+		}
+		buf[i] = pwAlphabet[k.Int64()]
+	}
+	return string(buf), nil
+}

server/internal/cli/reset_password.go

@@ -0,0 +1,176 @@
+// Package cli 命令行入口。
+package cli
+
+import (
+	"fmt"
+	"log"
+	"os"
+	"strconv"
+	"strings"
+
+	"github.com/openteam/server/internal/config"
+	"github.com/openteam/server/internal/pkg/crypto"
+	"github.com/openteam/server/internal/store"
+	"github.com/spf13/cobra"
+	"golang.org/x/term"
+	"gorm.io/gorm"
+)
+
+var (
+	resetPwUser     string
+	resetPwPassword string
+	resetPwUserID   string
+)
+
+var resetPasswordCmd = &cobra.Command{
+	Use:   "reset-password",
+	Short: "重置用户密码",
+	Long: `重置任意用户密码（管理员可重置任何用户，需在可加载配置的目录下执行）。
+
+交互式引导（默认）：
+  go run ./cmd/server reset-password
+
+非交互式（用于脚本）：
+  go run ./cmd/server reset-password --user admin --password "新密码"
+  go run ./cmd/server reset-password --user-id 123 --password "新密码"`,
+	Run: func(cmd *cobra.Command, args []string) {
+		runResetPassword()
+	},
+}
+
+func init() {
+	rootCmd.AddCommand(resetPasswordCmd)
+	resetPasswordCmd.Flags().StringVar(&resetPwUser, "user", "", "目标用户名（留空则交互选择）")
+	resetPasswordCmd.Flags().StringVar(&resetPwUserID, "user-id", "", "目标用户 ID（优先于 --user）")
+	resetPasswordCmd.Flags().StringVar(&resetPwPassword, "password", "", "新密码，至少 8 位（留空则交互输入）")
+}
+
+func runResetPassword() {
+	// --- 阶段 1：加载配置与数据库 ---
+	cfg, err := config.Load()
+	if err != nil {
+		log.Fatalf("加载配置失败: %v", err)
+	}
+	db, err := store.Open(cfg.DB.Driver, cfg.DB.DSN)
+	if err != nil {
+		log.Fatalf("连接数据库失败: %v", err)
+	}
+
+	// --- 阶段 2：选择目标用户 ---
+	target := targetUser(db)
+
+	// --- 阶段 3：获取新密码 ---
+	var pw string
+	if resetPwPassword != "" {
+		pw = resetPwPassword
+	} else {
+		pw = promptNewPassword()
+	}
+	if pw == "" {
+		log.Fatalf("密码不能为空")
+	}
+	if len(pw) < 8 || len(pw) > 72 {
+		log.Fatalf("密码长度必须在 8 到 72 位之间")
+	}
+
+	// --- 阶段 4：加密并更新 ---
+	hasher := crypto.NewPasswordHasher(cfg.Auth.Argon2Time, cfg.Auth.Argon2Memory, cfg.Auth.Argon2Threads, cfg.Auth.Argon2KeyLen, cfg.Auth.SaltLen)
+	hash, err := hasher.HashPassword(pw)
+	if err != nil {
+		log.Fatalf("加密密码失败: %v", err)
+	}
+	if err := db.Model(&store.User{}).Where("id = ?", target.ID).Update("password_hash", hash).Error; err != nil {
+		log.Fatalf("更新密码失败: %v", err)
+	}
+
+	fmt.Printf("✅ 已重置用户 %q 的密码！\n", target.Username)
+}
+
+// targetUser 根据 --user-id 或 --user 解析目标用户，否则交互选择。
+func targetUser(db *gorm.DB) *store.User {
+	// 优先使用 --user-id
+	if resetPwUserID != "" {
+		id, err := strconv.ParseUint(resetPwUserID, 10, 64)
+		if err != nil {
+			log.Fatalf("无效的用户 ID: %v", err)
+		}
+		var u store.User
+		if err := db.First(&u, id).Error; err != nil {
+			log.Fatalf("用户 ID %d 不存在", id)
+		}
+		return &u
+	}
+
+	// 使用 --user
+	if resetPwUser != "" {
+		var u store.User
+		if err := db.Where("username = ?", resetPwUser).First(&u).Error; err != nil {
+			log.Fatalf("用户 %q 不存在", resetPwUser)
+		}
+		return &u
+	}
+
+	// 交互选择
+	var users []struct {
+		ID       uint64
+		Username string
+		Role     string
+	}
+	db.Model(&store.User{}).Select("id, username, role").Order("id ASC").Find(&users)
+	if len(users) == 0 {
+		log.Fatalf("系统中没有用户")
+	}
+	if len(users) == 1 {
+		return &store.User{ID: users[0].ID, Username: users[0].Username}
+	}
+	return selectUser(users)
+}
+
+// selectUser 交互式选择用户。
+func selectUser(users []struct {
+	ID       uint64
+	Username string
+	Role     string
+}) *store.User {
+	fmt.Println("请选择要重置密码的用户：")
+	for i, u := range users {
+		fmt.Printf("  [%d] %s (%s)\n", i+1, u.Username, u.Role)
+	}
+	fmt.Printf("请输入编号 (1-%d): ", len(users))
+	var n int
+	if _, err := fmt.Scanf("%d", &n); err != nil || n < 1 || n > len(users) {
+		log.Fatalf("无效输入")
+	}
+	return &store.User{ID: users[n-1].ID, Username: users[n-1].Username}
+}
+
+// promptNewPassword 交互式输入密码。
+func promptNewPassword() string {
+	fmt.Println("请选择密码设置方式：")
+	fmt.Println("  [1] 手动输入密码")
+	fmt.Println("  [2] 自动生成随机密码（16 位，不含易混淆字符）")
+	fmt.Printf("请输入编号 (1-2): ")
+
+	var n int
+	if _, err := fmt.Scanf("%d", &n); err != nil || n < 1 || n > 2 {
+		log.Fatalf("无效输入，请输入 1 或 2")
+	}
+	if n == 2 {
+		pw, err := randomPassword(16)
+		if err != nil {
+			log.Fatalf("生成随机密码失败: %v", err)
+		}
+		fmt.Printf("生成的随机密码：%s\n", pw)
+		fmt.Println("（请立即保存，不会再次显示）")
+		return pw
+	}
+
+	// 手动输入：静默输入（不回显）
+	fmt.Print("请输入新密码（至少 8 位）：")
+	raw, err := term.ReadPassword(int(os.Stdin.Fd()))
+	fmt.Println()
+	if err != nil {
+		log.Fatalf("读取密码失败: %v", err)
+	}
+	return strings.TrimSpace(string(raw))
+}

server/internal/cli/root.go

@@ -0,0 +1,27 @@
+// Package cli 命令行入口（Cobra 子命令分发）。
+package cli
+
+import (
+	"github.com/spf13/cobra"
+)
+
+// Execute 启动根命令。
+func Execute() {
+	cobra.CheckErr(rootCmd.Execute())
+}
+
+var rootCmd = &cobra.Command{
+	Use:   "openteam",
+	Short: "openteam · LLM API 中转站",
+	Long: `openteam 大模型中转站 API 服务。
+
+直接运行（无子命令）启动 HTTP 服务：
+  go run ./cmd/server
+
+子命令：
+  reset-admin-password  重置管理员密码（交互式，无参数时自动提示）
+  reset-password        重置任意用户密码`,
+	RunE: func(c *cobra.Command, args []string) error {
+		return serve()
+	},
+}

server/internal/cli/serve.go

@@ -0,0 +1,62 @@
+// Package cli 命令行入口。
+package cli
+
+import (
+	"context"
+	"errors"
+	"log"
+	"net/http"
+	"os"
+	"os/signal"
+	"strconv"
+	"syscall"
+	"time"
+
+	"github.com/openteam/server/internal/api"
+	"github.com/openteam/server/internal/app"
+	"github.com/openteam/server/internal/config"
+	"github.com/openteam/server/internal/proxy"
+)
+
+// serve 启动 HTTP 服务（默认子命令）。
+func serve() error {
+	cfg, err := config.Load()
+	if err != nil {
+		return err
+	}
+
+	a, err := app.New(cfg)
+	if err != nil {
+		return err
+	}
+	defer a.Shutdown(context.Background())
+
+	gw := proxy.NewGateway(a.DB, a.Enc, a.Usage, a.Limit, cfg.RateLimit.UserRPS, cfg.Proxy.LogRaw, cfg.Proxy.Timeout)
+	router := api.NewRouter(a, gw)
+
+	srv := &http.Server{
+		Addr:              ":" + strconv.Itoa(cfg.Port),
+		Handler:           router,
+		ReadHeaderTimeout: 10 * time.Second,
+	}
+
+	go func() {
+		log.Printf("openteam listening on %s (env=%s)", srv.Addr, cfg.Env)
+		if err := srv.ListenAndServe(); err != nil && !errors.Is(err, http.ErrServerClosed) {
+			log.Fatalf("server: %v", err)
+		}
+	}()
+
+	quit := make(chan os.Signal, 1)
+	signal.Notify(quit, syscall.SIGINT, syscall.SIGTERM)
+	<-quit
+
+	log.Println("shutting down...")
+	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+	defer cancel()
+	if err := srv.Shutdown(ctx); err != nil {
+		log.Printf("server shutdown: %v", err)
+	}
+	a.Usage.Close()
+	return nil
+}

server/internal/config/config.go

@@ -66,6 +66,7 @@ type ProxyConfig struct {
 	Timeout             time.Duration
 	HealthInterval      time.Duration // 渠道健康检查周期
 	HealthFailThreshold int           // 连续失败 N 次进 cooldown
+	LogRaw              bool          // 记录管理员原始请求体+响应到 usage_logs（调试用，默认关）
 }
 
 // loadDotEnv 读取 .env 并把 KEY=VALUE 注入环境变量（AutomaticEnv 自动映射 OT_ 前缀）。
@@ -126,9 +127,10 @@ func Load() (*Config, error) {
 	v.SetDefault("proxy.upstream_base_url", "https://api.openai.com")
 	v.SetDefault("proxy.upstream_key", "")
 	v.SetDefault("proxy.default_model", "gpt-4o-mini")
-	v.SetDefault("proxy.timeout", "120s")
+	v.SetDefault("proxy.timeout", "300s")
 	v.SetDefault("proxy.health_interval", "60s")
 	v.SetDefault("proxy.health_fail_threshold", 2)
+	v.SetDefault("proxy.log_raw", false)
 
 	v.SetDefault("ratelimit.user_rps", 20)
 
@@ -168,6 +170,7 @@ func Load() (*Config, error) {
 			Timeout:             v.GetDuration("proxy.timeout"),
 			HealthInterval:      v.GetDuration("proxy.health_interval"),
 			HealthFailThreshold: v.GetInt("proxy.health_fail_threshold"),
+			LogRaw:              v.GetBool("proxy.log_raw"),
 		},
 		RateLimit: RateLimitConfig{
 			UserRPS: v.GetInt("ratelimit.user_rps"),

server/internal/passkey/passkey.go

@@ -235,7 +235,10 @@ func (s *Service) takeSession(challenge string) (webauthn.SessionData, bool) {
 		delete(s.sessions, challenge)
 	}
 	s.mu.Unlock()
-	if ok && time.Now().After(sess.Expires) {
+	// Expires 可能为零值：go-webauthn 默认 Enforce=false 不设过期时间。
+	// 零值时间恒早于 now，直接 After 会把每个 challenge 都判为过期，
+	// 与库内部一致，仅当显式设置了过期时间才做校验。
+	if ok && !sess.Expires.IsZero() && time.Now().After(sess.Expires) {
 		return webauthn.SessionData{}, false
 	}
 	return sess, ok

server/internal/passkey/passkey_test.go

@@ -0,0 +1,56 @@
+package passkey
+
+import (
+	"testing"
+
+	"github.com/glebarez/sqlite"
+	"github.com/openteam/server/internal/store"
+	"gorm.io/gorm"
+)
+
+func newTestService(t *testing.T) *Service {
+	t.Helper()
+	db, err := gorm.Open(sqlite.Open("file::memory:?cache=shared"), &gorm.Config{})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if err := db.AutoMigrate(&store.User{}, &store.Passkey{}); err != nil {
+		t.Fatal(err)
+	}
+	s, err := New(db, Config{RPID: "localhost", Origin: "http://localhost:8080", Name: "openteam test"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	return s
+}
+
+// TestChallengeSessionRoundTrip 验证 begin 产生的 challenge 能被 complete 取回。
+// go-webauthn 默认 Enforce=false，SessionData.Expires 为零值；takeSession 的
+// 过期判断必须跳过零值时间，否则每个 challenge 都被当成已过期（线上必现
+// "challenge 已过期或不存在"）。
+func TestChallengeSessionRoundTrip(t *testing.T) {
+	s := newTestService(t)
+	u := &store.User{Username: "alice", Email: "alice@example.com", Role: "user", Status: "active"}
+	if err := s.db.Create(u).Error; err != nil {
+		t.Fatal(err)
+	}
+
+	creation, err := s.BeginRegistration(u)
+	if err != nil {
+		t.Fatal(err)
+	}
+	challenge := creation.Response.Challenge.String() // 与前端回传一致的 base64url
+
+	sess, ok := s.takeSession(challenge)
+	if !ok {
+		t.Fatal("takeSession 返回 false：challenge 被误判为已过期（Expires 零值 bug）")
+	}
+	if sess.Challenge != challenge {
+		t.Fatalf("session challenge 不匹配: %q != %q", sess.Challenge, challenge)
+	}
+
+	// challenge 一次性消费
+	if _, ok := s.takeSession(challenge); ok {
+		t.Fatal("takeSession 应一次性消费 challenge，二次取应失败")
+	}
+}

server/internal/pkg/apikey/apikey.go

@@ -1,4 +1,4 @@
-// Package apikey 生成与管理 API Key：sk- + 48 位 base62 随机串。
+// Package apikey 生成与管理 API Key：sk-ot- + 48 位 base62 随机串。
 // 库中仅存 SHA-256 哈希与展示前缀（PLANNING §4.3.3）。
 package apikey
 
@@ -12,7 +12,7 @@ import (
 const (
 	alphabet = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz"
 	keyLen   = 48
-	prefix   = "sk-"
+	prefix   = "sk-ot-"
 )
 
 // Generate 生成明文 key（仅创建时展示一次）与哈希、前缀。
@@ -34,7 +34,7 @@ func Hash(key string) string {
 	return hex.EncodeToString(sum[:])
 }
 
-// Prefix 展示前缀：sk-aB3cD5…（前 12 字符）
+// Prefix 展示前缀：sk-ot-aB3c…（前 12 字符）
 func Prefix(key string) string {
 	if len(key) <= 12 {
 		return key

server/internal/pkg/apikey/apikey_test.go

@@ -3,32 +3,35 @@ package apikey
 import "testing"
 
 func TestGenerateValid(t *testing.T) {
-	plain, hash, prefix, err := Generate()
+	plain, hash, keyPrefix, err := Generate()
 	if err != nil {
 		t.Fatalf("Generate: %v", err)
 	}
 	if !Valid(plain) {
 		t.Fatalf("generated key invalid: %q", plain)
 	}
-	if len(plain) != len("sk-")+48 {
+	if len(plain) != len(prefix)+keyLen {
 		t.Fatalf("unexpected key length: %d", len(plain))
 	}
+	if keyPrefix != plain[:12] {
+		t.Fatalf("prefix mismatch: %s vs %s", keyPrefix, plain[:12])
+	}
 	if Hash(plain) != hash {
 		t.Fatal("hash mismatch")
 	}
-	if prefix != plain[:12] {
-		t.Fatalf("prefix mismatch: %s vs %s", prefix, plain[:12])
-	}
 }
 
 func TestHashStable(t *testing.T) {
-	if Hash("sk-test") != Hash("sk-test") {
+	if Hash("sk-ot-test") != Hash("sk-ot-test") {
 		t.Fatal("hash not stable")
 	}
 }
 
 func TestValidRejects(t *testing.T) {
-	cases := []string{"", "sk-abc", "abc-123456789012345678901234567890123456789012345678", "sk-1234567890123456789012345678901234567890123456789"}
+	tooShort := prefix + "abc"
+	tooLong := prefix + "1234567890123456789012345678901234567890123456789"
+	wrongPrefix := "abc-ot123456789012345678901234567890123456789012345678"
+	cases := []string{"", tooShort, wrongPrefix, tooLong}
 	for _, c := range cases {
 		if Valid(c) {
 			t.Fatalf("expected invalid: %q", c)

server/internal/proxy/convert/convert_test.go

@@ -202,6 +202,51 @@ func TestChatToMessagesResp(t *testing.T) {
 	}
 }
 
+// TestResponsesToMessagesRespToolCall 非流式 Responses 响应 → Messages 响应（经 chat 中转）。
+func TestResponsesToMessagesRespToolCall(t *testing.T) {
+	in := `{"id":"resp_abc","object":"response","model":"glm-4","status":"completed",
+		"output":[
+			{"type":"message","role":"assistant","content":[{"type":"output_text","text":"我来查询"}]},
+			{"type":"function_call","id":"fc_1","call_id":"call_9","name":"get_weather","arguments":"{\"city\":\"北京\"}"}
+		],
+		"usage":{"input_tokens":12,"output_tokens":8,"total_tokens":20}}`
+	out, err := ConvertResponse([]byte(in), ProtoResponses, ProtoMessages)
+	if err != nil {
+		t.Fatalf("convert: %v", err)
+	}
+	var m map[string]any
+	_ = json.Unmarshal(out, &m)
+	if m["stop_reason"] != "tool_use" {
+		t.Fatalf("stop_reason = %v", m["stop_reason"])
+	}
+	if m["model"] != "glm-4" {
+		t.Fatalf("model = %v", m["model"])
+	}
+	content := m["content"].([]any)
+	var text, tool string
+	for _, b := range content {
+		bm := b.(map[string]any)
+		switch bm["type"] {
+		case "text":
+			text, _ = bm["text"].(string)
+		case "tool_use":
+			tool = "y"
+			if bm["id"] != "call_9" || bm["name"] != "get_weather" {
+				t.Fatalf("tool block = %v", bm)
+			}
+			if in, _ := bm["input"].(map[string]any); in["city"] != "北京" {
+				t.Fatalf("tool input = %v", bm["input"])
+			}
+		}
+	}
+	if text != "我来查询" {
+		t.Fatalf("text = %q", text)
+	}
+	if tool != "y" {
+		t.Fatal("expected tool_use block")
+	}
+}
+
 // ---------------------------------------------------------------------------
 // 流式转换
 
@@ -549,3 +594,431 @@ func TestMessagesToResponsesReqImage(t *testing.T) {
 		t.Fatalf("input_image = %v", img)
 	}
 }
+
+// ---------------------------------------------------------------------------
+// 流式工具调用转换（chat tool_calls / messages tool_use / responses function_call 互转不丢）
+
+// sseEvent 一条解析出的 SSE 事件（dataLine 无 event 行时 typ 为空串）。
+type sseEvent struct {
+	typ  string
+	data map[string]any
+}
+
+// parseSSE 把转换器输出拆成事件列表。
+func parseSSE(t *testing.T, s string) []sseEvent {
+	t.Helper()
+	var evs []sseEvent
+	cur := ""
+	for _, ln := range strings.Split(s, "\n") {
+		switch {
+		case strings.HasPrefix(ln, "event: "):
+			cur = strings.TrimSpace(strings.TrimPrefix(ln, "event: "))
+		case strings.HasPrefix(ln, "data: "):
+			d := strings.TrimPrefix(ln, "data: ")
+			if d == "[DONE]" {
+				evs = append(evs, sseEvent{typ: "[DONE]"})
+				continue
+			}
+			var m map[string]any
+			if err := json.Unmarshal([]byte(d), &m); err != nil {
+				t.Fatalf("parse sse data: %v\n%s", err, d)
+			}
+			evs = append(evs, sseEvent{typ: cur, data: m})
+		}
+	}
+	return evs
+}
+
+// findSSE 返回第一个 typ 匹配且（index>=0 时）index 字段相符的事件。
+func findSSE(evs []sseEvent, typ string, index int) *sseEvent {
+	for i := range evs {
+		if evs[i].typ != typ {
+			continue
+		}
+		if index >= 0 {
+			if idx, ok := evs[i].data["index"].(float64); !ok || int(idx) != index {
+				continue
+			}
+		}
+		return &evs[i]
+	}
+	return nil
+}
+
+// inputJSONDeltas 汇总所有 input_json_delta：index → 拼接后的 partial_json。
+func inputJSONDeltas(evs []sseEvent) map[int]string {
+	out := map[int]string{}
+	for _, e := range evs {
+		if e.typ != "content_block_delta" {
+			continue
+		}
+		d, _ := e.data["delta"].(map[string]any)
+		if d == nil || d["type"] != "input_json_delta" {
+			continue
+		}
+		idx, _ := e.data["index"].(float64)
+		p, _ := d["partial_json"].(string)
+		out[int(idx)] += p
+	}
+	return out
+}
+
+func TestStreamChatToMessagesToolCall(t *testing.T) {
+	tf := newChatToMessages().line
+	out := feedLines(t, tf, []string{
+		`data: {"id":"chatcmpl-1","model":"glm-4.7-flash","choices":[{"index":0,"delta":{"role":"assistant","reasoning_content":"用户"}}]}` + "\n\n",
+		`data: {"id":"chatcmpl-1","model":"glm-4.7-flash","choices":[{"index":0,"delta":{"content":"我来帮您查询"}}]}` + "\n\n",
+		`data: {"id":"chatcmpl-1","model":"glm-4.7-flash","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_dbcef6a399ad47bb973fa83a","index":0,"type":"function","function":{"name":"get_weather","arguments":"{\"city\":\"北京\"}"}}]}}]}` + "\n\n",
+		`data: {"id":"chatcmpl-1","model":"glm-4.7-flash","choices":[{"index":0,"finish_reason":"tool_calls","delta":{"role":"assistant","content":""}}],"usage":{"prompt_tokens":167,"completion_tokens":48}}` + "\n\n",
+		"data: [DONE]\n\n",
+	})
+	evs := parseSSE(t, out)
+
+	// message_start
+	if ms := findSSE(evs, "message_start", -1); ms == nil {
+		t.Fatal("missing message_start")
+	}
+	// 文本块 index 0
+	if cs := findSSE(evs, "content_block_start", 0); cs == nil {
+		t.Fatal("missing text content_block_start")
+	} else if cb := cs.data["content_block"].(map[string]any); cb["type"] != "text" {
+		t.Fatalf("block 0 type = %v", cb["type"])
+	}
+	if !strings.Contains(out, `"text":"我来帮您查询"`) {
+		t.Fatalf("missing text delta: %s", out)
+	}
+	// 工具块 index 1
+	cs := findSSE(evs, "content_block_start", 1)
+	if cs == nil {
+		t.Fatal("missing tool content_block_start")
+	}
+	cb := cs.data["content_block"].(map[string]any)
+	if cb["type"] != "tool_use" || cb["id"] != "call_dbcef6a399ad47bb973fa83a" || cb["name"] != "get_weather" {
+		t.Fatalf("tool block = %v", cb)
+	}
+	if input, ok := cb["input"].(map[string]any); !ok || len(input) != 0 {
+		t.Fatalf("tool input should be empty object, got %v", cb["input"])
+	}
+	// input_json_delta 整段 args
+	if d := inputJSONDeltas(evs)[1]; d != `{"city":"北京"}` {
+		t.Fatalf("tool args = %q", d)
+	}
+	// 两个块都有 stop
+	if findSSE(evs, "content_block_stop", 0) == nil || findSSE(evs, "content_block_stop", 1) == nil {
+		t.Fatalf("missing per-block stops: %s", out)
+	}
+	// stop_reason tool_use
+	if md := findSSE(evs, "message_delta", -1); md == nil {
+		t.Fatal("missing message_delta")
+	} else if d := md.data["delta"].(map[string]any); d["stop_reason"] != "tool_use" {
+		t.Fatalf("stop_reason = %v", d["stop_reason"])
+	}
+	if findSSE(evs, "message_stop", -1) == nil {
+		t.Fatal("missing message_stop")
+	}
+}
+
+func TestStreamChatToMessagesToolOnly(t *testing.T) {
+	tf := newChatToMessages().line
+	out := feedLines(t, tf, []string{
+		`data: {"id":"chatcmpl-1","model":"glm-4.7-flash","choices":[{"index":0,"delta":{"role":"assistant","reasoning_content":"查询"}}]}` + "\n\n",
+		`data: {"id":"chatcmpl-1","model":"glm-4.7-flash","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_1","index":0,"type":"function","function":{"name":"get_weather","arguments":"{\"city\":\"北京\"}"}}]}}]}` + "\n\n",
+		`data: {"id":"chatcmpl-1","model":"glm-4.7-flash","choices":[{"index":0,"finish_reason":"tool_calls","delta":{"role":"assistant","content":""}}]}` + "\n\n",
+		"data: [DONE]\n\n",
+	})
+	evs := parseSSE(t, out)
+
+	// 首块即 tool_use index 0
+	cs := findSSE(evs, "content_block_start", 0)
+	if cs == nil {
+		t.Fatal("missing first block")
+	}
+	if cb := cs.data["content_block"].(map[string]any); cb["type"] != "tool_use" {
+		t.Fatalf("first block type = %v", cb["type"])
+	}
+	// 不应出现空文本块
+	for _, e := range evs {
+		if e.typ == "content_block_start" {
+			if cb := e.data["content_block"].(map[string]any); cb["type"] == "text" {
+				t.Fatalf("unexpected text block: %s", out)
+			}
+		}
+	}
+	// 单个 stop
+	if findSSE(evs, "content_block_stop", 0) == nil {
+		t.Fatal("missing stop 0")
+	}
+	if findSSE(evs, "content_block_stop", 1) != nil {
+		t.Fatalf("unexpected stop 1: %s", out)
+	}
+	if md := findSSE(evs, "message_delta", -1); md == nil {
+		t.Fatal("missing message_delta")
+	} else if d := md.data["delta"].(map[string]any); d["stop_reason"] != "tool_use" {
+		t.Fatalf("stop_reason = %v", d["stop_reason"])
+	}
+}
+
+func TestStreamChatToMessagesParallelTools(t *testing.T) {
+	tf := newChatToMessages().line
+	out := feedLines(t, tf, []string{
+		`data: {"id":"chatcmpl-1","model":"deepseek","choices":[{"index":0,"delta":{"role":"assistant","tool_calls":[{"id":"call_0","index":0,"type":"function","function":{"name":"search","arguments":""}}]}}]}` + "\n\n",
+		`data: {"id":"chatcmpl-1","model":"deepseek","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_1","index":1,"type":"function","function":{"name":"read","arguments":""}}]}}]}` + "\n\n",
+		`data: {"id":"chatcmpl-1","model":"deepseek","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{\"query\":"}}]}}]}` + "\n\n",
+		`data: {"id":"chatcmpl-1","model":"deepseek","choices":[{"index":0,"delta":{"tool_calls":[{"index":1,"function":{"arguments":"{\"path\":"}}]}}]}` + "\n\n",
+		`data: {"id":"chatcmpl-1","model":"deepseek","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"weather\"}"}}]}}]}` + "\n\n",
+		`data: {"id":"chatcmpl-1","model":"deepseek","choices":[{"index":0,"delta":{"tool_calls":[{"index":1,"function":{"arguments":"\"a.txt\"}"}}]}}]}` + "\n\n",
+		`data: {"id":"chatcmpl-1","model":"deepseek","choices":[{"index":0,"finish_reason":"tool_calls","delta":{}}]}` + "\n\n",
+		"data: [DONE]\n\n",
+	})
+	evs := parseSSE(t, out)
+
+	// 块 0 = search，块 1 = read（按首现顺序分配）
+	if cb := findSSE(evs, "content_block_start", 0); cb == nil {
+		t.Fatal("missing block 0")
+	} else if n := cb.data["content_block"].(map[string]any)["name"]; n != "search" {
+		t.Fatalf("block 0 name = %v", n)
+	}
+	if cb := findSSE(evs, "content_block_start", 1); cb == nil {
+		t.Fatal("missing block 1")
+	} else if n := cb.data["content_block"].(map[string]any)["name"]; n != "read" {
+		t.Fatalf("block 1 name = %v", n)
+	}
+	// 参数分段正确路由
+	deltas := inputJSONDeltas(evs)
+	if deltas[0] != `{"query":"weather"}` {
+		t.Fatalf("block 0 args = %q", deltas[0])
+	}
+	if deltas[1] != `{"path":"a.txt"}` {
+		t.Fatalf("block 1 args = %q", deltas[1])
+	}
+	if findSSE(evs, "content_block_stop", 0) == nil || findSSE(evs, "content_block_stop", 1) == nil {
+		t.Fatalf("missing stops: %s", out)
+	}
+}
+
+func TestStreamChatToMessagesTextAfterTool(t *testing.T) {
+	tf := newChatToMessages().line
+	out := feedLines(t, tf, []string{
+		`data: {"id":"chatcmpl-1","model":"deepseek","choices":[{"index":0,"delta":{"role":"assistant","tool_calls":[{"id":"call_1","index":0,"type":"function","function":{"name":"get_weather","arguments":""}}]}}]}` + "\n\n",
+		`data: {"id":"chatcmpl-1","model":"deepseek","choices":[{"index":0,"delta":{"content":"天气查询中"}}]}` + "\n\n",
+		`data: {"id":"chatcmpl-1","model":"deepseek","choices":[{"index":0,"finish_reason":"stop","delta":{}}]}` + "\n\n",
+		"data: [DONE]\n\n",
+	})
+	evs := parseSSE(t, out)
+
+	// 工具块 index 0、文本块 index 1
+	if cb := findSSE(evs, "content_block_start", 0); cb == nil {
+		t.Fatal("missing tool block")
+	} else if cb.data["content_block"].(map[string]any)["type"] != "tool_use" {
+		t.Fatalf("block 0 type = %v", cb.data["content_block"].(map[string]any)["type"])
+	}
+	if cb := findSSE(evs, "content_block_start", 1); cb == nil {
+		t.Fatal("missing text block")
+	} else if cb.data["content_block"].(map[string]any)["type"] != "text" {
+		t.Fatalf("block 1 type = %v", cb.data["content_block"].(map[string]any)["type"])
+	}
+	// stop 顺序 0,1
+	var stops []int
+	for _, e := range evs {
+		if e.typ == "content_block_stop" {
+			idx, _ := e.data["index"].(float64)
+			stops = append(stops, int(idx))
+		}
+	}
+	if len(stops) != 2 || stops[0] != 0 || stops[1] != 1 {
+		t.Fatalf("stop order = %v", stops)
+	}
+}
+
+func TestStreamMessagesToChatToolUse(t *testing.T) {
+	tf := newMessagesToChat().line
+	out := feedLines(t, tf, []string{
+		"event: message_start\n",
+		`data: {"type":"message_start","message":{"id":"msg_1","model":"glm-4.7-flash"}}` + "\n\n",
+		"event: content_block_delta\n",
+		`data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":"你好"}}` + "\n\n",
+		"event: content_block_start\n",
+		`data: {"type":"content_block_start","index":1,"content_block":{"type":"tool_use","id":"call_x","name":"get_weather","input":{}}}` + "\n\n",
+		"event: content_block_delta\n",
+		`data: {"type":"content_block_delta","index":1,"delta":{"type":"input_json_delta","partial_json":"{\""}}` + "\n\n",
+		"event: content_block_delta\n",
+		`data: {"type":"content_block_delta","index":1,"delta":{"type":"input_json_delta","partial_json":"city\":\"北京\"}"}}` + "\n\n",
+		"event: content_block_stop\n",
+		`data: {"type":"content_block_stop","index":1}` + "\n\n",
+		"event: message_delta\n",
+		`data: {"type":"message_delta","delta":{"stop_reason":"tool_use"}}` + "\n\n",
+		"event: message_stop\n",
+		`data: {"type":"message_stop"}` + "\n\n",
+	})
+	evs := parseSSE(t, out)
+
+	// 文本增量
+	if !strings.Contains(out, `"content":"你好"`) {
+		t.Fatalf("missing text chunk: %s", out)
+	}
+	// tool_use → chat tool_calls：messages 块 index 1 映射为 chat 工具索引 0
+	var toolStart, argsSeen, finishSeen bool
+	for _, e := range evs {
+		choices, _ := e.data["choices"].([]any)
+		if len(choices) == 0 {
+			continue
+		}
+		c0, _ := choices[0].(map[string]any)
+		delta, _ := c0["delta"].(map[string]any)
+		if tcs, ok := delta["tool_calls"].([]any); ok && len(tcs) > 0 {
+			tc, _ := tcs[0].(map[string]any)
+			idx, _ := tc["index"].(float64)
+			if int(idx) != 0 {
+				continue
+			}
+			fn, _ := tc["function"].(map[string]any)
+			if tc["id"] == "call_x" && fn["name"] == "get_weather" {
+				toolStart = true
+			}
+			if s, _ := fn["arguments"].(string); s != "" {
+				argsSeen = true
+			}
+		}
+		if fr, _ := c0["finish_reason"].(string); fr == "tool_calls" {
+			finishSeen = true
+		}
+	}
+	if !toolStart {
+		t.Fatalf("missing tool_calls start: %s", out)
+	}
+	if !argsSeen {
+		t.Fatalf("missing tool_calls arguments chunks: %s", out)
+	}
+	if !finishSeen {
+		t.Fatalf("missing finish_reason tool_calls: %s", out)
+	}
+	if !strings.Contains(out, "data: [DONE]") {
+		t.Fatalf("missing [DONE]: %s", out)
+	}
+}
+
+func TestStreamResponsesToMessagesFunctionCall(t *testing.T) {
+	tf := newResponsesToMessages().line
+	out := feedLines(t, tf, []string{
+		"event: response.created\n",
+		`data: {"type":"response.created","response":{"id":"resp_1","model":"glm-4.7-flash"}}` + "\n\n",
+		"event: response.output_item.added\n",
+		`data: {"type":"response.output_item.added","item":{"type":"function_call","id":"fc_call_1","call_id":"call_1","name":"get_weather","arguments":"","status":"in_progress"},"output_index":0}` + "\n\n",
+		"event: response.function_call_arguments.delta\n",
+		`data: {"type":"response.function_call_arguments.delta","delta":"{\"","item_id":"fc_call_1"}` + "\n\n",
+		"event: response.function_call_arguments.delta\n",
+		`data: {"type":"response.function_call_arguments.delta","delta":"city\":\"北京\"}","item_id":"fc_call_1"}` + "\n\n",
+		"event: response.output_item.done\n",
+		`data: {"type":"response.output_item.done","item":{"type":"function_call","id":"fc_call_1","arguments":"{\"city\":\"北京\"}","status":"completed"}}` + "\n\n",
+		"event: response.completed\n",
+		`data: {"type":"response.completed","response":{"id":"resp_1","status":"completed","usage":{"input_tokens":7,"output_tokens":8}}}` + "\n\n",
+	})
+	evs := parseSSE(t, out)
+
+	if findSSE(evs, "message_start", -1) == nil {
+		t.Fatal("missing message_start")
+	}
+	// 首块即 tool_use index 0
+	cs := findSSE(evs, "content_block_start", 0)
+	if cs == nil {
+		t.Fatal("missing tool block")
+	}
+	cb := cs.data["content_block"].(map[string]any)
+	if cb["type"] != "tool_use" || cb["id"] != "call_1" || cb["name"] != "get_weather" {
+		t.Fatalf("tool block = %v", cb)
+	}
+	if d := inputJSONDeltas(evs)[0]; d != `{"city":"北京"}` {
+		t.Fatalf("tool args = %q", d)
+	}
+	if findSSE(evs, "content_block_stop", 0) == nil {
+		t.Fatal("missing stop")
+	}
+	if md := findSSE(evs, "message_delta", -1); md == nil {
+		t.Fatal("missing message_delta")
+	} else if d := md.data["delta"].(map[string]any); d["stop_reason"] != "tool_use" {
+		t.Fatalf("stop_reason = %v", d["stop_reason"])
+	}
+	if findSSE(evs, "message_stop", -1) == nil {
+		t.Fatal("missing message_stop")
+	}
+}
+
+func TestStreamResponsesToMessagesTextThenFunction(t *testing.T) {
+	tf := newResponsesToMessages().line
+	out := feedLines(t, tf, []string{
+		"event: response.created\n",
+		`data: {"type":"response.created","response":{"id":"resp_1","model":"glm-4.7-flash"}}` + "\n\n",
+		"event: response.output_text.delta\n",
+		`data: {"type":"response.output_text.delta","delta":"我来查询"}` + "\n\n",
+		"event: response.output_item.added\n",
+		`data: {"type":"response.output_item.added","item":{"type":"function_call","id":"fc_call_1","call_id":"call_1","name":"get_weather","arguments":"","status":"in_progress"},"output_index":1}` + "\n\n",
+		"event: response.function_call_arguments.delta\n",
+		`data: {"type":"response.function_call_arguments.delta","delta":"{\"city\":\"北京\"}","item_id":"fc_call_1"}` + "\n\n",
+		"event: response.completed\n",
+		`data: {"type":"response.completed","response":{"id":"resp_1","status":"completed","usage":{"input_tokens":7,"output_tokens":8}}}` + "\n\n",
+	})
+	evs := parseSSE(t, out)
+
+	// 文本块 index 0、工具块 index 1
+	if cb := findSSE(evs, "content_block_start", 0); cb == nil {
+		t.Fatal("missing text block")
+	} else if cb.data["content_block"].(map[string]any)["type"] != "text" {
+		t.Fatalf("block 0 type = %v", cb.data["content_block"].(map[string]any)["type"])
+	}
+	if cb := findSSE(evs, "content_block_start", 1); cb == nil {
+		t.Fatal("missing tool block")
+	} else if cb.data["content_block"].(map[string]any)["type"] != "tool_use" {
+		t.Fatalf("block 1 type = %v", cb.data["content_block"].(map[string]any)["type"])
+	}
+	if !strings.Contains(out, `"text":"我来查询"`) {
+		t.Fatalf("missing text delta: %s", out)
+	}
+	// stop 顺序 0,1
+	var stops []int
+	for _, e := range evs {
+		if e.typ == "content_block_stop" {
+			idx, _ := e.data["index"].(float64)
+			stops = append(stops, int(idx))
+		}
+	}
+	if len(stops) != 2 || stops[0] != 0 || stops[1] != 1 {
+		t.Fatalf("stop order = %v", stops)
+	}
+	if md := findSSE(evs, "message_delta", -1); md == nil {
+		t.Fatal("missing message_delta")
+	} else if d := md.data["delta"].(map[string]any); d["stop_reason"] != "tool_use" {
+		t.Fatalf("stop_reason = %v", d["stop_reason"])
+	}
+}
+
+func TestMessagesToChatReqToolOrder(t *testing.T) {
+	in := mustJSON(t, map[string]any{
+		"model": "gpt-4o",
+		"messages": []any{map[string]any{"role": "assistant", "content": []any{
+			map[string]any{"type": "text", "text": "我来"},
+			map[string]any{"type": "tool_use", "id": "call_1", "name": "get_weather", "input": map[string]any{"city": "北京"}},
+			map[string]any{"type": "tool_result", "tool_use_id": "call_1", "content": "晴"},
+		}}},
+	})
+	out, err := messagesToChatReq([]byte(in))
+	if err != nil {
+		t.Fatalf("messagesToChatReq: %v", err)
+	}
+	var m map[string]any
+	if err := json.Unmarshal(out, &m); err != nil {
+		t.Fatalf("unmarshal: %v", err)
+	}
+	msgs := m["messages"].([]any)
+	if len(msgs) != 2 {
+		t.Fatalf("messages len = %d: %s", len(msgs), out)
+	}
+	if r := msgs[0].(map[string]any)["role"]; r != "assistant" {
+		t.Fatalf("msg[0] role = %v", r)
+	}
+	if r := msgs[1].(map[string]any)["role"]; r != "tool" {
+		t.Fatalf("msg[1] role = %v", r)
+	}
+	if _, ok := msgs[0].(map[string]any)["tool_calls"]; !ok {
+		t.Fatalf("msg[0] missing tool_calls: %s", out)
+	}
+}

server/internal/proxy/convert/json_chat.go

@@ -230,6 +230,7 @@ func anthropicMsgToChat(role string, content json.RawMessage) []any {
 	var blocks []map[string]any
 	if json.Unmarshal(content, &blocks) == nil && blocks != nil {
 		var out []any
+		var toolMsgs []any // tool_result 单独收集，保证排在 assistant(tool_calls) 之后
 		var textParts []string
 		var contentBlocks []any // text / image_url 块，保留原始顺序
 		var toolCalls []any
@@ -259,7 +260,7 @@ func anthropicMsgToChat(role string, content json.RawMessage) []any {
 			case "tool_result":
 				callID, _ := b["tool_use_id"].(string)
 				res := strField(b["content"])
-				out = append(out, map[string]any{"role": "tool", "tool_call_id": callID, "content": res})
+				toolMsgs = append(toolMsgs, map[string]any{"role": "tool", "tool_call_id": callID, "content": res})
 			}
 		}
 		hasImage := false
@@ -282,6 +283,7 @@ func anthropicMsgToChat(role string, content json.RawMessage) []any {
 			}
 			out = append(out, msg)
 		}
+		out = append(out, toolMsgs...)
 		if len(out) > 0 {
 			return out
 		}

server/internal/proxy/convert/json_responses.go

@@ -240,11 +240,9 @@ func chatToResponsesReq(body []byte) ([]byte, error) {
 	if len(system) > 0 {
 		out["instructions"] = strings.Join(system, "\n")
 	}
-	if len(input) == 1 {
-		out["input"] = input[0] // 单条消息项
-	} else {
+	// input 必须是数组：部分上游（如火山方舟）只接受 []*InputItem，
+	// 单对象会被拒（400 Mismatch type）。
 	out["input"] = input
-	}
 
 	if len(req.Tools) > 0 {
 		tools := make([]any, 0, len(req.Tools))
@@ -305,8 +303,11 @@ func responsesToChatResp(body []byte) ([]byte, error) {
 		msg["tool_calls"] = toolCalls
 	}
 	finish := "stop"
-	if string(rawJSON(m, "status")) == `"incomplete"` {
-		finish = "length"
+	switch {
+	case string(rawJSON(m, "status")) == `"incomplete"`:
+		finish = "length" // 截断优先，客户端可据此区分
+	case len(toolCalls) > 0:
+		finish = "tool_calls" // 有工具调用时映射 tool_calls，否则中转后 stop_reason 成 end_turn
 	}
 	var prompt, completion int64
 	if u := rawJSON(m, "usage"); u != nil {

server/internal/proxy/convert/stream.go

@@ -11,16 +11,17 @@ type sseState struct {
 }
 
 // parseLine 解析一行 SSE；返回是否 data 行及其内容、是否 [DONE]。
+// data: 后可跟空格（标准）或紧贴 JSON（上游如火山方舟会省略空格）。
 func (s *sseState) parseLine(line []byte) (isData bool, data string, done bool) {
 	str := strings.TrimRight(string(line), "\r\n")
 	switch {
 	case strings.HasPrefix(str, "event: "):
 		s.event = strings.TrimSpace(strings.TrimPrefix(str, "event: "))
 		return false, "", false
-	case str == "data: [DONE]":
+	case str == "data: [DONE]" || str == "data:[DONE]":
 		return true, "[DONE]", true
-	case strings.HasPrefix(str, "data: "):
-		return true, strings.TrimPrefix(str, "data: "), false
+	case strings.HasPrefix(str, "data:"):
+		return true, strings.TrimLeft(strings.TrimPrefix(str, "data:"), " "), false
 	default:
 		return false, "", false
 	}
@@ -49,9 +50,11 @@ func eventLine(name string, obj any) []byte {
 type messagesToChat struct {
 	sseState
 	id, model string
+	toolIdx   map[int]int // messages content block index → chat tool_calls index（顺序编号，避开文本块）
+	nextTool  int
 }
 
-func newMessagesToChat() *messagesToChat { return &messagesToChat{} }
+func newMessagesToChat() *messagesToChat { return &messagesToChat{toolIdx: map[int]int{}} }
 
 func (t *messagesToChat) line(line []byte) []byte {
 	isData, data, done := t.parseLine(line)
@@ -72,8 +75,43 @@ func (t *messagesToChat) line(line []byte) []byte {
 			"id": "chatcmpl-" + strings.TrimPrefix(t.id, "msg_"), "object": "chat.completion.chunk", "model": t.model,
 			"choices": []any{map[string]any{"index": 0, "delta": map[string]any{"role": "assistant"}, "finish_reason": nil}},
 		})
+	case "content_block_start":
+		cb, _ := m["content_block"].(map[string]any)
+		if cb == nil || cb["type"] != "tool_use" {
+			return nil
+		}
+		blockIdx, _ := m["index"].(float64)
+		tool := t.nextTool
+		t.nextTool++
+		t.toolIdx[int(blockIdx)] = tool
+		toolID, _ := cb["id"].(string)
+		name, _ := cb["name"].(string)
+		return dataLine(map[string]any{
+			"id": "chatcmpl-" + strings.TrimPrefix(t.id, "msg_"), "object": "chat.completion.chunk", "model": t.model,
+			"choices": []any{map[string]any{"index": 0, "delta": map[string]any{
+				"tool_calls": []any{map[string]any{"index": tool, "id": toolID, "type": "function", "function": map[string]any{"name": name, "arguments": ""}}},
+			}, "finish_reason": nil}},
+		})
 	case "content_block_delta":
 		delta, _ := m["delta"].(map[string]any)
+		deltaType, _ := delta["type"].(string)
+		if deltaType == "input_json_delta" {
+			blockIdx, _ := m["index"].(float64)
+			tool, ok := t.toolIdx[int(blockIdx)]
+			if !ok {
+				return nil
+			}
+			partial, _ := delta["partial_json"].(string)
+			if partial == "" {
+				return nil
+			}
+			return dataLine(map[string]any{
+				"id": "chatcmpl-" + strings.TrimPrefix(t.id, "msg_"), "object": "chat.completion.chunk", "model": t.model,
+				"choices": []any{map[string]any{"index": 0, "delta": map[string]any{
+					"tool_calls": []any{map[string]any{"index": tool, "function": map[string]any{"arguments": partial}}},
+				}, "finish_reason": nil}},
+			})
+		}
 		text, _ := delta["text"].(string)
 		if text == "" {
 			return nil
@@ -118,14 +156,19 @@ func joinLines(lines [][]byte) []byte {
 
 type chatToMessages struct {
 	sseState
-	started      bool
-	blockStarted bool
+	started    bool        // message_start 已发出
+	nextIndex  int         // 下一个 content block index（顺序分配）
+	textIndex  int         // 文本块 index；-1 = 未开始
+	toolIdx    map[int]int // chat delta.tool_calls[].index → messages block index
+	openBlocks []int       // 已开始未停止的 block index，按开始顺序
 	model      string
 	stopReason string
 	usage      any
 }
 
-func newChatToMessages() *chatToMessages { return &chatToMessages{} }
+func newChatToMessages() *chatToMessages {
+	return &chatToMessages{textIndex: -1, toolIdx: map[int]int{}}
+}
 
 func (t *chatToMessages) line(line []byte) []byte {
 	isData, data, done := t.parseLine(line)
@@ -133,18 +176,18 @@ func (t *chatToMessages) line(line []byte) []byte {
 		return nil
 	}
 	if done {
-		// 汇聚最终 message_delta + content_block_stop + message_stop
+		// 汇聚最终：先对每个已开始未停止的块发 content_block_stop，再 message_delta + message_stop
+		var out [][]byte
+		for _, idx := range t.openBlocks {
+			out = append(out, eventLine("content_block_stop", map[string]any{"type": "content_block_stop", "index": idx}))
+		}
 		md := map[string]any{"type": "message_delta", "delta": map[string]any{
 			"stop_reason": stopReasonOrEnd(t.stopReason), "stop_sequence": nil,
 		}}
 		if t.usage != nil {
 			md["usage"] = t.usage
 		}
-		var out [][]byte
 		out = append(out, eventLine("message_delta", md))
-		if t.blockStarted {
-			out = append(out, eventLine("content_block_stop", map[string]any{"type": "content_block_stop", "index": 0}))
-		}
 		out = append(out, eventLine("message_stop", map[string]any{"type": "message_stop"}))
 		return joinLines(out)
 	}
@@ -167,11 +210,12 @@ func (t *chatToMessages) line(line []byte) []byte {
 	id, _ := m["id"].(string)
 
 	var out [][]byte
-	// 首个包含内容或角色的块前，先发 message_start + content_block_start
-	if !t.started {
-		role, _ := delta["role"].(string)
-		content, _ := delta["content"].(string)
-		if role == "assistant" || content != "" {
+	// message_start 只在实际有内容（文本或工具）时发出，避免 GLM 等模型的
+	// reasoning_content 块（带 role 无 content）提前开出一个空文本块。
+	ensureStarted := func() {
+		if t.started {
+			return
+		}
 		t.started = true
 		out = append(out, eventLine("message_start", map[string]any{
 			"type": "message_start",
@@ -180,34 +224,83 @@ func (t *chatToMessages) line(line []byte) []byte {
 				"model": t.model, "content": []any{}, "usage": map[string]any{"input_tokens": 0, "output_tokens": 0},
 			},
 		}))
-			out = append(out, eventLine("content_block_start", map[string]any{
-				"type": "content_block_start", "index": 0, "content_block": map[string]any{"type": "text", "text": ""},
-			}))
-			t.blockStarted = true
 	}
-	}
-	if content, _ := delta["content"].(string); content != "" {
-		if !t.started {
-			t.started = true
-			t.blockStarted = true
-			out = append(out, eventLine("message_start", map[string]any{
-				"type": "message_start",
-				"message": map[string]any{"id": "msg_" + strings.TrimPrefix(id, "chatcmpl-"), "type": "message", "role": "assistant", "model": t.model, "content": []any{}},
-			}))
+
+	// 文本：delta.content（string；兼容 {type:text,text} 数组）
+	if content := deltaText(delta); content != "" {
+		if t.textIndex < 0 {
+			t.textIndex = t.nextIndex
+			t.nextIndex++
+			ensureStarted()
 			out = append(out, eventLine("content_block_start", map[string]any{
-				"type": "content_block_start", "index": 0, "content_block": map[string]any{"type": "text", "text": ""},
+				"type": "content_block_start", "index": t.textIndex, "content_block": map[string]any{"type": "text", "text": ""},
 			}))
+			t.openBlocks = append(t.openBlocks, t.textIndex)
 		}
 		out = append(out, eventLine("content_block_delta", map[string]any{
-			"type": "content_block_delta", "index": 0, "delta": map[string]any{"type": "text_delta", "text": content},
+			"type": "content_block_delta", "index": t.textIndex, "delta": map[string]any{"type": "text_delta", "text": content},
 		}))
 	}
+
+	// 工具调用：delta.tool_calls（并行调用各 index 独立成块；arguments 支持整段/分段两种流式）
+	if tcs, ok := delta["tool_calls"].([]any); ok {
+		for _, tc := range tcs {
+			call, ok := tc.(map[string]any)
+			if !ok {
+				continue
+			}
+			idx, _ := call["index"].(float64)
+			tcIdx := int(idx)
+			fn, _ := call["function"].(map[string]any)
+			name, _ := fn["name"].(string)
+			args, _ := fn["arguments"].(string)
+			blockIdx, seen := t.toolIdx[tcIdx]
+			if !seen {
+				blockIdx = t.nextIndex
+				t.nextIndex++
+				t.toolIdx[tcIdx] = blockIdx
+				toolID, _ := call["id"].(string)
+				ensureStarted()
+				out = append(out, eventLine("content_block_start", map[string]any{
+					"type": "content_block_start", "index": blockIdx, "content_block": map[string]any{
+						"type": "tool_use", "id": toolID, "name": name, "input": map[string]any{},
+					},
+				}))
+				t.openBlocks = append(t.openBlocks, blockIdx)
+			}
+			if args != "" {
+				out = append(out, eventLine("content_block_delta", map[string]any{
+					"type": "content_block_delta", "index": blockIdx, "delta": map[string]any{"type": "input_json_delta", "partial_json": args},
+				}))
+			}
+		}
+	}
+
 	if u, ok := m["usage"]; ok {
 		t.usage = u
 	}
 	return joinLines(out)
 }
 
+// deltaText 取 chat delta.content 文本（string 或 [{type:text,text}] 数组拼接）。
+func deltaText(delta map[string]any) string {
+	if s, ok := delta["content"].(string); ok {
+		return s
+	}
+	if arr, ok := delta["content"].([]any); ok {
+		var parts []string
+		for _, b := range arr {
+			if bm, ok := b.(map[string]any); ok {
+				if t, _ := bm["text"].(string); t != "" {
+					parts = append(parts, t)
+				}
+			}
+		}
+		return strings.Join(parts, "")
+	}
+	return ""
+}
+
 func stopReasonOrEnd(s string) string {
 	if s == "" {
 		return "end_turn"
@@ -223,9 +316,16 @@ type responsesToMessages struct {
 	started    bool
 	model      string
 	usage      any
+	nextIndex  int            // 下一个 content block index（顺序分配）
+	textIndex  int            // 文本块 index；-1 = 未开始
+	toolIdx    map[string]int // function_call item_id → messages block index
+	openBlocks []int          // 已开始未停止的 block index，按开始顺序
+	anyTool    bool
 }
 
-func newResponsesToMessages() *responsesToMessages { return &responsesToMessages{} }
+func newResponsesToMessages() *responsesToMessages {
+	return &responsesToMessages{textIndex: -1, toolIdx: map[string]int{}}
+}
 
 func (t *responsesToMessages) line(line []byte) []byte {
 	isData, data, done := t.parseLine(line)
@@ -243,14 +343,16 @@ func (t *responsesToMessages) line(line []byte) []byte {
 		}
 	}
 	var out [][]byte
-	switch evt {
-	case "response.created":
-		if !t.started {
+	// message_start 只在 response.created 时发出；文本/工具块在对应事件到达时再开，
+	// 避免纯函数调用响应提前开出一个空文本块。
+	ensureStarted := func() {
+		if t.started {
+			return
+		}
 		t.started = true
-			id, _ := m["response"].(map[string]any)
 		rid := ""
-			if id != nil {
-				rid, _ = id["id"].(string)
+		if resp, ok := m["response"].(map[string]any); ok {
+			rid, _ = resp["id"].(string)
 		}
 		out = append(out, eventLine("message_start", map[string]any{
 			"type": "message_start",
@@ -259,22 +361,75 @@ func (t *responsesToMessages) line(line []byte) []byte {
 				"model": t.model, "content": []any{},
 			},
 		}))
-			out = append(out, eventLine("content_block_start", map[string]any{
-				"type": "content_block_start", "index": 0, "content_block": map[string]any{"type": "text", "text": ""},
-			}))
 	}
+	switch evt {
+	case "response.created":
+		ensureStarted()
 	case "response.output_text.delta":
 		delta, _ := m["delta"].(string)
-		if delta != "" {
-			out = append(out, eventLine("content_block_delta", map[string]any{
-				"type": "content_block_delta", "index": 0, "delta": map[string]any{"type": "text_delta", "text": delta},
-			}))
+		if delta == "" {
+			return nil
 		}
-	case "response.completed":
-		out = append(out, eventLine("message_delta", map[string]any{
-			"type": "message_delta", "delta": map[string]any{"stop_reason": "end_turn", "stop_sequence": nil},
+		if t.textIndex < 0 {
+			t.textIndex = t.nextIndex
+			t.nextIndex++
+			ensureStarted()
+			out = append(out, eventLine("content_block_start", map[string]any{
+				"type": "content_block_start", "index": t.textIndex, "content_block": map[string]any{"type": "text", "text": ""},
 			}))
-		out = append(out, eventLine("content_block_stop", map[string]any{"type": "content_block_stop", "index": 0}))
+			t.openBlocks = append(t.openBlocks, t.textIndex)
+		}
+		out = append(out, eventLine("content_block_delta", map[string]any{
+			"type": "content_block_delta", "index": t.textIndex, "delta": map[string]any{"type": "text_delta", "text": delta},
+		}))
+	case "response.output_item.added":
+		item, _ := m["item"].(map[string]any)
+		if item == nil || item["type"] != "function_call" {
+			return nil
+		}
+		blockIdx := t.nextIndex
+		t.nextIndex++
+		t.anyTool = true
+		itemID, _ := item["id"].(string)
+		t.toolIdx[itemID] = blockIdx
+		toolUseID, _ := item["call_id"].(string)
+		if toolUseID == "" {
+			toolUseID = itemID
+		}
+		name, _ := item["name"].(string)
+		ensureStarted()
+		out = append(out, eventLine("content_block_start", map[string]any{
+			"type": "content_block_start", "index": blockIdx, "content_block": map[string]any{
+				"type": "tool_use", "id": toolUseID, "name": name, "input": map[string]any{},
+			},
+		}))
+		t.openBlocks = append(t.openBlocks, blockIdx)
+	case "response.function_call_arguments.delta":
+		itemID, _ := m["item_id"].(string)
+		blockIdx, ok := t.toolIdx[itemID]
+		if !ok {
+			return nil
+		}
+		delta, _ := m["delta"].(string)
+		if delta == "" {
+			return nil
+		}
+		out = append(out, eventLine("content_block_delta", map[string]any{
+			"type": "content_block_delta", "index": blockIdx, "delta": map[string]any{"type": "input_json_delta", "partial_json": delta},
+		}))
+	case "response.completed":
+		for _, idx := range t.openBlocks {
+			out = append(out, eventLine("content_block_stop", map[string]any{"type": "content_block_stop", "index": idx}))
+		}
+		stop := "end_turn"
+		if t.anyTool {
+			stop = "tool_use"
+		}
+		md := map[string]any{"type": "message_delta", "delta": map[string]any{"stop_reason": stop, "stop_sequence": nil}}
+		if t.usage != nil {
+			md["usage"] = t.usage
+		}
+		out = append(out, eventLine("message_delta", md))
 		out = append(out, eventLine("message_stop", map[string]any{"type": "message_stop"}))
 	}
 	return joinLines(out)

server/internal/proxy/gateway.go

@@ -35,6 +35,7 @@ type Gateway struct {
 	enc     *crypto.Encryptor
 	lim     *ratelimit.Limiter
 	userRPS int
+	logRaw  bool
 	hc      *http.Client
 
 	policyMu sync.Mutex
@@ -108,7 +109,7 @@ func contains(list []string, s string) bool {
 	return false
 }
 
-func NewGateway(db *gorm.DB, enc *crypto.Encryptor, rec *usage.Recorder, lim *ratelimit.Limiter, userRPS int) *Gateway {
+func NewGateway(db *gorm.DB, enc *crypto.Encryptor, rec *usage.Recorder, lim *ratelimit.Limiter, userRPS int, logRaw bool, gw_timeout time.Duration) *Gateway {
 	return &Gateway{
 		db:      db,
 		ch:      channel.NewService(db, enc),
@@ -116,7 +117,8 @@ func NewGateway(db *gorm.DB, enc *crypto.Encryptor, rec *usage.Recorder, lim *ra
 		enc:     enc,
 		lim:     lim,
 		userRPS: userRPS,
-		hc:      &http.Client{Timeout: 120 * time.Second},
+		logRaw:  logRaw,
+		hc:      &http.Client{Timeout: gw_timeout * time.Second},
 	}
 }
 
@@ -135,8 +137,12 @@ func (g *Gateway) Auth(c *gin.Context) {
 	auth := c.GetHeader("Authorization")
 	key := strings.TrimPrefix(auth, "Bearer ")
 	key = strings.TrimSpace(key)
+	if key == "" {
+		// Anthropic 客户端（Claude Code / SDK）用 x-api-key 头而不是 Authorization
+		key = strings.TrimSpace(c.GetHeader("x-api-key"))
+	}
 	if !apikey.Valid(key) {
-		apiError(c, http.StatusUnauthorized, "invalid_api_key", "Invalid API key format. Expected: Bearer sk-...")
+		apiError(c, http.StatusUnauthorized, "invalid_api_key", "Invalid API key format. Expected: Authorization: Bearer sk-ot-... or x-api-key: sk-ot-...")
 		c.Abort()
 		return
 	}
@@ -201,10 +207,15 @@ func (g *Gateway) Handle(c *gin.Context) {
 	}
 }
 
-// models GET /v1/models：返回启用的全局模型（OpenAI 风格）。
+// models GET /v1/models：返回对外可见的模型（启用且至少绑定到一个启用且健康的渠道；OpenAI 风格）。
 func (g *Gateway) models(c *gin.Context) {
+	ids := g.ch.AvailableModelIDs()
+	if len(ids) == 0 {
+		c.JSON(http.StatusOK, gin.H{"object": "list", "data": []gin.H{}})
+		return
+	}
 	var ms []store.Model
-	if err := g.db.Where("enabled = ?", true).Order("sort ASC, id ASC").Find(&ms).Error; err != nil {
+	if err := g.db.Where("id IN ? AND enabled = ?", ids, true).Order("sort ASC, id ASC").Find(&ms).Error; err != nil {
 		apiError(c, http.StatusInternalServerError, "internal_error", "failed to load models")
 		return
 	}

server/internal/proxy/handlers.go

@@ -5,6 +5,7 @@ import (
 
 	"github.com/gin-gonic/gin"
 	"github.com/openteam/server/internal/proxy/convert"
+	"github.com/openteam/server/internal/store"
 )
 
 // chatCompletions POST /v1/chat/completions
@@ -23,6 +24,7 @@ func (g *Gateway) chatCompletions(c *gin.Context) {
 	}
 	c.Set("protocol", convert.ProtoChat)
 	c.Set("model_name", br.Model)
+	g.recordRawRequest(c, u, body)
 	if !g.checkModelAllowed(u, br.Model) {
 		apiError(c, http.StatusForbidden, "model_not_allowed", "模型未对你开放，请联系管理员")
 		return
@@ -55,6 +57,7 @@ func (g *Gateway) responses(c *gin.Context) {
 	}
 	c.Set("protocol", convert.ProtoResponses)
 	c.Set("model_name", br.Model)
+	g.recordRawRequest(c, u, body)
 	if !g.checkModelAllowed(u, br.Model) {
 		apiError(c, http.StatusForbidden, "model_not_allowed", "模型未对你开放，请联系管理员")
 		return
@@ -87,6 +90,7 @@ func (g *Gateway) messages(c *gin.Context) {
 	}
 	c.Set("protocol", convert.ProtoMessages)
 	c.Set("model_name", br.Model)
+	g.recordRawRequest(c, u, body)
 	if !g.checkModelAllowed(u, br.Model) {
 		apiError(c, http.StatusForbidden, "model_not_allowed", "模型未对你开放，请联系管理员")
 		return
@@ -108,6 +112,14 @@ type sinkHolder struct {
 	sink *usageSink
 }
 
+// recordRawRequest 记录管理员原始请求体到 context（供 finishUsage 落库）。
+// 仅当开关开启且用户为管理员时记录；响应侧以 c.Get("raw_request") 是否非空判断是否需要捕获响应。
+func (g *Gateway) recordRawRequest(c *gin.Context, u *store.User, body []byte) {
+	if g.logRaw && u.Role == store.RoleAdmin {
+		c.Set("raw_request", string(body))
+	}
+}
+
 // apiError 按客户端协议返回错误体（PLANNING §5.1.4）。
 func apiError(c *gin.Context, status int, code, message string) {
 	if p, _ := c.Get("protocol"); p == convert.ProtoMessages {

server/internal/proxy/passthrough.go

@@ -3,7 +3,6 @@ package proxy
 import (
 	"bufio"
 	"bytes"
-	"context"
 	"crypto/rand"
 	"encoding/hex"
 	"encoding/json"
@@ -101,14 +100,32 @@ func requestText(body []byte) string {
 	return strings.Join(parts, "\n")
 }
 
-// sseContentText 提取一条 SSE data 行中的内容文本（chat delta.content / responses delta / messages delta.text）。
-func sseContentText(line []byte) string {
+// sseDataPayload 提取一条 SSE 的 JSON 载荷（去掉 data: 前缀与空白）。
+// 兼容三种写法：
+//   - 单 data: 行：data: {...} 或 data:{...}（上游如火山方舟会省略 data: 后的空格）
+//   - event:+data: 多行块：转换器 eventLine 产出的块（event: xxx\ndata: {...} 拼在一个 []byte）
+func sseDataPayload(line []byte) (string, bool) {
 	s := string(line)
-	if strings.HasPrefix(s, "data: ") {
-		s = strings.TrimPrefix(s, "data: ")
+	idx := strings.LastIndex(s, "\ndata:")
+	if idx >= 0 {
+		s = s[idx+len("\ndata:"):] // 跳过 event: 头，落在 data: 之后
+	} else if strings.HasPrefix(s, "data:") {
+		s = strings.TrimPrefix(s, "data:")
+	} else {
+		return "", false
 	}
+	s = strings.TrimLeft(s, " ") // data: 后的可选空格
 	s = strings.TrimSpace(s)
 	if s == "" || s == "[DONE]" {
+		return "", false
+	}
+	return s, true
+}
+
+// sseContentText 提取一条 SSE 中的内容文本（chat delta.content / responses delta / messages delta.text）。
+func sseContentText(line []byte) string {
+	s, ok := sseDataPayload(line)
+	if !ok {
 		return ""
 	}
 	var m map[string]any
@@ -146,11 +163,29 @@ func upstreamURL(ch *store.Channel, proto, path string) string {
 	return ch.UpstreamURL(proto, path)
 }
 
-// doProxy 通用代理（M5）：遍历候选渠道，按需转换；可安全重试的失败自动故障转移。
+// doProxy 通用代理（M5）：加权随机选一个候选渠道；失败自动故障转移。
 func (g *Gateway) doProxy(c *gin.Context, cands []channel.Candidate, clientProto string, body []byte, stream bool, sink *usageSink) {
+	if len(cands) == 0 {
+		apiError(c, http.StatusBadGateway, "upstream_error", "no available channel")
+		g.recordError(c, nil, nil, now(), "no_available_channel")
+		return
+	}
+	// 加权随机选择起始渠道
+	picked := g.ch.Pick(cands)
+	startIdx := 0
+	for i, cand := range cands {
+		if cand.Channel.ID == picked.ID {
+			startIdx = i
+			break
+		}
+	}
+
 	var lastStatus = http.StatusBadGateway
 	var lastBody = []byte("all upstream channels failed")
-	for _, cand := range cands {
+	// 从选中的渠道开始遍历，到末尾后再从头遍历到选中渠道之前
+	for offset := 0; offset < len(cands); offset++ {
+		idx := (startIdx + offset) % len(cands)
+		cand := cands[idx]
 		ch := cand.Channel
 		plan, err := prepareUpstream(ch, clientProto, body, cand.UpstreamModel)
 		if err != nil {
@@ -200,9 +235,7 @@ func (g *Gateway) proxyOne(c *gin.Context, ch *store.Channel, plan *upstreamPlan
 		}
 	}
 
-	ctx, cancel := context.WithTimeout(c.Request.Context(), time.Duration(ch.TimeoutMS)*time.Millisecond)
-	defer cancel()
-	req, err := http.NewRequestWithContext(ctx, http.MethodPost, upstreamURL(ch, plan.proto, plan.path), bytes.NewReader(upBody))
+	req, err := http.NewRequest(http.MethodPost, upstreamURL(ch, plan.proto, plan.path), bytes.NewReader(upBody))
 	if err != nil {
 		return false, false, http.StatusInternalServerError, []byte("failed to build upstream request")
 	}
@@ -224,9 +257,6 @@ func (g *Gateway) proxyOne(c *gin.Context, ch *store.Channel, plan *upstreamPlan
 	start := time.Now()
 	resp, err := g.hc.Do(req)
 	if err != nil {
-		if ctx.Err() == context.DeadlineExceeded {
-			return false, true, http.StatusGatewayTimeout, []byte("upstream request timed out")
-		}
 		return false, true, http.StatusBadGateway, []byte("upstream request failed: " + err.Error())
 	}
 	defer resp.Body.Close()
@@ -268,6 +298,9 @@ func (g *Gateway) copyAndCapture(c *gin.Context, ch *store.Channel, r io.Reader,
 		}
 	}
 	_, _ = c.Writer.Write(out)
+	if _, ok := c.Get("raw_request"); ok {
+		c.Set("raw_response", string(data)) // 上游原始响应（未转换）
+	}
 	g.finishUsage(c, ch, start, store.UsageStatusSuccess, "")
 }
 
@@ -279,10 +312,24 @@ func (g *Gateway) streamCopy(c *gin.Context, ch *store.Channel, r io.Reader, sta
 		flusher = nopFlusher{}
 	}
 
+	// 原始响应捕获：仅管理员且开关开启（raw_request 已 set）时累积上游原始行
+	_, capture := c.Get("raw_request")
+	var rawResp strings.Builder
+
+	// commitRaw 在记账前把已累积的原始响应写入 context
+	commitRaw := func() {
+		if capture {
+			c.Set("raw_response", rawResp.String())
+		}
+	}
+
 	scanner := newSSEScanner(r)
 	for {
 		line, err := scanner.Next()
 		if line != nil {
+			if capture {
+				rawResp.Write(line)
+			}
 			out := line
 			if lineConv != nil {
 				out = lineConv(line)
@@ -290,6 +337,7 @@ func (g *Gateway) streamCopy(c *gin.Context, ch *store.Channel, r io.Reader, sta
 			if out != nil {
 				if _, werr := w.Write(out); werr != nil {
 					// 客户端意外断开：按已生成部分收费（canceled）
+					commitRaw()
 					g.finishUsage(c, ch, start, store.UsageStatusCanceled, "client_disconnect")
 					return
 				}
@@ -303,6 +351,7 @@ func (g *Gateway) streamCopy(c *gin.Context, ch *store.Channel, r io.Reader, sta
 			}
 		}
 		if err != nil {
+			commitRaw()
 			if err == io.EOF {
 				g.finishUsage(c, ch, start, store.UsageStatusSuccess, "")
 			} else if c.Request.Context().Err() != nil {
@@ -371,15 +420,11 @@ func extractUsage(data []byte) json.RawMessage {
 
 // scanUsage 从 SSE 一行中提取 usage（OpenAI 末块 / responses completed / messages message_delta 等）。
 func scanUsage(line []byte) json.RawMessage {
-	s := string(line)
-	if !strings.Contains(s, `"usage"`) {
+	if !bytes.Contains(line, []byte(`"usage"`)) {
 		return nil
 	}
-	if strings.HasPrefix(s, "data: ") {
-		s = strings.TrimPrefix(s, "data: ")
-	}
-	s = strings.TrimSpace(s)
-	if s == "[DONE]" || s == "" {
+	s, ok := sseDataPayload(line)
+	if !ok {
 		return nil
 	}
 	var m map[string]json.RawMessage
@@ -400,7 +445,11 @@ func scanUsage(line []byte) json.RawMessage {
 	return nil
 }
 
-// usageFromMap 从 map 顶层或 message 子对象中取 usage。
+// usageFromMap 从 map 顶层、message 或 delta 子对象中取 usage。
+// 覆盖三种形态：
+//   - OpenAI 流式末块顶层 usage
+//   - Anthropic 非流式 / message_start 的 message.usage
+//   - Anthropic 流式 message_delta 的 delta.usage（真实 token 计数所在）
 func usageFromMap(m map[string]json.RawMessage) json.RawMessage {
 	if u, ok := m["usage"]; ok && string(u) != "null" {
 		return u
@@ -413,6 +462,14 @@ func usageFromMap(m map[string]json.RawMessage) json.RawMessage {
 			}
 		}
 	}
+	if deltaRaw, ok := m["delta"]; ok {
+		var delta map[string]json.RawMessage
+		if json.Unmarshal(deltaRaw, &delta) == nil {
+			if u, ok := delta["usage"]; ok && string(u) != "null" {
+				return u
+			}
+		}
+	}
 	return nil
 }
 
@@ -452,6 +509,14 @@ func (u *usageSink) push(raw json.RawMessage) {
 	if json.Unmarshal(raw, &t) != nil {
 		return
 	}
+	// messages 流式最终事件（message_delta 的 usage）带 cache_* 字段，是上游的最终计费口径，
+	// 其中 input_tokens 仅指"非缓存输入"（与 message_start 的"总输入"语义不同）。
+	// 整体替换而非字段合并，避免 delta 的非缓存 input 覆盖 start 的总 input 后语义错乱
+	// （实际消耗由 finishUsage 按 input + cache_read + cache_creation 汇总）。
+	if t.CacheReadInputTokens > 0 || t.CacheCreationInputTokens > 0 {
+		u.us = t
+		return
+	}
 	// 零值不覆盖：不同事件携带不同字段
 	if t.PromptTokens > 0 {
 		u.us.PromptTokens = t.PromptTokens
@@ -499,11 +564,15 @@ func (g *Gateway) finishUsage(c *gin.Context, ch *store.Channel, start time.Time
 	modelName, _ := c.Get("model_name")
 	mn, _ := modelName.(string)
 
-	// 流式中断（canceled）：上游最终 usage 可能未返回，按已生成内容用 tiktoken 估算
-	if status == store.UsageStatusCanceled {
-		if in == 0 {
+	// 上游未返回真实 usage 时估算（tiktoken）：
+	//   - 流式中断（canceled）：上游最终 usage 可能未返回
+	//   - 流式正常结束（success）：部分上游（如火山方舟 Anthropic 流式）message_start.usage 恒为 0、
+	//     message_delta 不带 usage，只能按已收发内容估算，否则记账为 0 消耗
+	// 非流式上游必返回 usage，此处 in/out 非 0 不受影响。
+	if status == store.UsageStatusCanceled || status == store.UsageStatusSuccess {
+		if in == 0 && cacheRead == 0 && cacheCreate == 0 {
 			if est, ok := c.Get("est_input_text"); ok {
-				if v, ok2 := est.(string); ok2 {
+				if v, ok2 := est.(string); ok2 && v != "" {
 					in = int64(tokenizer.Count(v, mn))
 				}
 			}
@@ -519,12 +588,18 @@ func (g *Gateway) finishUsage(c *gin.Context, ch *store.Channel, start time.Time
 	_ = g.db.Where("name = ?", mn).First(&model).Error
 	if model.ID > 0 {
 		modelID = model.ID
+		// 计价口径：in=非缓存输入、cacheRead=缓存读、cacheCreate=缓存写（Anthropic 语义，
+		// messages 流式 message_delta 的 input_tokens 即非缓存部分）。
+		// 缓存写按 1.25× 输入价（Anthropic 5m 口径）。
 		cost = float64(in)/1e6*model.InputPrice +
 			float64(out)/1e6*model.OutputPrice +
-			float64(cacheRead)/1e6*model.CacheReadPrice
+			float64(cacheRead)/1e6*model.CacheReadPrice +
+			float64(cacheCreate)/1e6*model.InputPrice*1.25
 	} else {
-		cost = float64(in)/1e6*0.15 + float64(out)/1e6*0.60 // 无定价模型时按示例价
+		cost = float64(in+cacheRead+cacheCreate)/1e6*0.15 + float64(out)/1e6*0.60 // 无定价模型时按示例价
 	}
+	// 落库的 input_tokens 存输入总量（含缓存读/写），与上游 message_start 口径一致，便于对账展示。
+	in += cacheRead + cacheCreate
 
 	proto, _ := c.Get("protocol")
 	p, _ := proto.(string)
@@ -555,6 +630,14 @@ func (g *Gateway) finishUsage(c *gin.Context, ch *store.Channel, start time.Time
 		chID = ch.ID
 	}
 
+	var rawReq, rawResp string
+	if v, ok := c.Get("raw_request"); ok {
+		rawReq, _ = v.(string)
+	}
+	if v, ok := c.Get("raw_response"); ok {
+		rawResp, _ = v.(string)
+	}
+
 	// 密钥今日 token 用量累计（配额检查用）
 	if g.lim != nil && kidVal > 0 {
 		g.lim.AddTokens(kidVal, in+out)
@@ -580,6 +663,8 @@ func (g *Gateway) finishUsage(c *gin.Context, ch *store.Channel, start time.Time
 		LatencyMS:           latency,
 		Status:              status,
 		ErrorCode:           errCodePtr,
+		RawRequest:          rawReq,
+		RawResponse:         rawResp,
 		CreatedAt:           time.Now().UTC(),
 	})
 }

server/internal/proxy/sse_fix_test.go

@@ -0,0 +1,101 @@
+package proxy
+
+import (
+	"encoding/json"
+	"testing"
+)
+
+// 复现线上火山方舟 qwen 流：data:{...} 无空格（省略 data: 后的空格）。
+func TestSSEDataPayloadNoSpace(t *testing.T) {
+	lines := []string{
+		`data:{"message":{"model":"qwen3.8-flash","id":"msg_1","role":"assistant","type":"message","content":[],"usage":{"input_tokens":31626,"output_tokens":0}},"type":"message_start"}`,
+		`data:{"delta":{"type":"text_delta","text":"你好"},"type":"content_block_delta","index":0}`,
+		`data:{"delta":{"type":"text_delta","text":"!"},"type":"content_block_delta","index":0}`,
+		`data:{"delta":{"stop_reason":"end_turn"},"type":"message_delta","usage":{"cache_creation":{"ephemeral_5m_input_tokens":33065},"output_tokens":8,"cache_creation_input_tokens":33065,"input_tokens":8,"cache_read_input_tokens":0}}`,
+	}
+	var out string
+	for _, l := range lines {
+		out += sseContentText([]byte(l))
+	}
+	if out != "你好!" {
+		t.Fatalf("outputText=%q, want %q", out, "你好!")
+	}
+	// message_delta 的 delta.usage 应能提取（output_tokens=8）
+	u := scanUsage([]byte(lines[3]))
+	if u == nil {
+		t.Fatal("scanUsage returned nil for message_delta with usage")
+	}
+	var sh usageShape
+	if err := json.Unmarshal(u, &sh); err != nil {
+		t.Fatalf("unmarshal usage: %v", err)
+	}
+	if sh.OutputTokens != 8 {
+		t.Fatalf("output_tokens=%d, want 8", sh.OutputTokens)
+	}
+}
+
+// 兼容带空格的单 data: 行（标准 SSE）与 event:+data: 多行块。
+func TestSSEDataPayloadSpacedAndMultiLine(t *testing.T) {
+	// 标准：data: {...}
+	if got := sseContentText([]byte(`data: {"delta":{"type":"text_delta","text":"hi"},"type":"content_block_delta","index":0}`)); got != "hi" {
+		t.Fatalf("spaced single line: got %q, want hi", got)
+	}
+	// 多行块：event: message_delta\ndata: {...}
+	block := []byte("event: message_delta\ndata: {\"delta\":{\"type\":\"text_delta\",\"text\":\"yo\"},\"type\":\"content_block_delta\",\"index\":0}\n")
+	if got := sseContentText(block); got != "yo" {
+		t.Fatalf("multiline block: got %q, want yo", got)
+	}
+}
+
+// 复现线上 qwen（dashscope）messages 流式缓存场景：
+// message_start.usage.input_tokens 是总输入，message_delta.usage.input_tokens 是非缓存输入
+// 且带 cache_read/cache_creation，是最终计费口径。合并后：
+//   in(落库)=input+cache_read+cache_creation，计价 in 只算非缓存部分。
+// 此前 delta 的 input 覆盖 start 的 input 导致总输入丢失（31790 → 8）。
+func TestUsageSinkMessageDeltaAuthoritative(t *testing.T) {
+	sink := &usageSink{}
+	// message_start：总输入 31790
+	start := json.RawMessage(`{"input_tokens":31790,"output_tokens":0}`)
+	sink.push(start)
+	if got := sink.us.InputTokens; got != 31790 {
+		t.Fatalf("after start: input=%d, want 31790", got)
+	}
+	// message_delta：非缓存输入 8 + 缓存写 33229（最终口径，整体替换）
+	delta := json.RawMessage(`{"output_tokens":8,"cache_creation_input_tokens":33229,"input_tokens":8,"cache_read_input_tokens":0}`)
+	sink.push(delta)
+	s := sink.Shape()
+	if s.InputTokens != 8 || s.CacheCreationInputTokens != 33229 || s.OutputTokens != 8 {
+		t.Fatalf("after delta: %+v, want input=8 cache_create=33229 output=8", s)
+	}
+	// finishUsage 口径：落库 input = 8 + 0 + 33229 = 33237（总量），计价 in=8、cacheCreate=33229
+	in := s.InputTokens + s.CacheReadInputTokens + s.CacheCreationInputTokens
+	if in != 33237 {
+		t.Fatalf("total input=%d, want 33237", in)
+	}
+}
+
+// 缓存命中场景（id=55）：delta input=76 非缓存 + cache_read=33229 + cache_creation=17。
+func TestUsageSinkCacheHitMerge(t *testing.T) {
+	sink := &usageSink{}
+	sink.push(json.RawMessage(`{"input_tokens":31862,"output_tokens":0}`))
+	sink.push(json.RawMessage(`{"output_tokens":32,"cache_creation_input_tokens":17,"input_tokens":76,"cache_read_input_tokens":33229}`))
+	s := sink.Shape()
+	total := s.InputTokens + s.CacheReadInputTokens + s.CacheCreationInputTokens
+	if total != 33322 {
+		t.Fatalf("total input=%d, want 33322 (76+33229+17)", total)
+	}
+	if s.OutputTokens != 32 {
+		t.Fatalf("output=%d, want 32", s.OutputTokens)
+	}
+}
+
+// OpenAI chat 末块（无 cache 字段）仍走零值不覆盖合并，不受整体替换影响。
+func TestUsageSinkChatLastChunkStillMerges(t *testing.T) {
+	sink := &usageSink{}
+	sink.push(json.RawMessage(`{"prompt_tokens":65,"completion_tokens":0}`))
+	sink.push(json.RawMessage(`{"prompt_tokens":65,"completion_tokens":82}`))
+	s := sink.Shape()
+	if s.PromptTokens != 65 || s.CompletionTokens != 82 {
+		t.Fatalf("chat merge broken: %+v", s)
+	}
+}

server/internal/store/db.go

@@ -1,6 +1,7 @@
 package store
 
 import (
+	"fmt"
 	"log"
 	"os"
 	"path/filepath"
@@ -11,6 +12,12 @@ import (
 	gormlogger "gorm.io/gorm/logger"
 )
 
+// schemaVersion 当前 schema 版本。struct 变更（加列/改列/删列）时递增，
+// 触发一次 AutoMigrate 并把新版本写入库（SQLite 用 PRAGMA user_version）。
+// AutoMigrate 对已有表的列判定不收敛（每次都重建表：CREATE __temp + INSERT SELECT + DROP），
+// 大表上一次重建数十秒且每次重启重演，所以之后版本未变就直接跳过。
+const schemaVersion = 2
+
 // Open 打开数据库连接并自动迁移。
 // 开发默认 SQLite（dsn 支持 file:...?_journal_mode=WAL），生产可切 postgres。
 func Open(driver, dsn string) (*gorm.DB, error) {
@@ -33,13 +40,35 @@ func Open(driver, dsn string) (*gorm.DB, error) {
 		return nil, err
 	}
 
+	if driver != "postgres" && currentSQLiteVersion(db) >= schemaVersion {
+		log.Printf("store: connected driver=%s (schema up-to-date v%d, skip migrate)", driver, schemaVersion)
+		return db, nil
+	}
+
 	if err := db.AutoMigrate(AllModels()...); err != nil {
 		return nil, err
 	}
-	log.Printf("store: connected driver=%s (migrated)", driver)
+	if driver != "postgres" {
+		setSQLiteVersion(db, schemaVersion)
+	}
+	log.Printf("store: connected driver=%s (migrated, schema v%d)", driver, schemaVersion)
+	// 将已有渠道的超时时间从 120000ms 更新为 300000ms（幂等操作）
+	db.Model(&Channel{}).Where("timeout_ms = ?", 120000).Update("timeout_ms", 300000)
 	return db, nil
 }
 
+// currentSQLiteVersion 读取 PRAGMA user_version。
+func currentSQLiteVersion(db *gorm.DB) int {
+	var v int
+	db.Raw("PRAGMA user_version").Scan(&v)
+	return v
+}
+
+// setSQLiteVersion 写入 PRAGMA user_version。
+func setSQLiteVersion(db *gorm.DB, v int) {
+	db.Exec(fmt.Sprintf("PRAGMA user_version = %d", v))
+}
+
 // sqliteDir 提取 SQLite DSN 中的目录部分（忽略 file: 前缀与查询参数）。
 func sqliteDir(dsn string) string {
 	d := dsn

server/internal/store/models.go

@@ -92,7 +92,7 @@ type Channel struct {
 	APIKeyEnc      string            `gorm:"size:1024;not null" json:"-"` // AES-GCM 密文
 	Weight         int       `gorm:"not null;default:1" json:"weight"`
 	Priority       int       `gorm:"not null;default:0" json:"priority"` // 数值小优先
-	TimeoutMS      int       `gorm:"not null;default:120000" json:"timeout_ms"`
+	TimeoutMS      int       `gorm:"not null;default:300000" json:"timeout_ms"`
 	MaxConcurrency int       `gorm:"not null;default:16" json:"max_concurrency"`
 	HealthStatus   string    `gorm:"size:16;not null;default:healthy" json:"health_status"`
 	Enabled        bool      `gorm:"not null;default:true" json:"enabled"`
@@ -145,7 +145,6 @@ func (c *Channel) UpstreamURL(proto, path string) string {
 type Model struct {
 	ID             uint64    `gorm:"primaryKey;autoIncrement" json:"id"`
 	Name           string    `gorm:"uniqueIndex;size:128;not null" json:"name"`
-	DisplayName    string    `gorm:"size:128" json:"display_name"`
 	InputPrice     float64   `gorm:"type:numeric(20,8);not null;default:0" json:"input_price"`
 	OutputPrice    float64   `gorm:"type:numeric(20,8);not null;default:0" json:"output_price"`
 	CacheReadPrice float64   `gorm:"type:numeric(20,8);not null;default:0" json:"cache_read_price"`
@@ -188,6 +187,8 @@ type UsageLog struct {
 	LatencyMS           int       `json:"latency_ms"`
 	Status              string    `gorm:"size:16;not null" json:"status"`
 	ErrorCode           *string   `json:"error_code,omitempty"`
+	RawRequest          string    `gorm:"type:text" json:"raw_request"`   // 客户端原始请求体（未转换）
+	RawResponse         string    `gorm:"type:text" json:"raw_response"`  // 上游原始响应（未转换；流式为全部 SSE 事件）
 	CreatedAt           time.Time `gorm:"index" json:"created_at"`
 }
 

web/src/components/layout/ShellLayout.vue

@@ -116,7 +116,7 @@ async function logout() {
       </div>
     </aside>
 
-    <div class="flex min-h-[100dvh] flex-1 flex-col md:ml-56">
+    <div class="flex min-h-[100dvh] min-w-0 flex-1 flex-col md:ml-56">
       <header class="flex h-14 items-center justify-between gap-2 border-b border-edge px-4 md:px-6">
         <div class="flex min-w-0 items-center gap-2">
           <button
@@ -142,7 +142,7 @@ async function logout() {
         </div>
       </header>
 
-      <main class="flex-1 px-4 py-6 md:px-6">
+      <main class="min-w-0 flex-1 px-4 py-6 md:px-6">
         <router-view />
       </main>
     </div>

web/src/components/ui/Input.vue

@@ -9,6 +9,7 @@ withDefaults(
     error?: string
     autocomplete?: string
     disabled?: boolean
+    maxlength?: number
   }>(),
   { type: 'text', modelValue: '', disabled: false },
 )
@@ -24,6 +25,7 @@ const emit = defineEmits<{ 'update:modelValue': [string | number] }>()
       :placeholder="placeholder"
       :autocomplete="autocomplete"
       :disabled="disabled"
+      :maxlength="maxlength"
       class="h-10 w-full rounded-md border border-edge2 bg-surface px-3 text-sm text-ink placeholder-muted outline-none transition focus:border-accent focus:ring-2 focus:ring-accent disabled:cursor-not-allowed disabled:opacity-50"
       :class="error && 'border-err focus:border-err focus:ring-err'"
       @input="emit('update:modelValue', ($event.target as HTMLInputElement).value as string | number)"

web/src/types.ts

@@ -59,7 +59,6 @@ export interface ModelBinding {
 export interface Model {
   id: number
   name: string
-  display_name: string
   input_price: number
   output_price: number
   cache_read_price: number
@@ -83,6 +82,8 @@ export interface UsageLog {
   request_id: string
   model: string
   protocol: string
+  channel_id?: number
+  channel?: string
   input_tokens: number
   output_tokens: number
   cache_read_tokens: number
@@ -90,8 +91,11 @@ export interface UsageLog {
   latency_ms: number
   status: string
   error_code: string | null
+  raw_request?: string
+  raw_response?: string
   created_at: string
   user?: string
+  user_id?: number
 }
 
 export interface Paged<T> {

web/src/views/LandingView.vue

@@ -2,6 +2,10 @@
 import { useAuthStore } from '@/stores/auth'
 import ThemeToggle from '@/components/ui/ThemeToggle.vue'
 const auth = useAuthStore()
+
+// 调用演示用浏览器当前 host（兼容端点），避免写死域名
+const host = window.location.host
+const baseURL = `${window.location.origin}/v1`
 </script>
 
 <template>
@@ -70,12 +74,12 @@ const auth = useAuthStore()
           <div class="flex items-center gap-1.5 border-b border-edge px-4 py-2.5">
             <span class="size-2.5 rounded-full bg-surface2" />
             <span class="size-2.5 rounded-full bg-surface2" />
-            <span class="ml-2 text-muted">curl api.openteam.dev</span>
+            <span class="ml-2 text-muted">curl {{ host }}</span>
           </div>
           <div class="space-y-3 p-4 leading-relaxed">
             <div>
-              <p class="text-muted"><span class="text-accent">$</span> curl https://api.openteam.dev/v1/chat/completions</p>
-              <p class="text-muted">  -H <span class="text-accent">"Authorization: Bearer sk-..."</span> \</p>
+              <p class="text-muted"><span class="text-accent">$</span> curl {{ baseURL }}/chat/completions</p>
+              <p class="text-muted">  -H <span class="text-accent">"Authorization: Bearer sk-ot-..."</span> \</p>
               <p class="text-muted">  -d <span class="text-ink">'{"model": "claude-sonnet-5", "messages": [{"role": "user", "content": "你好"}]}'</span></p>
             </div>
             <div class="border-t border-edge pt-3 text-muted">
@@ -132,7 +136,7 @@ const auth = useAuthStore()
             </p>
           </div>
           <div class="card p-5">
-            <p class="font-mono text-xs text-muted">sk-…</p>
+            <p class="font-mono text-xs text-muted">sk-ot-…</p>
             <h3 class="mt-2 text-lg font-semibold">API Key 管理</h3>
             <p class="mt-2 text-sm leading-relaxed text-muted">
               密钥仅存哈希，支持配额、过期与模型白名单，创建时一次性展示。

web/src/views/admin/ChannelsView.vue

@@ -31,7 +31,7 @@ const form = reactive({
   api_key: '',
   weight: 1,
   priority: 0,
-  timeout_ms: 120000,
+  timeout_ms: 300000,
   max_concurrency: 16,
   enabled: true,
 })
@@ -51,7 +51,7 @@ function openCreate() {
     name: '', formats: ['chat'], base_url: '',
     base_urls: { chat: '', responses: '', messages: '' },
     api_key: '',
-    weight: 1, priority: 0, timeout_ms: 120000, max_concurrency: 16, enabled: true,
+    weight: 1, priority: 0, timeout_ms: 300000, max_concurrency: 16, enabled: true,
   })
   editOpen.value = true
 }
@@ -142,7 +142,7 @@ onMounted(load)
 
     <!-- 移动端：卡片列表 -->
     <div class="space-y-3 md:hidden">
-      <div v-for="ch in channels" :key="ch.id" class="card p-4">
+      <div v-for="ch in channels" :key="ch.id" class="card p-4" :class="ch.enabled ? 'bg-accent-soft' : ''">
         <div class="flex flex-wrap items-start justify-between gap-2">
           <div class="min-w-0">
             <p class="text-sm font-medium text-ink">{{ ch.name }}</p>
@@ -206,7 +206,7 @@ onMounted(load)
           </thead>
           <tbody>
             <template v-for="ch in channels" :key="ch.id">
-            <tr class="table-row">
+            <tr class="table-row" :style="ch.enabled ? { borderLeft: '2px solid var(--color-accent)' } : {}">
               <td class="px-4 py-2.5">
                 <button class="inline-flex items-center gap-1.5 text-ink transition hover:text-accent" @click="toggleDrawer(ch)">
                   <span class="truncate">{{ ch.name }}</span>
@@ -222,7 +222,7 @@ onMounted(load)
                   >{{ protocolShort(f) }}</code>
                 </div>
               </td>
-              <td class="px-4 py-2.5 font-mono text-xs text-muted">{{ ch.base_url }}</td>
+              <td class="max-w-[220px] truncate px-4 py-2.5 font-mono text-xs text-muted">{{ ch.base_url }}</td>
               <td class="px-4 py-2.5 font-mono text-xs text-muted">{{ ch.api_key_masked || '****' }}</td>
               <td class="px-4 py-2.5">
                 <Badge :variant="ch.health_status === 'healthy' ? 'ok' : ch.health_status === 'cooldown' ? 'err' : 'warn'">
@@ -288,13 +288,14 @@ onMounted(load)
           v-model="form.base_url"
           label="Base URL（可选）"
           placeholder="https://api.openai.com/v1"
+          :maxlength="255"
           hint="支持前缀或完整端点，如 https://api.openai.com/v1 或 https://api.openai.com/v1/chat/completions；留空按供应商默认"
         />
         <div class="space-y-3 rounded-md border border-edge p-3">
           <p class="text-xs font-medium text-muted">分协议 Base URL（可选，如智谱三种格式不同）</p>
-          <Input v-model="form.base_urls.chat" label="OpenAI Chat Completions" placeholder="留空用主 Base URL" />
-          <Input v-model="form.base_urls.responses" label="OpenAI Responses" placeholder="留空用主 Base URL" />
-          <Input v-model="form.base_urls.messages" label="Anthropic Messages" placeholder="留空用主 Base URL" />
+          <Input v-model="form.base_urls.chat" label="OpenAI Chat Completions" placeholder="留空用主 Base URL" :maxlength="255" />
+          <Input v-model="form.base_urls.responses" label="OpenAI Responses" placeholder="留空用主 Base URL" :maxlength="255" />
+          <Input v-model="form.base_urls.messages" label="Anthropic Messages" placeholder="留空用主 Base URL" :maxlength="255" />
           <p class="text-xs text-muted">网关按协议选对应 base_url 直通，无需为每种格式建多个渠道</p>
         </div>
         <Input
@@ -308,6 +309,25 @@ onMounted(load)
           <Input v-model="form.timeout_ms" label="超时 (ms)" type="number" />
           <Input v-model="form.max_concurrency" label="最大并发" type="number" />
         </div>
+        <div class="flex items-center justify-between rounded-md border border-edge p-3">
+          <div>
+            <p class="text-sm font-medium text-ink">启用渠道</p>
+            <p class="text-xs text-muted">禁用后该渠道不会被用于请求转发</p>
+          </div>
+          <button
+            type="button"
+            role="switch"
+            :aria-checked="form.enabled"
+            class="relative inline-flex h-6 w-11 shrink-0 cursor-pointer items-center rounded-full transition-colors focus-visible:outline-2 focus-visible:outline-offset-2 focus-visible:outline-accent"
+            :class="form.enabled ? 'bg-accent' : 'bg-surface2'"
+            @click="form.enabled = !form.enabled"
+          >
+            <span
+              class="pointer-events-none inline-block h-4 w-4 rounded-full bg-white shadow-sm ring-0 transition-transform"
+              :class="form.enabled ? 'translate-x-6' : 'translate-x-1'"
+            />
+          </button>
+        </div>
       </div>
       <template #footer>
         <Button variant="ghost" @click="editOpen = false">取消</Button>

web/src/views/admin/ModelsView.vue

@@ -46,7 +46,6 @@ function quickAdd() {
 
 const form = reactive({
   name: '',
-  display_name: '',
   input_price: 0,
   output_price: 0,
   cache_read_price: 0,
@@ -65,14 +64,14 @@ async function load() {
 
 function openCreate() {
   editing.value = null
-  Object.assign(form, { name: '', display_name: '', input_price: 0, output_price: 0, cache_read_price: 0, enabled: true })
+  Object.assign(form, { name: '', input_price: 0, output_price: 0, cache_read_price: 0, enabled: true })
   editOpen.value = true
 }
 
 function openEdit(m: Model) {
   editing.value = m
   Object.assign(form, {
-    name: m.name, display_name: m.display_name,
+    name: m.name,
     input_price: m.input_price, output_price: m.output_price, cache_read_price: m.cache_read_price,
     enabled: m.enabled,
   })
@@ -82,7 +81,6 @@ function openEdit(m: Model) {
 async function save() {
   saving.value = true
   const payload = {
-    display_name: form.display_name || form.name,
     input_price: Number(form.input_price),
     output_price: Number(form.output_price),
     cache_read_price: Number(form.cache_read_price),
@@ -207,7 +205,6 @@ onMounted(load)
     <Modal :open="editOpen" :title="editing ? '编辑模型' : '添加模型'" @close="editOpen = false">
       <div class="space-y-4">
         <Input v-model="form.name" label="模型名" placeholder="claude-sonnet-5" :disabled="!!editing" />
-        <Input v-model="form.display_name" label="展示名" />
         <div class="grid grid-cols-1 sm:grid-cols-2 gap-4">
           <Input v-model="form.input_price" label="输入价格 /1M" type="number" />
           <Input v-model="form.output_price" label="输出价格 /1M" type="number" />

web/src/views/admin/UsageView.vue

@@ -6,20 +6,68 @@ import { protocolShort } from '@/lib/protocol'
 import { fmtCost, fmtTime } from '@/lib/format'
 import Badge from '@/components/ui/Badge.vue'
 import Button from '@/components/ui/Button.vue'
-import type { UsageLog } from '@/types'
+import Modal from '@/components/ui/Modal.vue'
+import type { UsageLog, User } from '@/types'
 
 const toast = useToastStore()
 const logs = ref<UsageLog[]>([])
 const total = ref(0)
 const page = ref(1)
 const modelFilter = ref('')
+const userFilter = ref('')
+const userCandidates = ref<User[]>([])
+let userSearchTimer: ReturnType<typeof setTimeout> | null = null
 const pageSize = 15
 
+const viewing = ref<UsageLog | null>(null)
+const viewTab = ref<'request' | 'response'>('request')
+
+function openRaw(l: UsageLog) {
+  viewing.value = l
+  viewTab.value = 'request'
+}
+
+// 用户自动补全候选（防抖 300ms）
+function onUserInput() {
+  if (userSearchTimer) clearTimeout(userSearchTimer)
+  const kw = userFilter.value.trim()
+  if (!kw) {
+    userCandidates.value = []
+    return
+  }
+  userSearchTimer = setTimeout(async () => {
+    try {
+      const { data } = await http.get(`/admin/users?q=${encodeURIComponent(kw)}&page_size=10`)
+      userCandidates.value = data.data.items ?? []
+    } catch {
+      userCandidates.value = []
+    }
+  }, 300)
+}
+
+// 用户名 → user_id（精确匹配），查不到返回 null
+async function resolveUserId(username: string): Promise<number | null> {
+  const { data } = await http.get(`/admin/users?q=${encodeURIComponent(username)}&page_size=10`)
+  const items: User[] = data.data.items ?? []
+  const hit = items.find((u) => u.username === username)
+  return hit ? hit.id : null
+}
+
 async function load() {
   try {
-    const { data } = await http.get(
-      `/admin/usage?page=${page.value}&page_size=${pageSize}${modelFilter.value ? '&model=' + modelFilter.value : ''}`,
-    )
+    const params = new URLSearchParams()
+    params.set('page', String(page.value))
+    params.set('page_size', String(pageSize))
+    if (modelFilter.value.trim()) params.set('model', modelFilter.value.trim())
+    if (userFilter.value.trim()) {
+      const uid = await resolveUserId(userFilter.value.trim())
+      if (uid === null) {
+        toast.err(`未找到用户：${userFilter.value.trim()}`)
+        return
+      }
+      params.set('user_id', String(uid))
+    }
+    const { data } = await http.get(`/admin/usage?${params.toString()}`)
     logs.value = data.data.items
     total.value = data.data.total
   } catch (e) {
@@ -32,6 +80,12 @@ function search() {
   load()
 }
 
+function clearUser() {
+  userFilter.value = ''
+  userCandidates.value = []
+  search()
+}
+
 function goPage(p: number) {
   page.value = p
   load()
@@ -47,7 +101,28 @@ onMounted(load)
         <h1 class="text-lg font-semibold">用量统计</h1>
         <p class="text-sm text-muted">全局请求明细与成本</p>
       </div>
-      <div class="flex gap-2">
+      <div class="flex flex-wrap gap-2">
+        <div class="relative">
+          <input
+            v-model="userFilter"
+            list="user-options"
+            placeholder="按用户过滤"
+            class="h-10 w-44 rounded-md border border-edge2 bg-surface px-3 font-mono text-xs outline-none focus:border-accent"
+            @input="onUserInput"
+            @keyup.enter="search"
+          />
+          <button
+            v-if="userFilter"
+            class="absolute right-2 top-1/2 -translate-y-1/2 text-xs text-muted hover:text-ink"
+            title="清除用户筛选"
+            @click="clearUser"
+          >
+            ✕
+          </button>
+        </div>
+        <datalist id="user-options">
+          <option v-for="u in userCandidates" :key="u.id" :value="u.username" />
+        </datalist>
         <input
           v-model="modelFilter"
           placeholder="按模型过滤"
@@ -65,7 +140,7 @@ onMounted(load)
           <div class="flex items-start justify-between gap-2">
             <div class="min-w-0">
               <p class="truncate font-mono text-sm text-ink">{{ l.model }}</p>
-              <p class="mt-0.5 truncate text-xs text-muted">{{ l.user }}</p>
+              <p class="mt-0.5 truncate text-xs text-muted">{{ l.channel }} · {{ l.user }}</p>
             </div>
             <div class="flex shrink-0 flex-col items-end gap-1">
               <Badge :variant="l.status === 'success' ? 'ok' : l.status === 'canceled' ? 'neutral' : 'err'">{{ l.status }}</Badge>
@@ -77,6 +152,7 @@ onMounted(load)
             <span class="mono-num">{{ l.input_tokens }}/{{ l.output_tokens }} tok</span>
             <span class="mono-num">{{ l.latency_ms }}ms</span>
             <span class="mono-num w-full">{{ fmtTime(l.created_at) }}</span>
+            <Button v-if="l.raw_request" size="sm" variant="ghost" class="ml-auto" @click="openRaw(l)">查看原始</Button>
           </div>
         </div>
         <p v-if="logs.length === 0" class="card px-4 py-8 text-center text-sm text-muted">暂无请求记录</p>
@@ -89,18 +165,21 @@ onMounted(load)
             <tr class="border-b border-edge text-left text-xs text-muted">
               <th scope="col" class="px-4 py-2.5 font-medium">用户</th>
               <th scope="col" class="px-4 py-2.5 font-medium">模型</th>
+              <th scope="col" class="px-4 py-2.5 font-medium">渠道</th>
               <th scope="col" class="px-4 py-2.5 font-medium">API 格式</th>
               <th scope="col" class="px-4 py-2.5 font-medium">Token 入/出</th>
               <th scope="col" class="px-4 py-2.5 font-medium">成本</th>
               <th scope="col" class="px-4 py-2.5 font-medium">耗时</th>
               <th scope="col" class="px-4 py-2.5 font-medium">状态</th>
               <th scope="col" class="px-4 py-2.5 font-medium">时间</th>
+              <th scope="col" class="px-4 py-2.5 font-medium">原始</th>
             </tr>
           </thead>
           <tbody>
             <tr v-for="l in logs" :key="l.id" class="table-row">
               <td class="px-4 py-2.5 text-xs text-ink">{{ l.user }}</td>
               <td class="px-4 py-2.5 font-mono text-xs text-ink">{{ l.model }}</td>
+              <td class="px-4 py-2.5 font-mono text-xs text-muted">{{ l.channel || '-' }}</td>
               <td class="px-4 py-2.5 font-mono text-xs text-muted">{{ protocolShort(l.protocol) }}</td>
               <td class="px-4 py-2.5 mono-num text-xs text-muted">{{ l.input_tokens }}/{{ l.output_tokens }}</td>
               <td class="px-4 py-2.5 mono-num text-xs text-ink">{{ fmtCost(l.cost) }}</td>
@@ -109,9 +188,13 @@ onMounted(load)
                 <Badge :variant="l.status === 'success' ? 'ok' : l.status === 'canceled' ? 'neutral' : 'err'">{{ l.status }}</Badge>
               </td>
               <td class="px-4 py-2.5 font-mono text-xs text-muted">{{ fmtTime(l.created_at) }}</td>
+              <td class="px-4 py-2.5">
+                <Button v-if="l.raw_request" size="sm" variant="ghost" @click="openRaw(l)">查看</Button>
+                <span v-else class="text-xs text-muted">-</span>
+              </td>
             </tr>
             <tr v-if="logs.length === 0">
-              <td colspan="8" class="px-4 py-10 text-center text-sm text-muted">暂无请求记录</td>
+              <td colspan="10" class="px-4 py-10 text-center text-sm text-muted">暂无请求记录</td>
             </tr>
           </tbody>
         </table>
@@ -124,5 +207,44 @@ onMounted(load)
         </div>
       </div>
     </div>
+
+    <Modal :open="!!viewing" title="原始请求与响应" width="max-w-3xl" @close="viewing = null">
+      <div v-if="viewing" class="space-y-3">
+        <div class="flex flex-wrap items-center justify-between gap-2">
+          <div class="text-xs text-muted">
+            <span class="font-mono text-ink">{{ viewing.model }}</span>
+            <span class="mx-1.5">·</span>
+            <span class="font-mono">{{ viewing.channel || '-' }}</span>
+            <span class="mx-1.5">·</span>
+            <span class="font-mono">{{ viewing.user || '-' }}</span>
+            <span class="mx-1.5">·</span>
+            <span class="mono-num">{{ fmtTime(viewing.created_at) }}</span>
+          </div>
+          <div class="flex gap-1 rounded-md border border-edge p-0.5">
+            <button
+              class="rounded px-2.5 py-1 text-xs transition"
+              :class="viewTab === 'request' ? 'bg-surface2 text-ink' : 'text-muted'"
+              @click="viewTab = 'request'"
+            >
+              请求
+            </button>
+            <button
+              class="rounded px-2.5 py-1 text-xs transition"
+              :class="viewTab === 'response' ? 'bg-surface2 text-ink' : 'text-muted'"
+              @click="viewTab = 'response'"
+            >
+              响应
+            </button>
+          </div>
+        </div>
+        <pre
+          v-if="viewTab === 'request' ? viewing.raw_request : viewing.raw_response"
+          class="max-h-[60vh] overflow-auto rounded-md border border-edge bg-surface p-3 font-mono text-xs leading-relaxed text-ink whitespace-pre-wrap break-all"
+        >{{ viewTab === 'request' ? viewing.raw_request : viewing.raw_response }}</pre>
+        <p v-else class="rounded-md border border-edge bg-surface p-4 text-center text-xs text-muted">
+          该请求未记录{{ viewTab === 'request' ? '原始请求' : '原始响应' }}
+        </p>
+      </div>
+    </Modal>
   </div>
 </template>

web/src/views/console/DashboardView.vue

@@ -3,7 +3,9 @@ import { onMounted, ref } from 'vue'
 import { http, errMsg } from '@/api/client'
 import { useToastStore } from '@/stores/toast'
 import { fmtMoney, fmtNum, fmtCost, fmtTime } from '@/lib/format'
+import { copyText } from '@/lib/clipboard'
 import Badge from '@/components/ui/Badge.vue'
+import Button from '@/components/ui/Button.vue'
 import Skeleton from '@/components/ui/Skeleton.vue'
 import TrendChart from '@/components/ui/TrendChart.vue'
 import type { UsageLog } from '@/types'
@@ -44,6 +46,24 @@ function daysAgo(n: number): string {
   return d.toISOString().slice(0, 10)
 }
 
+// 快速开始：Base URL 用浏览器当前 host（兼容端点），curl 示例
+const baseURL = `${window.location.origin}/v1`
+const curlCmd = `curl ${baseURL}/chat/completions \\
+  -H "Authorization: Bearer sk-ot-xxxx" \\
+  -H "Content-Type: application/json" \\
+  -d '{"model":"gpt-test","stream":true,"messages":[{"role":"user","content":"hi"}]}'`
+const copied = ref('')
+
+async function copy(text: string, key: string) {
+  const ok = await copyText(text)
+  if (ok) {
+    copied.value = key
+    setTimeout(() => (copied.value = ''), 2000)
+  } else {
+    toast.err('复制失败，请手动复制')
+  }
+}
+
 onMounted(load)
 </script>
 
@@ -102,16 +122,18 @@ onMounted(load)
           </li>
         </ul>
         <ul v-else class="divide-y divide-edge">
-          <li v-for="l in logs" :key="l.id" class="flex items-center justify-between py-2">
-            <div class="min-w-0">
-              <p class="truncate font-mono text-xs text-ink">{{ l.model }}</p>
-              <p class="text-[11px] text-muted">{{ fmtTime(l.created_at) }}</p>
-            </div>
-            <div class="flex items-center gap-2">
-              <span class="mono-num text-xs text-muted">{{ l.input_tokens }}/{{ l.output_tokens }}</span>
-              <span class="mono-num w-16 text-right text-xs text-muted">{{ fmtCost(l.cost) }}</span>
-              <Badge :variant="l.status === 'success' ? 'ok' : l.status === 'canceled' ? 'neutral' : 'err'">{{ l.status }}</Badge>
+          <li v-for="l in logs" :key="l.id" class="py-2">
+            <div class="flex items-center justify-between gap-2">
+              <p class="min-w-0 truncate font-mono text-xs text-ink">{{ l.model }}</p>
+              <Badge class="shrink-0" :variant="l.status === 'success' ? 'ok' : l.status === 'canceled' ? 'neutral' : 'err'">{{ l.status }}</Badge>
             </div>
+            <p class="mt-0.5 flex flex-wrap items-center gap-x-2 text-[11px] text-muted">
+              <span class="mono-num">{{ l.input_tokens }}/{{ l.output_tokens }}</span>
+              <span class="opacity-40">·</span>
+              <span class="mono-num">{{ fmtCost(l.cost) }}</span>
+              <span class="opacity-40">·</span>
+              <span>{{ fmtTime(l.created_at) }}</span>
+            </p>
           </li>
           <li v-if="logs.length === 0" class="py-6 text-center text-xs text-muted">
             还没有请求记录，去
@@ -121,5 +143,44 @@ onMounted(load)
         </ul>
       </div>
     </div>
+
+    <!-- 快速开始 -->
+    <div class="card min-w-0 p-5">
+      <div class="mb-4">
+        <h2 class="text-sm font-semibold">快速开始</h2>
+        <p class="text-xs text-muted">通过兼容端点接入，OpenAI / Anthropic 协议自动转换</p>
+      </div>
+
+      <div class="space-y-4">
+        <div class="min-w-0">
+          <p class="mb-1.5 text-[11px] font-medium text-muted">Base URL</p>
+          <div class="flex items-center gap-2">
+            <code class="min-w-0 flex-1 overflow-x-auto rounded-md border border-edge bg-surface px-3 py-2 font-mono text-xs whitespace-nowrap text-ink">{{ baseURL }}</code>
+            <Button size="sm" variant="ghost" class="shrink-0" @click="copy(baseURL, 'base')">
+              {{ copied === 'base' ? '已复制' : '复制' }}
+            </Button>
+          </div>
+        </div>
+
+        <div class="min-w-0">
+          <p class="mb-1.5 text-[11px] font-medium text-muted">兼容</p>
+          <div class="flex flex-wrap gap-1.5">
+            <Badge variant="neutral">messages</Badge>
+            <Badge variant="neutral">chat/completions</Badge>
+            <Badge variant="neutral">responses</Badge>
+          </div>
+        </div>
+
+        <div class="min-w-0">
+          <p class="mb-1.5 text-[11px] font-medium text-muted">curl 示例</p>
+          <div class="relative min-w-0 overflow-hidden rounded-md border border-edge bg-surface">
+            <pre class="overflow-x-auto px-3 py-2.5 pr-20 font-mono text-[11px] leading-relaxed text-ink">{{ curlCmd }}</pre>
+            <Button size="sm" variant="ghost" class="absolute top-2 right-2" @click="copy(curlCmd, 'curl')">
+              {{ copied === 'curl' ? '已复制' : '复制' }}
+            </Button>
+          </div>
+        </div>
+      </div>
+    </div>
   </div>
 </template>