fix(frontend): 修复运维设置对话框保存按钮始终禁用的问题

后端默认 alert.enabled=true 但 recipients 为空，前端验证将其视为
错误并阻断保存按钮。移除该阻断性验证，改为保存时自动禁用无收件人
的邮件通知配置。

.github/workflows/backend-ci.yml

@@ -19,7 +19,7 @@ jobs:
           cache: true
       - name: Verify Go version
         run: |
-          go version | grep -q 'go1.25.7'
+          go version | grep -q 'go1.26.1'
       - name: Unit tests
         working-directory: backend
         run: make test-unit
@@ -38,10 +38,10 @@ jobs:
           cache: true
       - name: Verify Go version
         run: |
-          go version | grep -q 'go1.25.7'
+          go version | grep -q 'go1.26.1'
       - name: golangci-lint
         uses: golangci/golangci-lint-action@v9
         with:
-          version: v2.7
+          version: v2.9
           args: --timeout=30m
           working-directory: backend

.github/workflows/release.yml

@@ -115,7 +115,7 @@ jobs:
 
       - name: Verify Go version
         run: |
-          go version | grep -q 'go1.25.7'
+          go version | grep -q 'go1.26.1'
 
       # Docker setup for GoReleaser
       - name: Set up QEMU

.github/workflows/security-scan.yml

@@ -23,7 +23,7 @@ jobs:
           cache-dependency-path: backend/go.sum
       - name: Verify Go version
         run: |
-          go version | grep -q 'go1.25.7'
+          go version | grep -q 'go1.26.1'
       - name: Run govulncheck
         working-directory: backend
         run: |

Dockerfile

@@ -7,8 +7,9 @@
 # =============================================================================
 
 ARG NODE_IMAGE=node:24-alpine
-ARG GOLANG_IMAGE=golang:1.25.7-alpine
+ARG GOLANG_IMAGE=golang:1.26.1-alpine
 ARG ALPINE_IMAGE=alpine:3.21
+ARG POSTGRES_IMAGE=postgres:18-alpine
 ARG GOPROXY=https://goproxy.cn,direct
 ARG GOSUMDB=sum.golang.google.cn
 
@@ -73,7 +74,12 @@ RUN VERSION_VALUE="${VERSION}" && \
     ./cmd/server
 
 # -----------------------------------------------------------------------------
-# Stage 3: Final Runtime Image
+# Stage 3: PostgreSQL Client (version-matched with docker-compose)
+# -----------------------------------------------------------------------------
+FROM ${POSTGRES_IMAGE} AS pg-client
+
+# -----------------------------------------------------------------------------
+# Stage 4: Final Runtime Image
 # -----------------------------------------------------------------------------
 FROM ${ALPINE_IMAGE}
 
@@ -86,8 +92,20 @@ LABEL org.opencontainers.image.source="https://github.com/Wei-Shaw/sub2api"
 RUN apk add --no-cache \
     ca-certificates \
     tzdata \
+    libpq \
+    zstd-libs \
+    lz4-libs \
+    krb5-libs \
+    libldap \
+    libedit \
     && rm -rf /var/cache/apk/*
 
+# Copy pg_dump and psql from the same postgres image used in docker-compose
+# This ensures version consistency between backup tools and the database server
+COPY --from=pg-client /usr/local/bin/pg_dump /usr/local/bin/pg_dump
+COPY --from=pg-client /usr/local/bin/psql /usr/local/bin/psql
+COPY --from=pg-client /usr/local/lib/libpq.so.5* /usr/local/lib/
+
 # Create non-root user
 RUN addgroup -g 1000 sub2api && \
     adduser -u 1000 -G sub2api -s /bin/sh -D sub2api

README.md

@@ -39,6 +39,16 @@ Sub2API is an AI API gateway platform designed to distribute and manage API quot
 - **Concurrency Control** - Per-user and per-account concurrency limits
 - **Rate Limiting** - Configurable request and token rate limits
 - **Admin Dashboard** - Web interface for monitoring and management
+- **External System Integration** - Embed external systems (e.g. payment, ticketing) via iframe to extend the admin dashboard
+
+## Ecosystem
+
+Community projects that extend or integrate with Sub2API:
+
+| Project | Description | Features |
+|---------|-------------|----------|
+| [Sub2ApiPay](https://github.com/touwaeriol/sub2apipay) | Self-service payment system | Self-service top-up and subscription purchase; supports YiPay protocol, WeChat Pay, Alipay, Stripe; embeddable via iframe |
+| [sub2api-mobile](https://github.com/ckken/sub2api-mobile) | Mobile admin console | Cross-platform app (iOS/Android/Web) for user management, account management, monitoring dashboard, and multi-backend switching; built with Expo + React Native |
 
 ## Tech Stack
 
@@ -150,14 +160,14 @@ mkdir -p sub2api-deploy && cd sub2api-deploy
 curl -sSL https://raw.githubusercontent.com/Wei-Shaw/sub2api/main/deploy/docker-deploy.sh | bash
 
 # Start services
-docker-compose -f docker-compose.local.yml up -d
+docker-compose up -d
 
 # View logs
-docker-compose -f docker-compose.local.yml logs -f sub2api
+docker-compose logs -f sub2api
 ```
 
 **What the script does:**
-- Downloads `docker-compose.local.yml` and `.env.example`
+- Downloads `docker-compose.local.yml` (saved as `docker-compose.yml`) and `.env.example`
 - Generates secure credentials (JWT_SECRET, TOTP_ENCRYPTION_KEY, POSTGRES_PASSWORD)
 - Creates `.env` file with auto-generated secrets
 - Creates data directories (uses local directories for easy backup/migration)
@@ -522,6 +532,28 @@ sub2api/
     └── install.sh            # One-click installation script
 ```
 
+## Disclaimer
+
+> **Please read carefully before using this project:**
+>
+> :rotating_light: **Terms of Service Risk**: Using this project may violate Anthropic's Terms of Service. Please read Anthropic's user agreement carefully before use. All risks arising from the use of this project are borne solely by the user.
+>
+> :book: **Disclaimer**: This project is for technical learning and research purposes only. The author assumes no responsibility for account suspension, service interruption, or any other losses caused by the use of this project.
+
+---
+
+## Star History
+
+<a href="https://star-history.com/#Wei-Shaw/sub2api&Date">
+ <picture>
+   <source media="(prefers-color-scheme: dark)" srcset="https://api.star-history.com/svg?repos=Wei-Shaw/sub2api&type=Date&theme=dark" />
+   <source media="(prefers-color-scheme: light)" srcset="https://api.star-history.com/svg?repos=Wei-Shaw/sub2api&type=Date" />
+   <img alt="Star History Chart" src="https://api.star-history.com/svg?repos=Wei-Shaw/sub2api&type=Date" />
+ </picture>
+</a>
+
+---
+
 ## License
 
 MIT License

README_CN.md

@@ -39,6 +39,16 @@ Sub2API 是一个 AI API 网关平台，用于分发和管理 AI 产品订阅（
 - **并发控制** - 用户级和账号级并发限制
 - **速率限制** - 可配置的请求和 Token 速率限制
 - **管理后台** - Web 界面进行监控和管理
+- **外部系统集成** - 支持通过 iframe 嵌入外部系统（如支付、工单等），扩展管理后台功能
+
+## 生态项目
+
+围绕 Sub2API 的社区扩展与集成项目：
+
+| 项目 | 说明 | 功能 |
+|------|------|------|
+| [Sub2ApiPay](https://github.com/touwaeriol/sub2apipay) | 自助支付系统 | 用户自助充值、自助订阅购买；兼容易支付协议、微信官方支付、支付宝官方支付、Stripe；支持 iframe 嵌入管理后台 |
+| [sub2api-mobile](https://github.com/ckken/sub2api-mobile) | 移动端管理控制台 | 跨平台应用（iOS/Android/Web），支持用户管理、账号管理、监控看板、多后端切换；基于 Expo + React Native 构建 |
 
 ## 技术栈
 
@@ -154,14 +164,14 @@ mkdir -p sub2api-deploy && cd sub2api-deploy
 curl -sSL https://raw.githubusercontent.com/Wei-Shaw/sub2api/main/deploy/docker-deploy.sh | bash
 
 # 启动服务
-docker-compose -f docker-compose.local.yml up -d
+docker-compose up -d
 
 # 查看日志
-docker-compose -f docker-compose.local.yml logs -f sub2api
+docker-compose logs -f sub2api
 ```
 
 **脚本功能：**
-- 下载 `docker-compose.local.yml` 和 `.env.example`
+- 下载 `docker-compose.local.yml`（本地保存为 `docker-compose.yml`）和 `.env.example`
 - 自动生成安全凭证（JWT_SECRET、TOTP_ENCRYPTION_KEY、POSTGRES_PASSWORD）
 - 创建 `.env` 文件并填充自动生成的密钥
 - 创建数据目录（使用本地目录，便于备份和迁移）
@@ -588,6 +598,28 @@ sub2api/
     └── install.sh            # 一键安装脚本
 ```
 
+## 免责声明
+
+> **使用本项目前请仔细阅读：**
+>
+> :rotating_light: **服务条款风险**: 使用本项目可能违反 Anthropic 的服务条款。请在使用前仔细阅读 Anthropic 的用户协议，使用本项目的一切风险由用户自行承担。
+>
+> :book: **免责声明**: 本项目仅供技术学习和研究使用，作者不对因使用本项目导致的账户封禁、服务中断或其他损失承担任何责任。
+
+---
+
+## Star History
+
+<a href="https://star-history.com/#Wei-Shaw/sub2api&Date">
+ <picture>
+   <source media="(prefers-color-scheme: dark)" srcset="https://api.star-history.com/svg?repos=Wei-Shaw/sub2api&type=Date&theme=dark" />
+   <source media="(prefers-color-scheme: light)" srcset="https://api.star-history.com/svg?repos=Wei-Shaw/sub2api&type=Date" />
+   <img alt="Star History Chart" src="https://api.star-history.com/svg?repos=Wei-Shaw/sub2api&type=Date" />
+ </picture>
+</a>
+
+---
+
 ## 许可证
 
 MIT License

backend/.golangci.yml

@@ -93,20 +93,13 @@ linters:
       check-escaping-errors: true
     staticcheck:
       # https://staticcheck.dev/docs/configuration/options/#dot_import_whitelist
-      # Default: ["github.com/mmcloughlin/avo/build", "github.com/mmcloughlin/avo/operand", "github.com/mmcloughlin/avo/reg"]
       dot-import-whitelist:
         - fmt
       # https://staticcheck.dev/docs/configuration/options/#initialisms
-      # Default: ["ACL", "API", "ASCII", "CPU", "CSS", "DNS", "EOF", "GUID", "HTML", "HTTP", "HTTPS", "ID", "IP", "JSON", "QPS", "RAM", "RPC", "SLA", "SMTP", "SQL", "SSH", "TCP", "TLS", "TTL", "UDP", "UI", "GID", "UID", "UUID", "URI", "URL", "UTF8", "VM", "XML", "XMPP", "XSRF", "XSS", "SIP", "RTP", "AMQP", "DB", "TS"]
       initialisms: [ "ACL", "API", "ASCII", "CPU", "CSS", "DNS", "EOF", "GUID", "HTML", "HTTP", "HTTPS", "ID", "IP", "JSON", "QPS", "RAM", "RPC", "SLA", "SMTP", "SQL", "SSH", "TCP", "TLS", "TTL", "UDP", "UI", "GID", "UID", "UUID", "URI", "URL", "UTF8", "VM", "XML", "XMPP", "XSRF", "XSS", "SIP", "RTP", "AMQP", "DB", "TS" ]
       # https://staticcheck.dev/docs/configuration/options/#http_status_code_whitelist
-      # Default: ["200", "400", "404", "500"]
       http-status-code-whitelist: [ "200", "400", "404", "500" ]
-      # SAxxxx checks in https://staticcheck.dev/docs/configuration/options/#checks
-      # Example (to disable some checks): [ "all", "-SA1000", "-SA1001"]
-      # Run `GL_DEBUG=staticcheck golangci-lint run --enable=staticcheck` to see all available checks and enabled by config checks.
-      # Default: ["all", "-ST1000", "-ST1003", "-ST1016", "-ST1020", "-ST1021", "-ST1022"]
-      # Temporarily disable style checks to allow CI to pass
+      # "all" enables every SA/ST/S/QF check; only list the ones to disable.
       checks:
         - all
         - -ST1000  # Package comment format
@@ -114,489 +107,19 @@ linters:
         - -ST1020  # Comment on exported method format
         - -ST1021  # Comment on exported type format
         - -ST1022  # Comment on exported variable format
-        # Invalid regular expression.
-        # https://staticcheck.dev/docs/checks/#SA1000
-        - SA1000
-        # Invalid template.
-        # https://staticcheck.dev/docs/checks/#SA1001
-        - SA1001
-        # Invalid format in 'time.Parse'.
-        # https://staticcheck.dev/docs/checks/#SA1002
-        - SA1002
-        # Unsupported argument to functions in 'encoding/binary'.
-        # https://staticcheck.dev/docs/checks/#SA1003
-        - SA1003
-        # Suspiciously small untyped constant in 'time.Sleep'.
-        # https://staticcheck.dev/docs/checks/#SA1004
-        - SA1004
-        # Invalid first argument to 'exec.Command'.
-        # https://staticcheck.dev/docs/checks/#SA1005
-        - SA1005
-        # 'Printf' with dynamic first argument and no further arguments.
-        # https://staticcheck.dev/docs/checks/#SA1006
-        - SA1006
-        # Invalid URL in 'net/url.Parse'.
-        # https://staticcheck.dev/docs/checks/#SA1007
-        - SA1007
-        # Non-canonical key in 'http.Header' map.
-        # https://staticcheck.dev/docs/checks/#SA1008
-        - SA1008
-        # '(*regexp.Regexp).FindAll' called with 'n == 0', which will always return zero results.
-        # https://staticcheck.dev/docs/checks/#SA1010
-        - SA1010
-        # Various methods in the "strings" package expect valid UTF-8, but invalid input is provided.
-        # https://staticcheck.dev/docs/checks/#SA1011
-        - SA1011
-        # A nil 'context.Context' is being passed to a function, consider using 'context.TODO' instead.
-        # https://staticcheck.dev/docs/checks/#SA1012
-        - SA1012
-        # 'io.Seeker.Seek' is being called with the whence constant as the first argument, but it should be the second.
-        # https://staticcheck.dev/docs/checks/#SA1013
-        - SA1013
-        # Non-pointer value passed to 'Unmarshal' or 'Decode'.
-        # https://staticcheck.dev/docs/checks/#SA1014
-        - SA1014
-        # Using 'time.Tick' in a way that will leak. Consider using 'time.NewTicker', and only use 'time.Tick' in tests, commands and endless functions.
-        # https://staticcheck.dev/docs/checks/#SA1015
-        - SA1015
-        # Trapping a signal that cannot be trapped.
-        # https://staticcheck.dev/docs/checks/#SA1016
-        - SA1016
-        # Channels used with 'os/signal.Notify' should be buffered.
-        # https://staticcheck.dev/docs/checks/#SA1017
-        - SA1017
-        # 'strings.Replace' called with 'n == 0', which does nothing.
-        # https://staticcheck.dev/docs/checks/#SA1018
-        - SA1018
-        # Using a deprecated function, variable, constant or field.
-        # https://staticcheck.dev/docs/checks/#SA1019
-        - SA1019
-        # Using an invalid host:port pair with a 'net.Listen'-related function.
-        # https://staticcheck.dev/docs/checks/#SA1020
-        - SA1020
-        # Using 'bytes.Equal' to compare two 'net.IP'.
-        # https://staticcheck.dev/docs/checks/#SA1021
-        - SA1021
-        # Modifying the buffer in an 'io.Writer' implementation.
-        # https://staticcheck.dev/docs/checks/#SA1023
-        - SA1023
-        # A string cutset contains duplicate characters.
-        # https://staticcheck.dev/docs/checks/#SA1024
-        - SA1024
-        # It is not possible to use '(*time.Timer).Reset''s return value correctly.
-        # https://staticcheck.dev/docs/checks/#SA1025
-        - SA1025
-        # Cannot marshal channels or functions.
-        # https://staticcheck.dev/docs/checks/#SA1026
-        - SA1026
-        # Atomic access to 64-bit variable must be 64-bit aligned.
-        # https://staticcheck.dev/docs/checks/#SA1027
-        - SA1027
-        # 'sort.Slice' can only be used on slices.
-        # https://staticcheck.dev/docs/checks/#SA1028
-        - SA1028
-        # Inappropriate key in call to 'context.WithValue'.
-        # https://staticcheck.dev/docs/checks/#SA1029
-        - SA1029
-        # Invalid argument in call to a 'strconv' function.
-        # https://staticcheck.dev/docs/checks/#SA1030
-        - SA1030
-        # Overlapping byte slices passed to an encoder.
-        # https://staticcheck.dev/docs/checks/#SA1031
-        - SA1031
-        # Wrong order of arguments to 'errors.Is'.
-        # https://staticcheck.dev/docs/checks/#SA1032
-        - SA1032
-        # 'sync.WaitGroup.Add' called inside the goroutine, leading to a race condition.
-        # https://staticcheck.dev/docs/checks/#SA2000
-        - SA2000
-        # Empty critical section, did you mean to defer the unlock?.
-        # https://staticcheck.dev/docs/checks/#SA2001
-        - SA2001
-        # Called 'testing.T.FailNow' or 'SkipNow' in a goroutine, which isn't allowed.
-        # https://staticcheck.dev/docs/checks/#SA2002
-        - SA2002
-        # Deferred 'Lock' right after locking, likely meant to defer 'Unlock' instead.
-        # https://staticcheck.dev/docs/checks/#SA2003
-        - SA2003
-        # 'TestMain' doesn't call 'os.Exit', hiding test failures.
-        # https://staticcheck.dev/docs/checks/#SA3000
-        - SA3000
-        # Assigning to 'b.N' in benchmarks distorts the results.
-        # https://staticcheck.dev/docs/checks/#SA3001
-        - SA3001
-        # Binary operator has identical expressions on both sides.
-        # https://staticcheck.dev/docs/checks/#SA4000
-        - SA4000
-        # '&*x' gets simplified to 'x', it does not copy 'x'.
-        # https://staticcheck.dev/docs/checks/#SA4001
-        - SA4001
-        # Comparing unsigned values against negative values is pointless.
-        # https://staticcheck.dev/docs/checks/#SA4003
-        - SA4003
-        # The loop exits unconditionally after one iteration.
-        # https://staticcheck.dev/docs/checks/#SA4004
-        - SA4004
-        # Field assignment that will never be observed. Did you mean to use a pointer receiver?.
-        # https://staticcheck.dev/docs/checks/#SA4005
-        - SA4005
-        # A value assigned to a variable is never read before being overwritten. Forgotten error check or dead code?.
-        # https://staticcheck.dev/docs/checks/#SA4006
-        - SA4006
-        # The variable in the loop condition never changes, are you incrementing the wrong variable?.
-        # https://staticcheck.dev/docs/checks/#SA4008
-        - SA4008
-        # A function argument is overwritten before its first use.
-        # https://staticcheck.dev/docs/checks/#SA4009
-        - SA4009
-        # The result of 'append' will never be observed anywhere.
-        # https://staticcheck.dev/docs/checks/#SA4010
-        - SA4010
-        # Break statement with no effect. Did you mean to break out of an outer loop?.
-        # https://staticcheck.dev/docs/checks/#SA4011
-        - SA4011
-        # Comparing a value against NaN even though no value is equal to NaN.
-        # https://staticcheck.dev/docs/checks/#SA4012
-        - SA4012
-        # Negating a boolean twice ('!!b') is the same as writing 'b'. This is either redundant, or a typo.
-        # https://staticcheck.dev/docs/checks/#SA4013
-        - SA4013
-        # An if/else if chain has repeated conditions and no side-effects; if the condition didn't match the first time, it won't match the second time, either.
-        # https://staticcheck.dev/docs/checks/#SA4014
-        - SA4014
-        # Calling functions like 'math.Ceil' on floats converted from integers doesn't do anything useful.
-        # https://staticcheck.dev/docs/checks/#SA4015
-        - SA4015
-        # Certain bitwise operations, such as 'x ^ 0', do not do anything useful.
-        # https://staticcheck.dev/docs/checks/#SA4016
-        - SA4016
-        # Discarding the return values of a function without side effects, making the call pointless.
-        # https://staticcheck.dev/docs/checks/#SA4017
-        - SA4017
-        # Self-assignment of variables.
-        # https://staticcheck.dev/docs/checks/#SA4018
-        - SA4018
-        # Multiple, identical build constraints in the same file.
-        # https://staticcheck.dev/docs/checks/#SA4019
-        - SA4019
-        # Unreachable case clause in a type switch.
-        # https://staticcheck.dev/docs/checks/#SA4020
-        - SA4020
-        # "x = append(y)" is equivalent to "x = y".
-        # https://staticcheck.dev/docs/checks/#SA4021
-        - SA4021
-        # Comparing the address of a variable against nil.
-        # https://staticcheck.dev/docs/checks/#SA4022
-        - SA4022
-        # Impossible comparison of interface value with untyped nil.
-        # https://staticcheck.dev/docs/checks/#SA4023
-        - SA4023
-        # Checking for impossible return value from a builtin function.
-        # https://staticcheck.dev/docs/checks/#SA4024
-        - SA4024
-        # Integer division of literals that results in zero.
-        # https://staticcheck.dev/docs/checks/#SA4025
-        - SA4025
-        # Go constants cannot express negative zero.
-        # https://staticcheck.dev/docs/checks/#SA4026
-        - SA4026
-        # '(*net/url.URL).Query' returns a copy, modifying it doesn't change the URL.
-        # https://staticcheck.dev/docs/checks/#SA4027
-        - SA4027
-        # 'x % 1' is always zero.
-        # https://staticcheck.dev/docs/checks/#SA4028
-        - SA4028
-        # Ineffective attempt at sorting slice.
-        # https://staticcheck.dev/docs/checks/#SA4029
-        - SA4029
-        # Ineffective attempt at generating random number.
-        # https://staticcheck.dev/docs/checks/#SA4030
-        - SA4030
-        # Checking never-nil value against nil.
-        # https://staticcheck.dev/docs/checks/#SA4031
-        - SA4031
-        # Comparing 'runtime.GOOS' or 'runtime.GOARCH' against impossible value.
-        # https://staticcheck.dev/docs/checks/#SA4032
-        - SA4032
-        # Assignment to nil map.
-        # https://staticcheck.dev/docs/checks/#SA5000
-        - SA5000
-        # Deferring 'Close' before checking for a possible error.
-        # https://staticcheck.dev/docs/checks/#SA5001
-        - SA5001
-        # The empty for loop ("for {}") spins and can block the scheduler.
-        # https://staticcheck.dev/docs/checks/#SA5002
-        - SA5002
-        # Defers in infinite loops will never execute.
-        # https://staticcheck.dev/docs/checks/#SA5003
-        - SA5003
-        # "for { select { ..." with an empty default branch spins.
-        # https://staticcheck.dev/docs/checks/#SA5004
-        - SA5004
-        # The finalizer references the finalized object, preventing garbage collection.
-        # https://staticcheck.dev/docs/checks/#SA5005
-        - SA5005
-        # Infinite recursive call.
-        # https://staticcheck.dev/docs/checks/#SA5007
-        - SA5007
-        # Invalid struct tag.
-        # https://staticcheck.dev/docs/checks/#SA5008
-        - SA5008
-        # Invalid Printf call.
-        # https://staticcheck.dev/docs/checks/#SA5009
-        - SA5009
-        # Impossible type assertion.
-        # https://staticcheck.dev/docs/checks/#SA5010
-        - SA5010
-        # Possible nil pointer dereference.
-        # https://staticcheck.dev/docs/checks/#SA5011
-        - SA5011
-        # Passing odd-sized slice to function expecting even size.
-        # https://staticcheck.dev/docs/checks/#SA5012
-        - SA5012
-        # Using 'regexp.Match' or related in a loop, should use 'regexp.Compile'.
-        # https://staticcheck.dev/docs/checks/#SA6000
-        - SA6000
-        # Missing an optimization opportunity when indexing maps by byte slices.
-        # https://staticcheck.dev/docs/checks/#SA6001
-        - SA6001
-        # Storing non-pointer values in 'sync.Pool' allocates memory.
-        # https://staticcheck.dev/docs/checks/#SA6002
-        - SA6002
-        # Converting a string to a slice of runes before ranging over it.
-        # https://staticcheck.dev/docs/checks/#SA6003
-        - SA6003
-        # Inefficient string comparison with 'strings.ToLower' or 'strings.ToUpper'.
-        # https://staticcheck.dev/docs/checks/#SA6005
-        - SA6005
-        # Using io.WriteString to write '[]byte'.
-        # https://staticcheck.dev/docs/checks/#SA6006
-        - SA6006
-        # Defers in range loops may not run when you expect them to.
-        # https://staticcheck.dev/docs/checks/#SA9001
-        - SA9001
-        # Using a non-octal 'os.FileMode' that looks like it was meant to be in octal.
-        # https://staticcheck.dev/docs/checks/#SA9002
-        - SA9002
-        # Empty body in an if or else branch.
-        # https://staticcheck.dev/docs/checks/#SA9003
-        - SA9003
-        # Only the first constant has an explicit type.
-        # https://staticcheck.dev/docs/checks/#SA9004
-        - SA9004
-        # Trying to marshal a struct with no public fields nor custom marshaling.
-        # https://staticcheck.dev/docs/checks/#SA9005
-        - SA9005
-        # Dubious bit shifting of a fixed size integer value.
-        # https://staticcheck.dev/docs/checks/#SA9006
-        - SA9006
-        # Deleting a directory that shouldn't be deleted.
-        # https://staticcheck.dev/docs/checks/#SA9007
-        - SA9007
-        # 'else' branch of a type assertion is probably not reading the right value.
-        # https://staticcheck.dev/docs/checks/#SA9008
-        - SA9008
-        # Ineffectual Go compiler directive.
-        # https://staticcheck.dev/docs/checks/#SA9009
-        - SA9009
-        # NOTE: ST1000, ST1001, ST1003, ST1020, ST1021, ST1022 are disabled above
-        # Incorrectly formatted error string.
-        # https://staticcheck.dev/docs/checks/#ST1005
-        - ST1005
-        # Poorly chosen receiver name.
-        # https://staticcheck.dev/docs/checks/#ST1006
-        - ST1006
-        # A function's error value should be its last return value.
-        # https://staticcheck.dev/docs/checks/#ST1008
-        - ST1008
-        # Poorly chosen name for variable of type 'time.Duration'.
-        # https://staticcheck.dev/docs/checks/#ST1011
-        - ST1011
-        # Poorly chosen name for error variable.
-        # https://staticcheck.dev/docs/checks/#ST1012
-        - ST1012
-        # Should use constants for HTTP error codes, not magic numbers.
-        # https://staticcheck.dev/docs/checks/#ST1013
-        - ST1013
-        # A switch's default case should be the first or last case.
-        # https://staticcheck.dev/docs/checks/#ST1015
-        - ST1015
-        # Use consistent method receiver names.
-        # https://staticcheck.dev/docs/checks/#ST1016
-        - ST1016
-        # Don't use Yoda conditions.
-        # https://staticcheck.dev/docs/checks/#ST1017
-        - ST1017
-        # Avoid zero-width and control characters in string literals.
-        # https://staticcheck.dev/docs/checks/#ST1018
-        - ST1018
-        # Importing the same package multiple times.
-        # https://staticcheck.dev/docs/checks/#ST1019
-        - ST1019
-        # NOTE: ST1020, ST1021, ST1022 removed (disabled above)
-        # Redundant type in variable declaration.
-        # https://staticcheck.dev/docs/checks/#ST1023
-        - ST1023
-        # Use plain channel send or receive instead of single-case select.
-        # https://staticcheck.dev/docs/checks/#S1000
-        - S1000
-        # Replace for loop with call to copy.
-        # https://staticcheck.dev/docs/checks/#S1001
-        - S1001
-        # Omit comparison with boolean constant.
-        # https://staticcheck.dev/docs/checks/#S1002
-        - S1002
-        # Replace call to 'strings.Index' with 'strings.Contains'.
-        # https://staticcheck.dev/docs/checks/#S1003
-        - S1003
-        # Replace call to 'bytes.Compare' with 'bytes.Equal'.
-        # https://staticcheck.dev/docs/checks/#S1004
-        - S1004
-        # Drop unnecessary use of the blank identifier.
-        # https://staticcheck.dev/docs/checks/#S1005
-        - S1005
-        # Use "for { ... }" for infinite loops.
-        # https://staticcheck.dev/docs/checks/#S1006
-        - S1006
-        # Simplify regular expression by using raw string literal.
-        # https://staticcheck.dev/docs/checks/#S1007
-        - S1007
-        # Simplify returning boolean expression.
-        # https://staticcheck.dev/docs/checks/#S1008
-        - S1008
-        # Omit redundant nil check on slices, maps, and channels.
-        # https://staticcheck.dev/docs/checks/#S1009
-        - S1009
-        # Omit default slice index.
-        # https://staticcheck.dev/docs/checks/#S1010
-        - S1010
-        # Use a single 'append' to concatenate two slices.
-        # https://staticcheck.dev/docs/checks/#S1011
-        - S1011
-        # Replace 'time.Now().Sub(x)' with 'time.Since(x)'.
-        # https://staticcheck.dev/docs/checks/#S1012
-        - S1012
-        # Use a type conversion instead of manually copying struct fields.
-        # https://staticcheck.dev/docs/checks/#S1016
-        - S1016
-        # Replace manual trimming with 'strings.TrimPrefix'.
-        # https://staticcheck.dev/docs/checks/#S1017
-        - S1017
-        # Use "copy" for sliding elements.
-        # https://staticcheck.dev/docs/checks/#S1018
-        - S1018
-        # Simplify "make" call by omitting redundant arguments.
-        # https://staticcheck.dev/docs/checks/#S1019
-        - S1019
-        # Omit redundant nil check in type assertion.
-        # https://staticcheck.dev/docs/checks/#S1020
-        - S1020
-        # Merge variable declaration and assignment.
-        # https://staticcheck.dev/docs/checks/#S1021
-        - S1021
-        # Omit redundant control flow.
-        # https://staticcheck.dev/docs/checks/#S1023
-        - S1023
-        # Replace 'x.Sub(time.Now())' with 'time.Until(x)'.
-        # https://staticcheck.dev/docs/checks/#S1024
-        - S1024
-        # Don't use 'fmt.Sprintf("%s", x)' unnecessarily.
-        # https://staticcheck.dev/docs/checks/#S1025
-        - S1025
-        # Simplify error construction with 'fmt.Errorf'.
-        # https://staticcheck.dev/docs/checks/#S1028
-        - S1028
-        # Range over the string directly.
-        # https://staticcheck.dev/docs/checks/#S1029
-        - S1029
-        # Use 'bytes.Buffer.String' or 'bytes.Buffer.Bytes'.
-        # https://staticcheck.dev/docs/checks/#S1030
-        - S1030
-        # Omit redundant nil check around loop.
-        # https://staticcheck.dev/docs/checks/#S1031
-        - S1031
-        # Use 'sort.Ints(x)', 'sort.Float64s(x)', and 'sort.Strings(x)'.
-        # https://staticcheck.dev/docs/checks/#S1032
-        - S1032
-        # Unnecessary guard around call to "delete".
-        # https://staticcheck.dev/docs/checks/#S1033
-        - S1033
-        # Use result of type assertion to simplify cases.
-        # https://staticcheck.dev/docs/checks/#S1034
-        - S1034
-        # Redundant call to 'net/http.CanonicalHeaderKey' in method call on 'net/http.Header'.
-        # https://staticcheck.dev/docs/checks/#S1035
-        - S1035
-        # Unnecessary guard around map access.
-        # https://staticcheck.dev/docs/checks/#S1036
-        - S1036
-        # Elaborate way of sleeping.
-        # https://staticcheck.dev/docs/checks/#S1037
-        - S1037
-        # Unnecessarily complex way of printing formatted string.
-        # https://staticcheck.dev/docs/checks/#S1038
-        - S1038
-        # Unnecessary use of 'fmt.Sprint'.
-        # https://staticcheck.dev/docs/checks/#S1039
-        - S1039
-        # Type assertion to current type.
-        # https://staticcheck.dev/docs/checks/#S1040
-        - S1040
-        # Apply De Morgan's law.
-        # https://staticcheck.dev/docs/checks/#QF1001
-        - QF1001
-        # Convert untagged switch to tagged switch.
-        # https://staticcheck.dev/docs/checks/#QF1002
-        - QF1002
-        # Convert if/else-if chain to tagged switch.
-        # https://staticcheck.dev/docs/checks/#QF1003
-        - QF1003
-        # Use 'strings.ReplaceAll' instead of 'strings.Replace' with 'n == -1'.
-        # https://staticcheck.dev/docs/checks/#QF1004
-        - QF1004
-        # Expand call to 'math.Pow'.
-        # https://staticcheck.dev/docs/checks/#QF1005
-        - QF1005
-        # Lift 'if'+'break' into loop condition.
-        # https://staticcheck.dev/docs/checks/#QF1006
-        - QF1006
-        # Merge conditional assignment into variable declaration.
-        # https://staticcheck.dev/docs/checks/#QF1007
-        - QF1007
-        # Omit embedded fields from selector expression.
-        # https://staticcheck.dev/docs/checks/#QF1008
-        - QF1008
-        # Use 'time.Time.Equal' instead of '==' operator.
-        # https://staticcheck.dev/docs/checks/#QF1009
-        - QF1009
-        # Convert slice of bytes to string when printing it.
-        # https://staticcheck.dev/docs/checks/#QF1010
-        - QF1010
-        # Omit redundant type from variable declaration.
-        # https://staticcheck.dev/docs/checks/#QF1011
-        - QF1011
-        # Use 'fmt.Fprintf(x, ...)' instead of 'x.Write(fmt.Sprintf(...))'.
-        # https://staticcheck.dev/docs/checks/#QF1012
-        - QF1012
     unused:
-      # Mark all struct fields that have been written to as used.
       # Default: true
-      field-writes-are-uses: false
-      # Treat IncDec statement (e.g. `i++` or `i--`) as both read and write operation instead of just write.
+      field-writes-are-uses: true
       # Default: false
       post-statements-are-reads: true
-      # Mark all exported fields as used.
-      # default: true
-      exported-fields-are-used: false
-      # Mark all function parameters as used.
-      # default: true
-      parameters-are-used: true
-      # Mark all local variables as used.
-      # default: true
-      local-variables-are-used: false
-      # Mark all identifiers inside generated files as used.
       # Default: true
-      generated-is-used: false
+      exported-fields-are-used: true
+      # Default: true
+      parameters-are-used: true
+      # Default: true
+      local-variables-are-used: false
+      # Default: true — must be true, ent generates 130K+ lines of code
+      generated-is-used: true
 
 formatters:
   enable:

backend/cmd/jwtgen/main.go

@@ -33,7 +33,7 @@ func main() {
 	}()
 
 	userRepo := repository.NewUserRepository(client, sqlDB)
-	authService := service.NewAuthService(userRepo, nil, nil, cfg, nil, nil, nil, nil, nil, nil)
+	authService := service.NewAuthService(client, userRepo, nil, nil, cfg, nil, nil, nil, nil, nil, nil)
 
 	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
 	defer cancel()

backend/cmd/server/wire.go

@@ -41,6 +41,9 @@ func initializeApplication(buildInfo handler.BuildInfo) (*Application, error) {
 		// Server layer ProviderSet
 		server.ProviderSet,
 
+		// Privacy client factory for OpenAI training opt-out
+		providePrivacyClientFactory,
+
 		// BuildInfo provider
 		provideServiceBuildInfo,
 
@@ -53,6 +56,10 @@ func initializeApplication(buildInfo handler.BuildInfo) (*Application, error) {
 	return nil, nil
 }
 
+func providePrivacyClientFactory() service.PrivacyClientFactory {
+	return repository.CreatePrivacyReqClient
+}
+
 func provideServiceBuildInfo(buildInfo handler.BuildInfo) service.BuildInfo {
 	return service.BuildInfo{
 		Version:   buildInfo.Version,
@@ -87,6 +94,7 @@ func provideCleanup(
 	antigravityOAuth *service.AntigravityOAuthService,
 	openAIGateway *service.OpenAIGatewayService,
 	scheduledTestRunner *service.ScheduledTestRunnerService,
+	backupSvc *service.BackupService,
 ) func() {
 	return func() {
 		ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
@@ -223,6 +231,12 @@ func provideCleanup(
 				}
 				return nil
 			}},
+			{"BackupService", func() error {
+				if backupSvc != nil {
+					backupSvc.Stop()
+				}
+				return nil
+			}},
 		}
 
 		infraSteps := []cleanupStep{

backend/cmd/server/wire_gen.go

@@ -67,7 +67,7 @@ func initializeApplication(buildInfo handler.BuildInfo) (*Application, error) {
 	apiKeyAuthCacheInvalidator := service.ProvideAPIKeyAuthCacheInvalidator(apiKeyService)
 	promoService := service.NewPromoService(promoCodeRepository, userRepository, billingCacheService, client, apiKeyAuthCacheInvalidator)
 	subscriptionService := service.NewSubscriptionService(groupRepository, userSubscriptionRepository, billingCacheService, client, configConfig)
-	authService := service.NewAuthService(userRepository, redeemCodeRepository, refreshTokenCache, configConfig, settingService, emailService, turnstileService, emailQueueService, promoService, subscriptionService)
+	authService := service.NewAuthService(client, userRepository, redeemCodeRepository, refreshTokenCache, configConfig, settingService, emailService, turnstileService, emailQueueService, promoService, subscriptionService)
 	userService := service.NewUserService(userRepository, apiKeyAuthCacheInvalidator, billingCache)
 	redeemCache := repository.NewRedeemCache(redisClient)
 	redeemService := service.NewRedeemService(redeemCodeRepository, userRepository, subscriptionService, redeemCache, billingCacheService, client, apiKeyAuthCacheInvalidator)
@@ -81,6 +81,7 @@ func initializeApplication(buildInfo handler.BuildInfo) (*Application, error) {
 	userHandler := handler.NewUserHandler(userService)
 	apiKeyHandler := handler.NewAPIKeyHandler(apiKeyService)
 	usageLogRepository := repository.NewUsageLogRepository(client, db)
+	usageBillingRepository := repository.NewUsageBillingRepository(client, db)
 	usageService := service.NewUsageService(usageLogRepository, userRepository, client, apiKeyAuthCacheInvalidator)
 	usageHandler := handler.NewUsageHandler(usageService, apiKeyService)
 	redeemHandler := handler.NewRedeemHandler(redeemService)
@@ -104,7 +105,8 @@ func initializeApplication(buildInfo handler.BuildInfo) (*Application, error) {
 	proxyRepository := repository.NewProxyRepository(client, db)
 	proxyExitInfoProber := repository.NewProxyExitInfoProber(configConfig)
 	proxyLatencyCache := repository.NewProxyLatencyCache(redisClient)
-	adminService := service.NewAdminService(userRepository, groupRepository, accountRepository, soraAccountRepository, proxyRepository, apiKeyRepository, redeemCodeRepository, userGroupRateRepository, billingCacheService, proxyExitInfoProber, proxyLatencyCache, apiKeyAuthCacheInvalidator, client, settingService, subscriptionService)
+	privacyClientFactory := providePrivacyClientFactory()
+	adminService := service.NewAdminService(userRepository, groupRepository, accountRepository, soraAccountRepository, proxyRepository, apiKeyRepository, redeemCodeRepository, userGroupRateRepository, billingCacheService, proxyExitInfoProber, proxyLatencyCache, apiKeyAuthCacheInvalidator, client, settingService, subscriptionService, userSubscriptionRepository, privacyClientFactory)
 	concurrencyCache := repository.ProvideConcurrencyCache(redisClient, configConfig)
 	concurrencyService := service.ProvideConcurrencyService(concurrencyCache, accountRepository, configConfig)
 	adminUserHandler := admin.NewUserHandler(adminService, concurrencyService)
@@ -144,6 +146,10 @@ func initializeApplication(buildInfo handler.BuildInfo) (*Application, error) {
 	adminAnnouncementHandler := admin.NewAnnouncementHandler(announcementService)
 	dataManagementService := service.NewDataManagementService()
 	dataManagementHandler := admin.NewDataManagementHandler(dataManagementService)
+	backupObjectStoreFactory := repository.NewS3BackupStoreFactory()
+	dbDumper := repository.NewPgDumper(configConfig)
+	backupService := service.ProvideBackupService(settingRepository, configConfig, secretEncryptor, backupObjectStoreFactory, dbDumper)
+	backupHandler := admin.NewBackupHandler(backupService, userService)
 	oAuthHandler := admin.NewOAuthHandler(oAuthService)
 	openAIOAuthHandler := admin.NewOpenAIOAuthHandler(openAIOAuthService, adminService)
 	geminiOAuthHandler := admin.NewGeminiOAuthHandler(geminiOAuthService)
@@ -162,9 +168,9 @@ func initializeApplication(buildInfo handler.BuildInfo) (*Application, error) {
 	deferredService := service.ProvideDeferredService(accountRepository, timingWheelService)
 	claudeTokenProvider := service.NewClaudeTokenProvider(accountRepository, geminiTokenCache, oAuthService)
 	digestSessionStore := service.NewDigestSessionStore()
-	gatewayService := service.NewGatewayService(accountRepository, groupRepository, usageLogRepository, userRepository, userSubscriptionRepository, userGroupRateRepository, gatewayCache, configConfig, schedulerSnapshotService, concurrencyService, billingService, rateLimitService, billingCacheService, identityService, httpUpstream, deferredService, claudeTokenProvider, sessionLimitCache, rpmCache, digestSessionStore)
+	gatewayService := service.NewGatewayService(accountRepository, groupRepository, usageLogRepository, usageBillingRepository, userRepository, userSubscriptionRepository, userGroupRateRepository, gatewayCache, configConfig, schedulerSnapshotService, concurrencyService, billingService, rateLimitService, billingCacheService, identityService, httpUpstream, deferredService, claudeTokenProvider, sessionLimitCache, rpmCache, digestSessionStore, settingService)
 	openAITokenProvider := service.NewOpenAITokenProvider(accountRepository, geminiTokenCache, openAIOAuthService)
-	openAIGatewayService := service.NewOpenAIGatewayService(accountRepository, usageLogRepository, userRepository, userSubscriptionRepository, gatewayCache, configConfig, schedulerSnapshotService, concurrencyService, billingService, rateLimitService, billingCacheService, httpUpstream, deferredService, openAITokenProvider)
+	openAIGatewayService := service.NewOpenAIGatewayService(accountRepository, usageLogRepository, usageBillingRepository, userRepository, userSubscriptionRepository, userGroupRateRepository, gatewayCache, configConfig, schedulerSnapshotService, concurrencyService, billingService, rateLimitService, billingCacheService, httpUpstream, deferredService, openAITokenProvider)
 	geminiMessagesCompatService := service.NewGeminiMessagesCompatService(accountRepository, groupRepository, gatewayCache, schedulerSnapshotService, geminiTokenProvider, rateLimitService, httpUpstream, antigravityGatewayService, configConfig)
 	opsSystemLogSink := service.ProvideOpsSystemLogSink(opsRepository)
 	opsService := service.NewOpsService(opsRepository, settingRepository, configConfig, accountRepository, userRepository, concurrencyService, gatewayService, openAIGatewayService, geminiMessagesCompatService, antigravityGatewayService, opsSystemLogSink)
@@ -199,7 +205,7 @@ func initializeApplication(buildInfo handler.BuildInfo) (*Application, error) {
 	scheduledTestResultRepository := repository.NewScheduledTestResultRepository(db)
 	scheduledTestService := service.ProvideScheduledTestService(scheduledTestPlanRepository, scheduledTestResultRepository)
 	scheduledTestHandler := admin.NewScheduledTestHandler(scheduledTestService)
-	adminHandlers := handler.ProvideAdminHandlers(dashboardHandler, adminUserHandler, groupHandler, accountHandler, adminAnnouncementHandler, dataManagementHandler, oAuthHandler, openAIOAuthHandler, geminiOAuthHandler, antigravityOAuthHandler, proxyHandler, adminRedeemHandler, promoHandler, settingHandler, opsHandler, systemHandler, adminSubscriptionHandler, adminUsageHandler, userAttributeHandler, errorPassthroughHandler, adminAPIKeyHandler, scheduledTestHandler)
+	adminHandlers := handler.ProvideAdminHandlers(dashboardHandler, adminUserHandler, groupHandler, accountHandler, adminAnnouncementHandler, dataManagementHandler, backupHandler, oAuthHandler, openAIOAuthHandler, geminiOAuthHandler, antigravityOAuthHandler, proxyHandler, adminRedeemHandler, promoHandler, settingHandler, opsHandler, systemHandler, adminSubscriptionHandler, adminUsageHandler, userAttributeHandler, errorPassthroughHandler, adminAPIKeyHandler, scheduledTestHandler)
 	usageRecordWorkerPool := service.NewUsageRecordWorkerPool(configConfig)
 	userMsgQueueCache := repository.NewUserMsgQueueCache(redisClient)
 	userMessageQueueService := service.ProvideUserMessageQueueService(userMsgQueueCache, rpmCache, configConfig)
@@ -226,11 +232,11 @@ func initializeApplication(buildInfo handler.BuildInfo) (*Application, error) {
 	opsCleanupService := service.ProvideOpsCleanupService(opsRepository, db, redisClient, configConfig)
 	opsScheduledReportService := service.ProvideOpsScheduledReportService(opsService, userService, emailService, redisClient, configConfig)
 	soraMediaCleanupService := service.ProvideSoraMediaCleanupService(soraMediaStorage, configConfig)
-	tokenRefreshService := service.ProvideTokenRefreshService(accountRepository, soraAccountRepository, oAuthService, openAIOAuthService, geminiOAuthService, antigravityOAuthService, compositeTokenCacheInvalidator, schedulerCache, configConfig, tempUnschedCache)
+	tokenRefreshService := service.ProvideTokenRefreshService(accountRepository, soraAccountRepository, oAuthService, openAIOAuthService, geminiOAuthService, antigravityOAuthService, compositeTokenCacheInvalidator, schedulerCache, configConfig, tempUnschedCache, privacyClientFactory, proxyRepository)
 	accountExpiryService := service.ProvideAccountExpiryService(accountRepository)
 	subscriptionExpiryService := service.ProvideSubscriptionExpiryService(userSubscriptionRepository)
-	scheduledTestRunnerService := service.ProvideScheduledTestRunnerService(scheduledTestPlanRepository, scheduledTestService, accountTestService, configConfig)
-	v := provideCleanup(client, redisClient, opsMetricsCollector, opsAggregationService, opsAlertEvaluatorService, opsCleanupService, opsScheduledReportService, opsSystemLogSink, soraMediaCleanupService, schedulerSnapshotService, tokenRefreshService, accountExpiryService, subscriptionExpiryService, usageCleanupService, idempotencyCleanupService, pricingService, emailQueueService, billingCacheService, usageRecordWorkerPool, subscriptionService, oAuthService, openAIOAuthService, geminiOAuthService, antigravityOAuthService, openAIGatewayService, scheduledTestRunnerService)
+	scheduledTestRunnerService := service.ProvideScheduledTestRunnerService(scheduledTestPlanRepository, scheduledTestService, accountTestService, rateLimitService, configConfig)
+	v := provideCleanup(client, redisClient, opsMetricsCollector, opsAggregationService, opsAlertEvaluatorService, opsCleanupService, opsScheduledReportService, opsSystemLogSink, soraMediaCleanupService, schedulerSnapshotService, tokenRefreshService, accountExpiryService, subscriptionExpiryService, usageCleanupService, idempotencyCleanupService, pricingService, emailQueueService, billingCacheService, usageRecordWorkerPool, subscriptionService, oAuthService, openAIOAuthService, geminiOAuthService, antigravityOAuthService, openAIGatewayService, scheduledTestRunnerService, backupService)
 	application := &Application{
 		Server:  httpServer,
 		Cleanup: v,
@@ -245,6 +251,10 @@ type Application struct {
 	Cleanup func()
 }
 
+func providePrivacyClientFactory() service.PrivacyClientFactory {
+	return repository.CreatePrivacyReqClient
+}
+
 func provideServiceBuildInfo(buildInfo handler.BuildInfo) service.BuildInfo {
 	return service.BuildInfo{
 		Version:   buildInfo.Version,
@@ -279,6 +289,7 @@ func provideCleanup(
 	antigravityOAuth *service.AntigravityOAuthService,
 	openAIGateway *service.OpenAIGatewayService,
 	scheduledTestRunner *service.ScheduledTestRunnerService,
+	backupSvc *service.BackupService,
 ) func() {
 	return func() {
 		ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
@@ -414,6 +425,12 @@ func provideCleanup(
 				}
 				return nil
 			}},
+			{"BackupService", func() error {
+				if backupSvc != nil {
+					backupSvc.Stop()
+				}
+				return nil
+			}},
 		}
 
 		infraSteps := []cleanupStep{

backend/cmd/server/wire_gen_test.go

@@ -75,6 +75,7 @@ func TestProvideCleanup_WithMinimalDependencies_NoPanic(t *testing.T) {
 		antigravityOAuthSvc,
 		nil, // openAIGateway
 		nil, // scheduledTestRunner
+		nil, // backupSvc
 	)
 
 	require.NotPanics(t, func() {

backend/ent/account.go

@@ -41,6 +41,8 @@ type Account struct {
 	ProxyID *int64 `json:"proxy_id,omitempty"`
 	// Concurrency holds the value of the "concurrency" field.
 	Concurrency int `json:"concurrency,omitempty"`
+	// LoadFactor holds the value of the "load_factor" field.
+	LoadFactor *int `json:"load_factor,omitempty"`
 	// Priority holds the value of the "priority" field.
 	Priority int `json:"priority,omitempty"`
 	// RateMultiplier holds the value of the "rate_multiplier" field.
@@ -143,7 +145,7 @@ func (*Account) scanValues(columns []string) ([]any, error) {
 			values[i] = new(sql.NullBool)
 		case account.FieldRateMultiplier:
 			values[i] = new(sql.NullFloat64)
-		case account.FieldID, account.FieldProxyID, account.FieldConcurrency, account.FieldPriority:
+		case account.FieldID, account.FieldProxyID, account.FieldConcurrency, account.FieldLoadFactor, account.FieldPriority:
 			values[i] = new(sql.NullInt64)
 		case account.FieldName, account.FieldNotes, account.FieldPlatform, account.FieldType, account.FieldStatus, account.FieldErrorMessage, account.FieldTempUnschedulableReason, account.FieldSessionWindowStatus:
 			values[i] = new(sql.NullString)
@@ -243,6 +245,13 @@ func (_m *Account) assignValues(columns []string, values []any) error {
 			} else if value.Valid {
 				_m.Concurrency = int(value.Int64)
 			}
+		case account.FieldLoadFactor:
+			if value, ok := values[i].(*sql.NullInt64); !ok {
+				return fmt.Errorf("unexpected type %T for field load_factor", values[i])
+			} else if value.Valid {
+				_m.LoadFactor = new(int)
+				*_m.LoadFactor = int(value.Int64)
+			}
 		case account.FieldPriority:
 			if value, ok := values[i].(*sql.NullInt64); !ok {
 				return fmt.Errorf("unexpected type %T for field priority", values[i])
@@ -445,6 +454,11 @@ func (_m *Account) String() string {
 	builder.WriteString("concurrency=")
 	builder.WriteString(fmt.Sprintf("%v", _m.Concurrency))
 	builder.WriteString(", ")
+	if v := _m.LoadFactor; v != nil {
+		builder.WriteString("load_factor=")
+		builder.WriteString(fmt.Sprintf("%v", *v))
+	}
+	builder.WriteString(", ")
 	builder.WriteString("priority=")
 	builder.WriteString(fmt.Sprintf("%v", _m.Priority))
 	builder.WriteString(", ")

backend/ent/account/account.go

@@ -37,6 +37,8 @@ const (
 	FieldProxyID = "proxy_id"
 	// FieldConcurrency holds the string denoting the concurrency field in the database.
 	FieldConcurrency = "concurrency"
+	// FieldLoadFactor holds the string denoting the load_factor field in the database.
+	FieldLoadFactor = "load_factor"
 	// FieldPriority holds the string denoting the priority field in the database.
 	FieldPriority = "priority"
 	// FieldRateMultiplier holds the string denoting the rate_multiplier field in the database.
@@ -121,6 +123,7 @@ var Columns = []string{
 	FieldExtra,
 	FieldProxyID,
 	FieldConcurrency,
+	FieldLoadFactor,
 	FieldPriority,
 	FieldRateMultiplier,
 	FieldStatus,
@@ -250,6 +253,11 @@ func ByConcurrency(opts ...sql.OrderTermOption) OrderOption {
 	return sql.OrderByField(FieldConcurrency, opts...).ToFunc()
 }
 
+// ByLoadFactor orders the results by the load_factor field.
+func ByLoadFactor(opts ...sql.OrderTermOption) OrderOption {
+	return sql.OrderByField(FieldLoadFactor, opts...).ToFunc()
+}
+
 // ByPriority orders the results by the priority field.
 func ByPriority(opts ...sql.OrderTermOption) OrderOption {
 	return sql.OrderByField(FieldPriority, opts...).ToFunc()

backend/ent/account/where.go

@@ -100,6 +100,11 @@ func Concurrency(v int) predicate.Account {
 	return predicate.Account(sql.FieldEQ(FieldConcurrency, v))
 }
 
+// LoadFactor applies equality check predicate on the "load_factor" field. It's identical to LoadFactorEQ.
+func LoadFactor(v int) predicate.Account {
+	return predicate.Account(sql.FieldEQ(FieldLoadFactor, v))
+}
+
 // Priority applies equality check predicate on the "priority" field. It's identical to PriorityEQ.
 func Priority(v int) predicate.Account {
 	return predicate.Account(sql.FieldEQ(FieldPriority, v))
@@ -650,6 +655,56 @@ func ConcurrencyLTE(v int) predicate.Account {
 	return predicate.Account(sql.FieldLTE(FieldConcurrency, v))
 }
 
+// LoadFactorEQ applies the EQ predicate on the "load_factor" field.
+func LoadFactorEQ(v int) predicate.Account {
+	return predicate.Account(sql.FieldEQ(FieldLoadFactor, v))
+}
+
+// LoadFactorNEQ applies the NEQ predicate on the "load_factor" field.
+func LoadFactorNEQ(v int) predicate.Account {
+	return predicate.Account(sql.FieldNEQ(FieldLoadFactor, v))
+}
+
+// LoadFactorIn applies the In predicate on the "load_factor" field.
+func LoadFactorIn(vs ...int) predicate.Account {
+	return predicate.Account(sql.FieldIn(FieldLoadFactor, vs...))
+}
+
+// LoadFactorNotIn applies the NotIn predicate on the "load_factor" field.
+func LoadFactorNotIn(vs ...int) predicate.Account {
+	return predicate.Account(sql.FieldNotIn(FieldLoadFactor, vs...))
+}
+
+// LoadFactorGT applies the GT predicate on the "load_factor" field.
+func LoadFactorGT(v int) predicate.Account {
+	return predicate.Account(sql.FieldGT(FieldLoadFactor, v))
+}
+
+// LoadFactorGTE applies the GTE predicate on the "load_factor" field.
+func LoadFactorGTE(v int) predicate.Account {
+	return predicate.Account(sql.FieldGTE(FieldLoadFactor, v))
+}
+
+// LoadFactorLT applies the LT predicate on the "load_factor" field.
+func LoadFactorLT(v int) predicate.Account {
+	return predicate.Account(sql.FieldLT(FieldLoadFactor, v))
+}
+
+// LoadFactorLTE applies the LTE predicate on the "load_factor" field.
+func LoadFactorLTE(v int) predicate.Account {
+	return predicate.Account(sql.FieldLTE(FieldLoadFactor, v))
+}
+
+// LoadFactorIsNil applies the IsNil predicate on the "load_factor" field.
+func LoadFactorIsNil() predicate.Account {
+	return predicate.Account(sql.FieldIsNull(FieldLoadFactor))
+}
+
+// LoadFactorNotNil applies the NotNil predicate on the "load_factor" field.
+func LoadFactorNotNil() predicate.Account {
+	return predicate.Account(sql.FieldNotNull(FieldLoadFactor))
+}
+
 // PriorityEQ applies the EQ predicate on the "priority" field.
 func PriorityEQ(v int) predicate.Account {
 	return predicate.Account(sql.FieldEQ(FieldPriority, v))

backend/ent/account_create.go

@@ -139,6 +139,20 @@ func (_c *AccountCreate) SetNillableConcurrency(v *int) *AccountCreate {
 	return _c
 }
 
+// SetLoadFactor sets the "load_factor" field.
+func (_c *AccountCreate) SetLoadFactor(v int) *AccountCreate {
+	_c.mutation.SetLoadFactor(v)
+	return _c
+}
+
+// SetNillableLoadFactor sets the "load_factor" field if the given value is not nil.
+func (_c *AccountCreate) SetNillableLoadFactor(v *int) *AccountCreate {
+	if v != nil {
+		_c.SetLoadFactor(*v)
+	}
+	return _c
+}
+
 // SetPriority sets the "priority" field.
 func (_c *AccountCreate) SetPriority(v int) *AccountCreate {
 	_c.mutation.SetPriority(v)
@@ -623,6 +637,10 @@ func (_c *AccountCreate) createSpec() (*Account, *sqlgraph.CreateSpec) {
 		_spec.SetField(account.FieldConcurrency, field.TypeInt, value)
 		_node.Concurrency = value
 	}
+	if value, ok := _c.mutation.LoadFactor(); ok {
+		_spec.SetField(account.FieldLoadFactor, field.TypeInt, value)
+		_node.LoadFactor = &value
+	}
 	if value, ok := _c.mutation.Priority(); ok {
 		_spec.SetField(account.FieldPriority, field.TypeInt, value)
 		_node.Priority = value
@@ -936,6 +954,30 @@ func (u *AccountUpsert) AddConcurrency(v int) *AccountUpsert {
 	return u
 }
 
+// SetLoadFactor sets the "load_factor" field.
+func (u *AccountUpsert) SetLoadFactor(v int) *AccountUpsert {
+	u.Set(account.FieldLoadFactor, v)
+	return u
+}
+
+// UpdateLoadFactor sets the "load_factor" field to the value that was provided on create.
+func (u *AccountUpsert) UpdateLoadFactor() *AccountUpsert {
+	u.SetExcluded(account.FieldLoadFactor)
+	return u
+}
+
+// AddLoadFactor adds v to the "load_factor" field.
+func (u *AccountUpsert) AddLoadFactor(v int) *AccountUpsert {
+	u.Add(account.FieldLoadFactor, v)
+	return u
+}
+
+// ClearLoadFactor clears the value of the "load_factor" field.
+func (u *AccountUpsert) ClearLoadFactor() *AccountUpsert {
+	u.SetNull(account.FieldLoadFactor)
+	return u
+}
+
 // SetPriority sets the "priority" field.
 func (u *AccountUpsert) SetPriority(v int) *AccountUpsert {
 	u.Set(account.FieldPriority, v)
@@ -1419,6 +1461,34 @@ func (u *AccountUpsertOne) UpdateConcurrency() *AccountUpsertOne {
 	})
 }
 
+// SetLoadFactor sets the "load_factor" field.
+func (u *AccountUpsertOne) SetLoadFactor(v int) *AccountUpsertOne {
+	return u.Update(func(s *AccountUpsert) {
+		s.SetLoadFactor(v)
+	})
+}
+
+// AddLoadFactor adds v to the "load_factor" field.
+func (u *AccountUpsertOne) AddLoadFactor(v int) *AccountUpsertOne {
+	return u.Update(func(s *AccountUpsert) {
+		s.AddLoadFactor(v)
+	})
+}
+
+// UpdateLoadFactor sets the "load_factor" field to the value that was provided on create.
+func (u *AccountUpsertOne) UpdateLoadFactor() *AccountUpsertOne {
+	return u.Update(func(s *AccountUpsert) {
+		s.UpdateLoadFactor()
+	})
+}
+
+// ClearLoadFactor clears the value of the "load_factor" field.
+func (u *AccountUpsertOne) ClearLoadFactor() *AccountUpsertOne {
+	return u.Update(func(s *AccountUpsert) {
+		s.ClearLoadFactor()
+	})
+}
+
 // SetPriority sets the "priority" field.
 func (u *AccountUpsertOne) SetPriority(v int) *AccountUpsertOne {
 	return u.Update(func(s *AccountUpsert) {
@@ -2113,6 +2183,34 @@ func (u *AccountUpsertBulk) UpdateConcurrency() *AccountUpsertBulk {
 	})
 }
 
+// SetLoadFactor sets the "load_factor" field.
+func (u *AccountUpsertBulk) SetLoadFactor(v int) *AccountUpsertBulk {
+	return u.Update(func(s *AccountUpsert) {
+		s.SetLoadFactor(v)
+	})
+}
+
+// AddLoadFactor adds v to the "load_factor" field.
+func (u *AccountUpsertBulk) AddLoadFactor(v int) *AccountUpsertBulk {
+	return u.Update(func(s *AccountUpsert) {
+		s.AddLoadFactor(v)
+	})
+}
+
+// UpdateLoadFactor sets the "load_factor" field to the value that was provided on create.
+func (u *AccountUpsertBulk) UpdateLoadFactor() *AccountUpsertBulk {
+	return u.Update(func(s *AccountUpsert) {
+		s.UpdateLoadFactor()
+	})
+}
+
+// ClearLoadFactor clears the value of the "load_factor" field.
+func (u *AccountUpsertBulk) ClearLoadFactor() *AccountUpsertBulk {
+	return u.Update(func(s *AccountUpsert) {
+		s.ClearLoadFactor()
+	})
+}
+
 // SetPriority sets the "priority" field.
 func (u *AccountUpsertBulk) SetPriority(v int) *AccountUpsertBulk {
 	return u.Update(func(s *AccountUpsert) {

backend/ent/account_update.go

@@ -172,6 +172,33 @@ func (_u *AccountUpdate) AddConcurrency(v int) *AccountUpdate {
 	return _u
 }
 
+// SetLoadFactor sets the "load_factor" field.
+func (_u *AccountUpdate) SetLoadFactor(v int) *AccountUpdate {
+	_u.mutation.ResetLoadFactor()
+	_u.mutation.SetLoadFactor(v)
+	return _u
+}
+
+// SetNillableLoadFactor sets the "load_factor" field if the given value is not nil.
+func (_u *AccountUpdate) SetNillableLoadFactor(v *int) *AccountUpdate {
+	if v != nil {
+		_u.SetLoadFactor(*v)
+	}
+	return _u
+}
+
+// AddLoadFactor adds value to the "load_factor" field.
+func (_u *AccountUpdate) AddLoadFactor(v int) *AccountUpdate {
+	_u.mutation.AddLoadFactor(v)
+	return _u
+}
+
+// ClearLoadFactor clears the value of the "load_factor" field.
+func (_u *AccountUpdate) ClearLoadFactor() *AccountUpdate {
+	_u.mutation.ClearLoadFactor()
+	return _u
+}
+
 // SetPriority sets the "priority" field.
 func (_u *AccountUpdate) SetPriority(v int) *AccountUpdate {
 	_u.mutation.ResetPriority()
@@ -684,6 +711,15 @@ func (_u *AccountUpdate) sqlSave(ctx context.Context) (_node int, err error) {
 	if value, ok := _u.mutation.AddedConcurrency(); ok {
 		_spec.AddField(account.FieldConcurrency, field.TypeInt, value)
 	}
+	if value, ok := _u.mutation.LoadFactor(); ok {
+		_spec.SetField(account.FieldLoadFactor, field.TypeInt, value)
+	}
+	if value, ok := _u.mutation.AddedLoadFactor(); ok {
+		_spec.AddField(account.FieldLoadFactor, field.TypeInt, value)
+	}
+	if _u.mutation.LoadFactorCleared() {
+		_spec.ClearField(account.FieldLoadFactor, field.TypeInt)
+	}
 	if value, ok := _u.mutation.Priority(); ok {
 		_spec.SetField(account.FieldPriority, field.TypeInt, value)
 	}
@@ -1063,6 +1099,33 @@ func (_u *AccountUpdateOne) AddConcurrency(v int) *AccountUpdateOne {
 	return _u
 }
 
+// SetLoadFactor sets the "load_factor" field.
+func (_u *AccountUpdateOne) SetLoadFactor(v int) *AccountUpdateOne {
+	_u.mutation.ResetLoadFactor()
+	_u.mutation.SetLoadFactor(v)
+	return _u
+}
+
+// SetNillableLoadFactor sets the "load_factor" field if the given value is not nil.
+func (_u *AccountUpdateOne) SetNillableLoadFactor(v *int) *AccountUpdateOne {
+	if v != nil {
+		_u.SetLoadFactor(*v)
+	}
+	return _u
+}
+
+// AddLoadFactor adds value to the "load_factor" field.
+func (_u *AccountUpdateOne) AddLoadFactor(v int) *AccountUpdateOne {
+	_u.mutation.AddLoadFactor(v)
+	return _u
+}
+
+// ClearLoadFactor clears the value of the "load_factor" field.
+func (_u *AccountUpdateOne) ClearLoadFactor() *AccountUpdateOne {
+	_u.mutation.ClearLoadFactor()
+	return _u
+}
+
 // SetPriority sets the "priority" field.
 func (_u *AccountUpdateOne) SetPriority(v int) *AccountUpdateOne {
 	_u.mutation.ResetPriority()
@@ -1605,6 +1668,15 @@ func (_u *AccountUpdateOne) sqlSave(ctx context.Context) (_node *Account, err er
 	if value, ok := _u.mutation.AddedConcurrency(); ok {
 		_spec.AddField(account.FieldConcurrency, field.TypeInt, value)
 	}
+	if value, ok := _u.mutation.LoadFactor(); ok {
+		_spec.SetField(account.FieldLoadFactor, field.TypeInt, value)
+	}
+	if value, ok := _u.mutation.AddedLoadFactor(); ok {
+		_spec.AddField(account.FieldLoadFactor, field.TypeInt, value)
+	}
+	if _u.mutation.LoadFactorCleared() {
+		_spec.ClearField(account.FieldLoadFactor, field.TypeInt)
+	}
 	if value, ok := _u.mutation.Priority(); ok {
 		_spec.SetField(account.FieldPriority, field.TypeInt, value)
 	}

backend/ent/announcement.go

@@ -25,6 +25,8 @@ type Announcement struct {
 	Content string `json:"content,omitempty"`
 	// 状态: draft, active, archived
 	Status string `json:"status,omitempty"`
+	// 通知模式: silent(仅铃铛), popup(弹窗提醒)
+	NotifyMode string `json:"notify_mode,omitempty"`
 	// 展示条件（JSON 规则）
 	Targeting domain.AnnouncementTargeting `json:"targeting,omitempty"`
 	// 开始展示时间（为空表示立即生效）
@@ -72,7 +74,7 @@ func (*Announcement) scanValues(columns []string) ([]any, error) {
 			values[i] = new([]byte)
 		case announcement.FieldID, announcement.FieldCreatedBy, announcement.FieldUpdatedBy:
 			values[i] = new(sql.NullInt64)
-		case announcement.FieldTitle, announcement.FieldContent, announcement.FieldStatus:
+		case announcement.FieldTitle, announcement.FieldContent, announcement.FieldStatus, announcement.FieldNotifyMode:
 			values[i] = new(sql.NullString)
 		case announcement.FieldStartsAt, announcement.FieldEndsAt, announcement.FieldCreatedAt, announcement.FieldUpdatedAt:
 			values[i] = new(sql.NullTime)
@@ -115,6 +117,12 @@ func (_m *Announcement) assignValues(columns []string, values []any) error {
 			} else if value.Valid {
 				_m.Status = value.String
 			}
+		case announcement.FieldNotifyMode:
+			if value, ok := values[i].(*sql.NullString); !ok {
+				return fmt.Errorf("unexpected type %T for field notify_mode", values[i])
+			} else if value.Valid {
+				_m.NotifyMode = value.String
+			}
 		case announcement.FieldTargeting:
 			if value, ok := values[i].(*[]byte); !ok {
 				return fmt.Errorf("unexpected type %T for field targeting", values[i])
@@ -213,6 +221,9 @@ func (_m *Announcement) String() string {
 	builder.WriteString("status=")
 	builder.WriteString(_m.Status)
 	builder.WriteString(", ")
+	builder.WriteString("notify_mode=")
+	builder.WriteString(_m.NotifyMode)
+	builder.WriteString(", ")
 	builder.WriteString("targeting=")
 	builder.WriteString(fmt.Sprintf("%v", _m.Targeting))
 	builder.WriteString(", ")

backend/ent/announcement/announcement.go

@@ -20,6 +20,8 @@ const (
 	FieldContent = "content"
 	// FieldStatus holds the string denoting the status field in the database.
 	FieldStatus = "status"
+	// FieldNotifyMode holds the string denoting the notify_mode field in the database.
+	FieldNotifyMode = "notify_mode"
 	// FieldTargeting holds the string denoting the targeting field in the database.
 	FieldTargeting = "targeting"
 	// FieldStartsAt holds the string denoting the starts_at field in the database.
@@ -53,6 +55,7 @@ var Columns = []string{
 	FieldTitle,
 	FieldContent,
 	FieldStatus,
+	FieldNotifyMode,
 	FieldTargeting,
 	FieldStartsAt,
 	FieldEndsAt,
@@ -81,6 +84,10 @@ var (
 	DefaultStatus string
 	// StatusValidator is a validator for the "status" field. It is called by the builders before save.
 	StatusValidator func(string) error
+	// DefaultNotifyMode holds the default value on creation for the "notify_mode" field.
+	DefaultNotifyMode string
+	// NotifyModeValidator is a validator for the "notify_mode" field. It is called by the builders before save.
+	NotifyModeValidator func(string) error
 	// DefaultCreatedAt holds the default value on creation for the "created_at" field.
 	DefaultCreatedAt func() time.Time
 	// DefaultUpdatedAt holds the default value on creation for the "updated_at" field.
@@ -112,6 +119,11 @@ func ByStatus(opts ...sql.OrderTermOption) OrderOption {
 	return sql.OrderByField(FieldStatus, opts...).ToFunc()
 }
 
+// ByNotifyMode orders the results by the notify_mode field.
+func ByNotifyMode(opts ...sql.OrderTermOption) OrderOption {
+	return sql.OrderByField(FieldNotifyMode, opts...).ToFunc()
+}
+
 // ByStartsAt orders the results by the starts_at field.
 func ByStartsAt(opts ...sql.OrderTermOption) OrderOption {
 	return sql.OrderByField(FieldStartsAt, opts...).ToFunc()

backend/ent/announcement/where.go

@@ -70,6 +70,11 @@ func Status(v string) predicate.Announcement {
 	return predicate.Announcement(sql.FieldEQ(FieldStatus, v))
 }
 
+// NotifyMode applies equality check predicate on the "notify_mode" field. It's identical to NotifyModeEQ.
+func NotifyMode(v string) predicate.Announcement {
+	return predicate.Announcement(sql.FieldEQ(FieldNotifyMode, v))
+}
+
 // StartsAt applies equality check predicate on the "starts_at" field. It's identical to StartsAtEQ.
 func StartsAt(v time.Time) predicate.Announcement {
 	return predicate.Announcement(sql.FieldEQ(FieldStartsAt, v))
@@ -295,6 +300,71 @@ func StatusContainsFold(v string) predicate.Announcement {
 	return predicate.Announcement(sql.FieldContainsFold(FieldStatus, v))
 }
 
+// NotifyModeEQ applies the EQ predicate on the "notify_mode" field.
+func NotifyModeEQ(v string) predicate.Announcement {
+	return predicate.Announcement(sql.FieldEQ(FieldNotifyMode, v))
+}
+
+// NotifyModeNEQ applies the NEQ predicate on the "notify_mode" field.
+func NotifyModeNEQ(v string) predicate.Announcement {
+	return predicate.Announcement(sql.FieldNEQ(FieldNotifyMode, v))
+}
+
+// NotifyModeIn applies the In predicate on the "notify_mode" field.
+func NotifyModeIn(vs ...string) predicate.Announcement {
+	return predicate.Announcement(sql.FieldIn(FieldNotifyMode, vs...))
+}
+
+// NotifyModeNotIn applies the NotIn predicate on the "notify_mode" field.
+func NotifyModeNotIn(vs ...string) predicate.Announcement {
+	return predicate.Announcement(sql.FieldNotIn(FieldNotifyMode, vs...))
+}
+
+// NotifyModeGT applies the GT predicate on the "notify_mode" field.
+func NotifyModeGT(v string) predicate.Announcement {
+	return predicate.Announcement(sql.FieldGT(FieldNotifyMode, v))
+}
+
+// NotifyModeGTE applies the GTE predicate on the "notify_mode" field.
+func NotifyModeGTE(v string) predicate.Announcement {
+	return predicate.Announcement(sql.FieldGTE(FieldNotifyMode, v))
+}
+
+// NotifyModeLT applies the LT predicate on the "notify_mode" field.
+func NotifyModeLT(v string) predicate.Announcement {
+	return predicate.Announcement(sql.FieldLT(FieldNotifyMode, v))
+}
+
+// NotifyModeLTE applies the LTE predicate on the "notify_mode" field.
+func NotifyModeLTE(v string) predicate.Announcement {
+	return predicate.Announcement(sql.FieldLTE(FieldNotifyMode, v))
+}
+
+// NotifyModeContains applies the Contains predicate on the "notify_mode" field.
+func NotifyModeContains(v string) predicate.Announcement {
+	return predicate.Announcement(sql.FieldContains(FieldNotifyMode, v))
+}
+
+// NotifyModeHasPrefix applies the HasPrefix predicate on the "notify_mode" field.
+func NotifyModeHasPrefix(v string) predicate.Announcement {
+	return predicate.Announcement(sql.FieldHasPrefix(FieldNotifyMode, v))
+}
+
+// NotifyModeHasSuffix applies the HasSuffix predicate on the "notify_mode" field.
+func NotifyModeHasSuffix(v string) predicate.Announcement {
+	return predicate.Announcement(sql.FieldHasSuffix(FieldNotifyMode, v))
+}
+
+// NotifyModeEqualFold applies the EqualFold predicate on the "notify_mode" field.
+func NotifyModeEqualFold(v string) predicate.Announcement {
+	return predicate.Announcement(sql.FieldEqualFold(FieldNotifyMode, v))
+}
+
+// NotifyModeContainsFold applies the ContainsFold predicate on the "notify_mode" field.
+func NotifyModeContainsFold(v string) predicate.Announcement {
+	return predicate.Announcement(sql.FieldContainsFold(FieldNotifyMode, v))
+}
+
 // TargetingIsNil applies the IsNil predicate on the "targeting" field.
 func TargetingIsNil() predicate.Announcement {
 	return predicate.Announcement(sql.FieldIsNull(FieldTargeting))

backend/ent/announcement_create.go

@@ -50,6 +50,20 @@ func (_c *AnnouncementCreate) SetNillableStatus(v *string) *AnnouncementCreate {
 	return _c
 }
 
+// SetNotifyMode sets the "notify_mode" field.
+func (_c *AnnouncementCreate) SetNotifyMode(v string) *AnnouncementCreate {
+	_c.mutation.SetNotifyMode(v)
+	return _c
+}
+
+// SetNillableNotifyMode sets the "notify_mode" field if the given value is not nil.
+func (_c *AnnouncementCreate) SetNillableNotifyMode(v *string) *AnnouncementCreate {
+	if v != nil {
+		_c.SetNotifyMode(*v)
+	}
+	return _c
+}
+
 // SetTargeting sets the "targeting" field.
 func (_c *AnnouncementCreate) SetTargeting(v domain.AnnouncementTargeting) *AnnouncementCreate {
 	_c.mutation.SetTargeting(v)
@@ -202,6 +216,10 @@ func (_c *AnnouncementCreate) defaults() {
 		v := announcement.DefaultStatus
 		_c.mutation.SetStatus(v)
 	}
+	if _, ok := _c.mutation.NotifyMode(); !ok {
+		v := announcement.DefaultNotifyMode
+		_c.mutation.SetNotifyMode(v)
+	}
 	if _, ok := _c.mutation.CreatedAt(); !ok {
 		v := announcement.DefaultCreatedAt()
 		_c.mutation.SetCreatedAt(v)
@@ -238,6 +256,14 @@ func (_c *AnnouncementCreate) check() error {
 			return &ValidationError{Name: "status", err: fmt.Errorf(`ent: validator failed for field "Announcement.status": %w`, err)}
 		}
 	}
+	if _, ok := _c.mutation.NotifyMode(); !ok {
+		return &ValidationError{Name: "notify_mode", err: errors.New(`ent: missing required field "Announcement.notify_mode"`)}
+	}
+	if v, ok := _c.mutation.NotifyMode(); ok {
+		if err := announcement.NotifyModeValidator(v); err != nil {
+			return &ValidationError{Name: "notify_mode", err: fmt.Errorf(`ent: validator failed for field "Announcement.notify_mode": %w`, err)}
+		}
+	}
 	if _, ok := _c.mutation.CreatedAt(); !ok {
 		return &ValidationError{Name: "created_at", err: errors.New(`ent: missing required field "Announcement.created_at"`)}
 	}
@@ -283,6 +309,10 @@ func (_c *AnnouncementCreate) createSpec() (*Announcement, *sqlgraph.CreateSpec)
 		_spec.SetField(announcement.FieldStatus, field.TypeString, value)
 		_node.Status = value
 	}
+	if value, ok := _c.mutation.NotifyMode(); ok {
+		_spec.SetField(announcement.FieldNotifyMode, field.TypeString, value)
+		_node.NotifyMode = value
+	}
 	if value, ok := _c.mutation.Targeting(); ok {
 		_spec.SetField(announcement.FieldTargeting, field.TypeJSON, value)
 		_node.Targeting = value
@@ -415,6 +445,18 @@ func (u *AnnouncementUpsert) UpdateStatus() *AnnouncementUpsert {
 	return u
 }
 
+// SetNotifyMode sets the "notify_mode" field.
+func (u *AnnouncementUpsert) SetNotifyMode(v string) *AnnouncementUpsert {
+	u.Set(announcement.FieldNotifyMode, v)
+	return u
+}
+
+// UpdateNotifyMode sets the "notify_mode" field to the value that was provided on create.
+func (u *AnnouncementUpsert) UpdateNotifyMode() *AnnouncementUpsert {
+	u.SetExcluded(announcement.FieldNotifyMode)
+	return u
+}
+
 // SetTargeting sets the "targeting" field.
 func (u *AnnouncementUpsert) SetTargeting(v domain.AnnouncementTargeting) *AnnouncementUpsert {
 	u.Set(announcement.FieldTargeting, v)
@@ -616,6 +658,20 @@ func (u *AnnouncementUpsertOne) UpdateStatus() *AnnouncementUpsertOne {
 	})
 }
 
+// SetNotifyMode sets the "notify_mode" field.
+func (u *AnnouncementUpsertOne) SetNotifyMode(v string) *AnnouncementUpsertOne {
+	return u.Update(func(s *AnnouncementUpsert) {
+		s.SetNotifyMode(v)
+	})
+}
+
+// UpdateNotifyMode sets the "notify_mode" field to the value that was provided on create.
+func (u *AnnouncementUpsertOne) UpdateNotifyMode() *AnnouncementUpsertOne {
+	return u.Update(func(s *AnnouncementUpsert) {
+		s.UpdateNotifyMode()
+	})
+}
+
 // SetTargeting sets the "targeting" field.
 func (u *AnnouncementUpsertOne) SetTargeting(v domain.AnnouncementTargeting) *AnnouncementUpsertOne {
 	return u.Update(func(s *AnnouncementUpsert) {
@@ -1002,6 +1058,20 @@ func (u *AnnouncementUpsertBulk) UpdateStatus() *AnnouncementUpsertBulk {
 	})
 }
 
+// SetNotifyMode sets the "notify_mode" field.
+func (u *AnnouncementUpsertBulk) SetNotifyMode(v string) *AnnouncementUpsertBulk {
+	return u.Update(func(s *AnnouncementUpsert) {
+		s.SetNotifyMode(v)
+	})
+}
+
+// UpdateNotifyMode sets the "notify_mode" field to the value that was provided on create.
+func (u *AnnouncementUpsertBulk) UpdateNotifyMode() *AnnouncementUpsertBulk {
+	return u.Update(func(s *AnnouncementUpsert) {
+		s.UpdateNotifyMode()
+	})
+}
+
 // SetTargeting sets the "targeting" field.
 func (u *AnnouncementUpsertBulk) SetTargeting(v domain.AnnouncementTargeting) *AnnouncementUpsertBulk {
 	return u.Update(func(s *AnnouncementUpsert) {

backend/ent/announcement_update.go

@@ -72,6 +72,20 @@ func (_u *AnnouncementUpdate) SetNillableStatus(v *string) *AnnouncementUpdate {
 	return _u
 }
 
+// SetNotifyMode sets the "notify_mode" field.
+func (_u *AnnouncementUpdate) SetNotifyMode(v string) *AnnouncementUpdate {
+	_u.mutation.SetNotifyMode(v)
+	return _u
+}
+
+// SetNillableNotifyMode sets the "notify_mode" field if the given value is not nil.
+func (_u *AnnouncementUpdate) SetNillableNotifyMode(v *string) *AnnouncementUpdate {
+	if v != nil {
+		_u.SetNotifyMode(*v)
+	}
+	return _u
+}
+
 // SetTargeting sets the "targeting" field.
 func (_u *AnnouncementUpdate) SetTargeting(v domain.AnnouncementTargeting) *AnnouncementUpdate {
 	_u.mutation.SetTargeting(v)
@@ -286,6 +300,11 @@ func (_u *AnnouncementUpdate) check() error {
 			return &ValidationError{Name: "status", err: fmt.Errorf(`ent: validator failed for field "Announcement.status": %w`, err)}
 		}
 	}
+	if v, ok := _u.mutation.NotifyMode(); ok {
+		if err := announcement.NotifyModeValidator(v); err != nil {
+			return &ValidationError{Name: "notify_mode", err: fmt.Errorf(`ent: validator failed for field "Announcement.notify_mode": %w`, err)}
+		}
+	}
 	return nil
 }
 
@@ -310,6 +329,9 @@ func (_u *AnnouncementUpdate) sqlSave(ctx context.Context) (_node int, err error
 	if value, ok := _u.mutation.Status(); ok {
 		_spec.SetField(announcement.FieldStatus, field.TypeString, value)
 	}
+	if value, ok := _u.mutation.NotifyMode(); ok {
+		_spec.SetField(announcement.FieldNotifyMode, field.TypeString, value)
+	}
 	if value, ok := _u.mutation.Targeting(); ok {
 		_spec.SetField(announcement.FieldTargeting, field.TypeJSON, value)
 	}
@@ -456,6 +478,20 @@ func (_u *AnnouncementUpdateOne) SetNillableStatus(v *string) *AnnouncementUpdat
 	return _u
 }
 
+// SetNotifyMode sets the "notify_mode" field.
+func (_u *AnnouncementUpdateOne) SetNotifyMode(v string) *AnnouncementUpdateOne {
+	_u.mutation.SetNotifyMode(v)
+	return _u
+}
+
+// SetNillableNotifyMode sets the "notify_mode" field if the given value is not nil.
+func (_u *AnnouncementUpdateOne) SetNillableNotifyMode(v *string) *AnnouncementUpdateOne {
+	if v != nil {
+		_u.SetNotifyMode(*v)
+	}
+	return _u
+}
+
 // SetTargeting sets the "targeting" field.
 func (_u *AnnouncementUpdateOne) SetTargeting(v domain.AnnouncementTargeting) *AnnouncementUpdateOne {
 	_u.mutation.SetTargeting(v)
@@ -683,6 +719,11 @@ func (_u *AnnouncementUpdateOne) check() error {
 			return &ValidationError{Name: "status", err: fmt.Errorf(`ent: validator failed for field "Announcement.status": %w`, err)}
 		}
 	}
+	if v, ok := _u.mutation.NotifyMode(); ok {
+		if err := announcement.NotifyModeValidator(v); err != nil {
+			return &ValidationError{Name: "notify_mode", err: fmt.Errorf(`ent: validator failed for field "Announcement.notify_mode": %w`, err)}
+		}
+	}
 	return nil
 }
 
@@ -724,6 +765,9 @@ func (_u *AnnouncementUpdateOne) sqlSave(ctx context.Context) (_node *Announceme
 	if value, ok := _u.mutation.Status(); ok {
 		_spec.SetField(announcement.FieldStatus, field.TypeString, value)
 	}
+	if value, ok := _u.mutation.NotifyMode(); ok {
+		_spec.SetField(announcement.FieldNotifyMode, field.TypeString, value)
+	}
 	if value, ok := _u.mutation.Targeting(); ok {
 		_spec.SetField(announcement.FieldTargeting, field.TypeJSON, value)
 	}

backend/ent/group.go

@@ -78,6 +78,10 @@ type Group struct {
 	SupportedModelScopes []string `json:"supported_model_scopes,omitempty"`
 	// 分组显示排序，数值越小越靠前
 	SortOrder int `json:"sort_order,omitempty"`
+	// 是否允许 /v1/messages 调度到此 OpenAI 分组
+	AllowMessagesDispatch bool `json:"allow_messages_dispatch,omitempty"`
+	// 默认映射模型 ID，当账号级映射找不到时使用此值
+	DefaultMappedModel string `json:"default_mapped_model,omitempty"`
 	// Edges holds the relations/edges for other nodes in the graph.
 	// The values are being populated by the GroupQuery when eager-loading is set.
 	Edges        GroupEdges `json:"edges"`
@@ -186,13 +190,13 @@ func (*Group) scanValues(columns []string) ([]any, error) {
 		switch columns[i] {
 		case group.FieldModelRouting, group.FieldSupportedModelScopes:
 			values[i] = new([]byte)
-		case group.FieldIsExclusive, group.FieldClaudeCodeOnly, group.FieldModelRoutingEnabled, group.FieldMcpXMLInject:
+		case group.FieldIsExclusive, group.FieldClaudeCodeOnly, group.FieldModelRoutingEnabled, group.FieldMcpXMLInject, group.FieldAllowMessagesDispatch:
 			values[i] = new(sql.NullBool)
 		case group.FieldRateMultiplier, group.FieldDailyLimitUsd, group.FieldWeeklyLimitUsd, group.FieldMonthlyLimitUsd, group.FieldImagePrice1k, group.FieldImagePrice2k, group.FieldImagePrice4k, group.FieldSoraImagePrice360, group.FieldSoraImagePrice540, group.FieldSoraVideoPricePerRequest, group.FieldSoraVideoPricePerRequestHd:
 			values[i] = new(sql.NullFloat64)
 		case group.FieldID, group.FieldDefaultValidityDays, group.FieldSoraStorageQuotaBytes, group.FieldFallbackGroupID, group.FieldFallbackGroupIDOnInvalidRequest, group.FieldSortOrder:
 			values[i] = new(sql.NullInt64)
-		case group.FieldName, group.FieldDescription, group.FieldStatus, group.FieldPlatform, group.FieldSubscriptionType:
+		case group.FieldName, group.FieldDescription, group.FieldStatus, group.FieldPlatform, group.FieldSubscriptionType, group.FieldDefaultMappedModel:
 			values[i] = new(sql.NullString)
 		case group.FieldCreatedAt, group.FieldUpdatedAt, group.FieldDeletedAt:
 			values[i] = new(sql.NullTime)
@@ -415,6 +419,18 @@ func (_m *Group) assignValues(columns []string, values []any) error {
 			} else if value.Valid {
 				_m.SortOrder = int(value.Int64)
 			}
+		case group.FieldAllowMessagesDispatch:
+			if value, ok := values[i].(*sql.NullBool); !ok {
+				return fmt.Errorf("unexpected type %T for field allow_messages_dispatch", values[i])
+			} else if value.Valid {
+				_m.AllowMessagesDispatch = value.Bool
+			}
+		case group.FieldDefaultMappedModel:
+			if value, ok := values[i].(*sql.NullString); !ok {
+				return fmt.Errorf("unexpected type %T for field default_mapped_model", values[i])
+			} else if value.Valid {
+				_m.DefaultMappedModel = value.String
+			}
 		default:
 			_m.selectValues.Set(columns[i], values[i])
 		}
@@ -608,6 +624,12 @@ func (_m *Group) String() string {
 	builder.WriteString(", ")
 	builder.WriteString("sort_order=")
 	builder.WriteString(fmt.Sprintf("%v", _m.SortOrder))
+	builder.WriteString(", ")
+	builder.WriteString("allow_messages_dispatch=")
+	builder.WriteString(fmt.Sprintf("%v", _m.AllowMessagesDispatch))
+	builder.WriteString(", ")
+	builder.WriteString("default_mapped_model=")
+	builder.WriteString(_m.DefaultMappedModel)
 	builder.WriteByte(')')
 	return builder.String()
 }

backend/ent/group/group.go

@@ -75,6 +75,10 @@ const (
 	FieldSupportedModelScopes = "supported_model_scopes"
 	// FieldSortOrder holds the string denoting the sort_order field in the database.
 	FieldSortOrder = "sort_order"
+	// FieldAllowMessagesDispatch holds the string denoting the allow_messages_dispatch field in the database.
+	FieldAllowMessagesDispatch = "allow_messages_dispatch"
+	// FieldDefaultMappedModel holds the string denoting the default_mapped_model field in the database.
+	FieldDefaultMappedModel = "default_mapped_model"
 	// EdgeAPIKeys holds the string denoting the api_keys edge name in mutations.
 	EdgeAPIKeys = "api_keys"
 	// EdgeRedeemCodes holds the string denoting the redeem_codes edge name in mutations.
@@ -180,6 +184,8 @@ var Columns = []string{
 	FieldMcpXMLInject,
 	FieldSupportedModelScopes,
 	FieldSortOrder,
+	FieldAllowMessagesDispatch,
+	FieldDefaultMappedModel,
 }
 
 var (
@@ -247,6 +253,12 @@ var (
 	DefaultSupportedModelScopes []string
 	// DefaultSortOrder holds the default value on creation for the "sort_order" field.
 	DefaultSortOrder int
+	// DefaultAllowMessagesDispatch holds the default value on creation for the "allow_messages_dispatch" field.
+	DefaultAllowMessagesDispatch bool
+	// DefaultDefaultMappedModel holds the default value on creation for the "default_mapped_model" field.
+	DefaultDefaultMappedModel string
+	// DefaultMappedModelValidator is a validator for the "default_mapped_model" field. It is called by the builders before save.
+	DefaultMappedModelValidator func(string) error
 )
 
 // OrderOption defines the ordering options for the Group queries.
@@ -397,6 +409,16 @@ func BySortOrder(opts ...sql.OrderTermOption) OrderOption {
 	return sql.OrderByField(FieldSortOrder, opts...).ToFunc()
 }
 
+// ByAllowMessagesDispatch orders the results by the allow_messages_dispatch field.
+func ByAllowMessagesDispatch(opts ...sql.OrderTermOption) OrderOption {
+	return sql.OrderByField(FieldAllowMessagesDispatch, opts...).ToFunc()
+}
+
+// ByDefaultMappedModel orders the results by the default_mapped_model field.
+func ByDefaultMappedModel(opts ...sql.OrderTermOption) OrderOption {
+	return sql.OrderByField(FieldDefaultMappedModel, opts...).ToFunc()
+}
+
 // ByAPIKeysCount orders the results by api_keys count.
 func ByAPIKeysCount(opts ...sql.OrderTermOption) OrderOption {
 	return func(s *sql.Selector) {

backend/ent/group/where.go

@@ -195,6 +195,16 @@ func SortOrder(v int) predicate.Group {
 	return predicate.Group(sql.FieldEQ(FieldSortOrder, v))
 }
 
+// AllowMessagesDispatch applies equality check predicate on the "allow_messages_dispatch" field. It's identical to AllowMessagesDispatchEQ.
+func AllowMessagesDispatch(v bool) predicate.Group {
+	return predicate.Group(sql.FieldEQ(FieldAllowMessagesDispatch, v))
+}
+
+// DefaultMappedModel applies equality check predicate on the "default_mapped_model" field. It's identical to DefaultMappedModelEQ.
+func DefaultMappedModel(v string) predicate.Group {
+	return predicate.Group(sql.FieldEQ(FieldDefaultMappedModel, v))
+}
+
 // CreatedAtEQ applies the EQ predicate on the "created_at" field.
 func CreatedAtEQ(v time.Time) predicate.Group {
 	return predicate.Group(sql.FieldEQ(FieldCreatedAt, v))
@@ -1470,6 +1480,81 @@ func SortOrderLTE(v int) predicate.Group {
 	return predicate.Group(sql.FieldLTE(FieldSortOrder, v))
 }
 
+// AllowMessagesDispatchEQ applies the EQ predicate on the "allow_messages_dispatch" field.
+func AllowMessagesDispatchEQ(v bool) predicate.Group {
+	return predicate.Group(sql.FieldEQ(FieldAllowMessagesDispatch, v))
+}
+
+// AllowMessagesDispatchNEQ applies the NEQ predicate on the "allow_messages_dispatch" field.
+func AllowMessagesDispatchNEQ(v bool) predicate.Group {
+	return predicate.Group(sql.FieldNEQ(FieldAllowMessagesDispatch, v))
+}
+
+// DefaultMappedModelEQ applies the EQ predicate on the "default_mapped_model" field.
+func DefaultMappedModelEQ(v string) predicate.Group {
+	return predicate.Group(sql.FieldEQ(FieldDefaultMappedModel, v))
+}
+
+// DefaultMappedModelNEQ applies the NEQ predicate on the "default_mapped_model" field.
+func DefaultMappedModelNEQ(v string) predicate.Group {
+	return predicate.Group(sql.FieldNEQ(FieldDefaultMappedModel, v))
+}
+
+// DefaultMappedModelIn applies the In predicate on the "default_mapped_model" field.
+func DefaultMappedModelIn(vs ...string) predicate.Group {
+	return predicate.Group(sql.FieldIn(FieldDefaultMappedModel, vs...))
+}
+
+// DefaultMappedModelNotIn applies the NotIn predicate on the "default_mapped_model" field.
+func DefaultMappedModelNotIn(vs ...string) predicate.Group {
+	return predicate.Group(sql.FieldNotIn(FieldDefaultMappedModel, vs...))
+}
+
+// DefaultMappedModelGT applies the GT predicate on the "default_mapped_model" field.
+func DefaultMappedModelGT(v string) predicate.Group {
+	return predicate.Group(sql.FieldGT(FieldDefaultMappedModel, v))
+}
+
+// DefaultMappedModelGTE applies the GTE predicate on the "default_mapped_model" field.
+func DefaultMappedModelGTE(v string) predicate.Group {
+	return predicate.Group(sql.FieldGTE(FieldDefaultMappedModel, v))
+}
+
+// DefaultMappedModelLT applies the LT predicate on the "default_mapped_model" field.
+func DefaultMappedModelLT(v string) predicate.Group {
+	return predicate.Group(sql.FieldLT(FieldDefaultMappedModel, v))
+}
+
+// DefaultMappedModelLTE applies the LTE predicate on the "default_mapped_model" field.
+func DefaultMappedModelLTE(v string) predicate.Group {
+	return predicate.Group(sql.FieldLTE(FieldDefaultMappedModel, v))
+}
+
+// DefaultMappedModelContains applies the Contains predicate on the "default_mapped_model" field.
+func DefaultMappedModelContains(v string) predicate.Group {
+	return predicate.Group(sql.FieldContains(FieldDefaultMappedModel, v))
+}
+
+// DefaultMappedModelHasPrefix applies the HasPrefix predicate on the "default_mapped_model" field.
+func DefaultMappedModelHasPrefix(v string) predicate.Group {
+	return predicate.Group(sql.FieldHasPrefix(FieldDefaultMappedModel, v))
+}
+
+// DefaultMappedModelHasSuffix applies the HasSuffix predicate on the "default_mapped_model" field.
+func DefaultMappedModelHasSuffix(v string) predicate.Group {
+	return predicate.Group(sql.FieldHasSuffix(FieldDefaultMappedModel, v))
+}
+
+// DefaultMappedModelEqualFold applies the EqualFold predicate on the "default_mapped_model" field.
+func DefaultMappedModelEqualFold(v string) predicate.Group {
+	return predicate.Group(sql.FieldEqualFold(FieldDefaultMappedModel, v))
+}
+
+// DefaultMappedModelContainsFold applies the ContainsFold predicate on the "default_mapped_model" field.
+func DefaultMappedModelContainsFold(v string) predicate.Group {
+	return predicate.Group(sql.FieldContainsFold(FieldDefaultMappedModel, v))
+}
+
 // HasAPIKeys applies the HasEdge predicate on the "api_keys" edge.
 func HasAPIKeys() predicate.Group {
 	return predicate.Group(func(s *sql.Selector) {

backend/ent/group_create.go

@@ -424,6 +424,34 @@ func (_c *GroupCreate) SetNillableSortOrder(v *int) *GroupCreate {
 	return _c
 }
 
+// SetAllowMessagesDispatch sets the "allow_messages_dispatch" field.
+func (_c *GroupCreate) SetAllowMessagesDispatch(v bool) *GroupCreate {
+	_c.mutation.SetAllowMessagesDispatch(v)
+	return _c
+}
+
+// SetNillableAllowMessagesDispatch sets the "allow_messages_dispatch" field if the given value is not nil.
+func (_c *GroupCreate) SetNillableAllowMessagesDispatch(v *bool) *GroupCreate {
+	if v != nil {
+		_c.SetAllowMessagesDispatch(*v)
+	}
+	return _c
+}
+
+// SetDefaultMappedModel sets the "default_mapped_model" field.
+func (_c *GroupCreate) SetDefaultMappedModel(v string) *GroupCreate {
+	_c.mutation.SetDefaultMappedModel(v)
+	return _c
+}
+
+// SetNillableDefaultMappedModel sets the "default_mapped_model" field if the given value is not nil.
+func (_c *GroupCreate) SetNillableDefaultMappedModel(v *string) *GroupCreate {
+	if v != nil {
+		_c.SetDefaultMappedModel(*v)
+	}
+	return _c
+}
+
 // AddAPIKeyIDs adds the "api_keys" edge to the APIKey entity by IDs.
 func (_c *GroupCreate) AddAPIKeyIDs(ids ...int64) *GroupCreate {
 	_c.mutation.AddAPIKeyIDs(ids...)
@@ -613,6 +641,14 @@ func (_c *GroupCreate) defaults() error {
 		v := group.DefaultSortOrder
 		_c.mutation.SetSortOrder(v)
 	}
+	if _, ok := _c.mutation.AllowMessagesDispatch(); !ok {
+		v := group.DefaultAllowMessagesDispatch
+		_c.mutation.SetAllowMessagesDispatch(v)
+	}
+	if _, ok := _c.mutation.DefaultMappedModel(); !ok {
+		v := group.DefaultDefaultMappedModel
+		_c.mutation.SetDefaultMappedModel(v)
+	}
 	return nil
 }
 
@@ -683,6 +719,17 @@ func (_c *GroupCreate) check() error {
 	if _, ok := _c.mutation.SortOrder(); !ok {
 		return &ValidationError{Name: "sort_order", err: errors.New(`ent: missing required field "Group.sort_order"`)}
 	}
+	if _, ok := _c.mutation.AllowMessagesDispatch(); !ok {
+		return &ValidationError{Name: "allow_messages_dispatch", err: errors.New(`ent: missing required field "Group.allow_messages_dispatch"`)}
+	}
+	if _, ok := _c.mutation.DefaultMappedModel(); !ok {
+		return &ValidationError{Name: "default_mapped_model", err: errors.New(`ent: missing required field "Group.default_mapped_model"`)}
+	}
+	if v, ok := _c.mutation.DefaultMappedModel(); ok {
+		if err := group.DefaultMappedModelValidator(v); err != nil {
+			return &ValidationError{Name: "default_mapped_model", err: fmt.Errorf(`ent: validator failed for field "Group.default_mapped_model": %w`, err)}
+		}
+	}
 	return nil
 }
 
@@ -830,6 +877,14 @@ func (_c *GroupCreate) createSpec() (*Group, *sqlgraph.CreateSpec) {
 		_spec.SetField(group.FieldSortOrder, field.TypeInt, value)
 		_node.SortOrder = value
 	}
+	if value, ok := _c.mutation.AllowMessagesDispatch(); ok {
+		_spec.SetField(group.FieldAllowMessagesDispatch, field.TypeBool, value)
+		_node.AllowMessagesDispatch = value
+	}
+	if value, ok := _c.mutation.DefaultMappedModel(); ok {
+		_spec.SetField(group.FieldDefaultMappedModel, field.TypeString, value)
+		_node.DefaultMappedModel = value
+	}
 	if nodes := _c.mutation.APIKeysIDs(); len(nodes) > 0 {
 		edge := &sqlgraph.EdgeSpec{
 			Rel:     sqlgraph.O2M,
@@ -1520,6 +1575,30 @@ func (u *GroupUpsert) AddSortOrder(v int) *GroupUpsert {
 	return u
 }
 
+// SetAllowMessagesDispatch sets the "allow_messages_dispatch" field.
+func (u *GroupUpsert) SetAllowMessagesDispatch(v bool) *GroupUpsert {
+	u.Set(group.FieldAllowMessagesDispatch, v)
+	return u
+}
+
+// UpdateAllowMessagesDispatch sets the "allow_messages_dispatch" field to the value that was provided on create.
+func (u *GroupUpsert) UpdateAllowMessagesDispatch() *GroupUpsert {
+	u.SetExcluded(group.FieldAllowMessagesDispatch)
+	return u
+}
+
+// SetDefaultMappedModel sets the "default_mapped_model" field.
+func (u *GroupUpsert) SetDefaultMappedModel(v string) *GroupUpsert {
+	u.Set(group.FieldDefaultMappedModel, v)
+	return u
+}
+
+// UpdateDefaultMappedModel sets the "default_mapped_model" field to the value that was provided on create.
+func (u *GroupUpsert) UpdateDefaultMappedModel() *GroupUpsert {
+	u.SetExcluded(group.FieldDefaultMappedModel)
+	return u
+}
+
 // UpdateNewValues updates the mutable fields using the new values that were set on create.
 // Using this option is equivalent to using:
 //
@@ -2188,6 +2267,34 @@ func (u *GroupUpsertOne) UpdateSortOrder() *GroupUpsertOne {
 	})
 }
 
+// SetAllowMessagesDispatch sets the "allow_messages_dispatch" field.
+func (u *GroupUpsertOne) SetAllowMessagesDispatch(v bool) *GroupUpsertOne {
+	return u.Update(func(s *GroupUpsert) {
+		s.SetAllowMessagesDispatch(v)
+	})
+}
+
+// UpdateAllowMessagesDispatch sets the "allow_messages_dispatch" field to the value that was provided on create.
+func (u *GroupUpsertOne) UpdateAllowMessagesDispatch() *GroupUpsertOne {
+	return u.Update(func(s *GroupUpsert) {
+		s.UpdateAllowMessagesDispatch()
+	})
+}
+
+// SetDefaultMappedModel sets the "default_mapped_model" field.
+func (u *GroupUpsertOne) SetDefaultMappedModel(v string) *GroupUpsertOne {
+	return u.Update(func(s *GroupUpsert) {
+		s.SetDefaultMappedModel(v)
+	})
+}
+
+// UpdateDefaultMappedModel sets the "default_mapped_model" field to the value that was provided on create.
+func (u *GroupUpsertOne) UpdateDefaultMappedModel() *GroupUpsertOne {
+	return u.Update(func(s *GroupUpsert) {
+		s.UpdateDefaultMappedModel()
+	})
+}
+
 // Exec executes the query.
 func (u *GroupUpsertOne) Exec(ctx context.Context) error {
 	if len(u.create.conflict) == 0 {
@@ -3022,6 +3129,34 @@ func (u *GroupUpsertBulk) UpdateSortOrder() *GroupUpsertBulk {
 	})
 }
 
+// SetAllowMessagesDispatch sets the "allow_messages_dispatch" field.
+func (u *GroupUpsertBulk) SetAllowMessagesDispatch(v bool) *GroupUpsertBulk {
+	return u.Update(func(s *GroupUpsert) {
+		s.SetAllowMessagesDispatch(v)
+	})
+}
+
+// UpdateAllowMessagesDispatch sets the "allow_messages_dispatch" field to the value that was provided on create.
+func (u *GroupUpsertBulk) UpdateAllowMessagesDispatch() *GroupUpsertBulk {
+	return u.Update(func(s *GroupUpsert) {
+		s.UpdateAllowMessagesDispatch()
+	})
+}
+
+// SetDefaultMappedModel sets the "default_mapped_model" field.
+func (u *GroupUpsertBulk) SetDefaultMappedModel(v string) *GroupUpsertBulk {
+	return u.Update(func(s *GroupUpsert) {
+		s.SetDefaultMappedModel(v)
+	})
+}
+
+// UpdateDefaultMappedModel sets the "default_mapped_model" field to the value that was provided on create.
+func (u *GroupUpsertBulk) UpdateDefaultMappedModel() *GroupUpsertBulk {
+	return u.Update(func(s *GroupUpsert) {
+		s.UpdateDefaultMappedModel()
+	})
+}
+
 // Exec executes the query.
 func (u *GroupUpsertBulk) Exec(ctx context.Context) error {
 	if u.create.err != nil {

backend/ent/group_update.go

@@ -625,6 +625,34 @@ func (_u *GroupUpdate) AddSortOrder(v int) *GroupUpdate {
 	return _u
 }
 
+// SetAllowMessagesDispatch sets the "allow_messages_dispatch" field.
+func (_u *GroupUpdate) SetAllowMessagesDispatch(v bool) *GroupUpdate {
+	_u.mutation.SetAllowMessagesDispatch(v)
+	return _u
+}
+
+// SetNillableAllowMessagesDispatch sets the "allow_messages_dispatch" field if the given value is not nil.
+func (_u *GroupUpdate) SetNillableAllowMessagesDispatch(v *bool) *GroupUpdate {
+	if v != nil {
+		_u.SetAllowMessagesDispatch(*v)
+	}
+	return _u
+}
+
+// SetDefaultMappedModel sets the "default_mapped_model" field.
+func (_u *GroupUpdate) SetDefaultMappedModel(v string) *GroupUpdate {
+	_u.mutation.SetDefaultMappedModel(v)
+	return _u
+}
+
+// SetNillableDefaultMappedModel sets the "default_mapped_model" field if the given value is not nil.
+func (_u *GroupUpdate) SetNillableDefaultMappedModel(v *string) *GroupUpdate {
+	if v != nil {
+		_u.SetDefaultMappedModel(*v)
+	}
+	return _u
+}
+
 // AddAPIKeyIDs adds the "api_keys" edge to the APIKey entity by IDs.
 func (_u *GroupUpdate) AddAPIKeyIDs(ids ...int64) *GroupUpdate {
 	_u.mutation.AddAPIKeyIDs(ids...)
@@ -910,6 +938,11 @@ func (_u *GroupUpdate) check() error {
 			return &ValidationError{Name: "subscription_type", err: fmt.Errorf(`ent: validator failed for field "Group.subscription_type": %w`, err)}
 		}
 	}
+	if v, ok := _u.mutation.DefaultMappedModel(); ok {
+		if err := group.DefaultMappedModelValidator(v); err != nil {
+			return &ValidationError{Name: "default_mapped_model", err: fmt.Errorf(`ent: validator failed for field "Group.default_mapped_model": %w`, err)}
+		}
+	}
 	return nil
 }
 
@@ -1110,6 +1143,12 @@ func (_u *GroupUpdate) sqlSave(ctx context.Context) (_node int, err error) {
 	if value, ok := _u.mutation.AddedSortOrder(); ok {
 		_spec.AddField(group.FieldSortOrder, field.TypeInt, value)
 	}
+	if value, ok := _u.mutation.AllowMessagesDispatch(); ok {
+		_spec.SetField(group.FieldAllowMessagesDispatch, field.TypeBool, value)
+	}
+	if value, ok := _u.mutation.DefaultMappedModel(); ok {
+		_spec.SetField(group.FieldDefaultMappedModel, field.TypeString, value)
+	}
 	if _u.mutation.APIKeysCleared() {
 		edge := &sqlgraph.EdgeSpec{
 			Rel:     sqlgraph.O2M,
@@ -2014,6 +2053,34 @@ func (_u *GroupUpdateOne) AddSortOrder(v int) *GroupUpdateOne {
 	return _u
 }
 
+// SetAllowMessagesDispatch sets the "allow_messages_dispatch" field.
+func (_u *GroupUpdateOne) SetAllowMessagesDispatch(v bool) *GroupUpdateOne {
+	_u.mutation.SetAllowMessagesDispatch(v)
+	return _u
+}
+
+// SetNillableAllowMessagesDispatch sets the "allow_messages_dispatch" field if the given value is not nil.
+func (_u *GroupUpdateOne) SetNillableAllowMessagesDispatch(v *bool) *GroupUpdateOne {
+	if v != nil {
+		_u.SetAllowMessagesDispatch(*v)
+	}
+	return _u
+}
+
+// SetDefaultMappedModel sets the "default_mapped_model" field.
+func (_u *GroupUpdateOne) SetDefaultMappedModel(v string) *GroupUpdateOne {
+	_u.mutation.SetDefaultMappedModel(v)
+	return _u
+}
+
+// SetNillableDefaultMappedModel sets the "default_mapped_model" field if the given value is not nil.
+func (_u *GroupUpdateOne) SetNillableDefaultMappedModel(v *string) *GroupUpdateOne {
+	if v != nil {
+		_u.SetDefaultMappedModel(*v)
+	}
+	return _u
+}
+
 // AddAPIKeyIDs adds the "api_keys" edge to the APIKey entity by IDs.
 func (_u *GroupUpdateOne) AddAPIKeyIDs(ids ...int64) *GroupUpdateOne {
 	_u.mutation.AddAPIKeyIDs(ids...)
@@ -2312,6 +2379,11 @@ func (_u *GroupUpdateOne) check() error {
 			return &ValidationError{Name: "subscription_type", err: fmt.Errorf(`ent: validator failed for field "Group.subscription_type": %w`, err)}
 		}
 	}
+	if v, ok := _u.mutation.DefaultMappedModel(); ok {
+		if err := group.DefaultMappedModelValidator(v); err != nil {
+			return &ValidationError{Name: "default_mapped_model", err: fmt.Errorf(`ent: validator failed for field "Group.default_mapped_model": %w`, err)}
+		}
+	}
 	return nil
 }
 
@@ -2529,6 +2601,12 @@ func (_u *GroupUpdateOne) sqlSave(ctx context.Context) (_node *Group, err error)
 	if value, ok := _u.mutation.AddedSortOrder(); ok {
 		_spec.AddField(group.FieldSortOrder, field.TypeInt, value)
 	}
+	if value, ok := _u.mutation.AllowMessagesDispatch(); ok {
+		_spec.SetField(group.FieldAllowMessagesDispatch, field.TypeBool, value)
+	}
+	if value, ok := _u.mutation.DefaultMappedModel(); ok {
+		_spec.SetField(group.FieldDefaultMappedModel, field.TypeString, value)
+	}
 	if _u.mutation.APIKeysCleared() {
 		edge := &sqlgraph.EdgeSpec{
 			Rel:     sqlgraph.O2M,

backend/ent/migrate/schema.go

@@ -106,6 +106,7 @@ var (
 		{Name: "credentials", Type: field.TypeJSON, SchemaType: map[string]string{"postgres": "jsonb"}},
 		{Name: "extra", Type: field.TypeJSON, SchemaType: map[string]string{"postgres": "jsonb"}},
 		{Name: "concurrency", Type: field.TypeInt, Default: 3},
+		{Name: "load_factor", Type: field.TypeInt, Nullable: true},
 		{Name: "priority", Type: field.TypeInt, Default: 50},
 		{Name: "rate_multiplier", Type: field.TypeFloat64, Default: 1, SchemaType: map[string]string{"postgres": "decimal(10,4)"}},
 		{Name: "status", Type: field.TypeString, Size: 20, Default: "active"},
@@ -132,7 +133,7 @@ var (
 		ForeignKeys: []*schema.ForeignKey{
 			{
 				Symbol:     "accounts_proxies_proxy",
-				Columns:    []*schema.Column{AccountsColumns[27]},
+				Columns:    []*schema.Column{AccountsColumns[28]},
 				RefColumns: []*schema.Column{ProxiesColumns[0]},
 				OnDelete:   schema.SetNull,
 			},
@@ -151,52 +152,52 @@ var (
 			{
 				Name:    "account_status",
 				Unique:  false,
-				Columns: []*schema.Column{AccountsColumns[13]},
+				Columns: []*schema.Column{AccountsColumns[14]},
 			},
 			{
 				Name:    "account_proxy_id",
 				Unique:  false,
-				Columns: []*schema.Column{AccountsColumns[27]},
+				Columns: []*schema.Column{AccountsColumns[28]},
 			},
 			{
 				Name:    "account_priority",
 				Unique:  false,
-				Columns: []*schema.Column{AccountsColumns[11]},
+				Columns: []*schema.Column{AccountsColumns[12]},
 			},
 			{
 				Name:    "account_last_used_at",
 				Unique:  false,
-				Columns: []*schema.Column{AccountsColumns[15]},
+				Columns: []*schema.Column{AccountsColumns[16]},
 			},
 			{
 				Name:    "account_schedulable",
 				Unique:  false,
-				Columns: []*schema.Column{AccountsColumns[18]},
+				Columns: []*schema.Column{AccountsColumns[19]},
 			},
 			{
 				Name:    "account_rate_limited_at",
 				Unique:  false,
-				Columns: []*schema.Column{AccountsColumns[19]},
+				Columns: []*schema.Column{AccountsColumns[20]},
 			},
 			{
 				Name:    "account_rate_limit_reset_at",
 				Unique:  false,
-				Columns: []*schema.Column{AccountsColumns[20]},
+				Columns: []*schema.Column{AccountsColumns[21]},
 			},
 			{
 				Name:    "account_overload_until",
 				Unique:  false,
-				Columns: []*schema.Column{AccountsColumns[21]},
+				Columns: []*schema.Column{AccountsColumns[22]},
 			},
 			{
 				Name:    "account_platform_priority",
 				Unique:  false,
-				Columns: []*schema.Column{AccountsColumns[6], AccountsColumns[11]},
+				Columns: []*schema.Column{AccountsColumns[6], AccountsColumns[12]},
 			},
 			{
 				Name:    "account_priority_status",
 				Unique:  false,
-				Columns: []*schema.Column{AccountsColumns[11], AccountsColumns[13]},
+				Columns: []*schema.Column{AccountsColumns[12], AccountsColumns[14]},
 			},
 			{
 				Name:    "account_deleted_at",
@@ -250,6 +251,7 @@ var (
 		{Name: "title", Type: field.TypeString, Size: 200},
 		{Name: "content", Type: field.TypeString, SchemaType: map[string]string{"postgres": "text"}},
 		{Name: "status", Type: field.TypeString, Size: 20, Default: "draft"},
+		{Name: "notify_mode", Type: field.TypeString, Size: 20, Default: "silent"},
 		{Name: "targeting", Type: field.TypeJSON, Nullable: true, SchemaType: map[string]string{"postgres": "jsonb"}},
 		{Name: "starts_at", Type: field.TypeTime, Nullable: true, SchemaType: map[string]string{"postgres": "timestamptz"}},
 		{Name: "ends_at", Type: field.TypeTime, Nullable: true, SchemaType: map[string]string{"postgres": "timestamptz"}},
@@ -272,17 +274,17 @@ var (
 			{
 				Name:    "announcement_created_at",
 				Unique:  false,
-				Columns: []*schema.Column{AnnouncementsColumns[9]},
+				Columns: []*schema.Column{AnnouncementsColumns[10]},
 			},
 			{
 				Name:    "announcement_starts_at",
 				Unique:  false,
-				Columns: []*schema.Column{AnnouncementsColumns[5]},
+				Columns: []*schema.Column{AnnouncementsColumns[6]},
 			},
 			{
 				Name:    "announcement_ends_at",
 				Unique:  false,
-				Columns: []*schema.Column{AnnouncementsColumns[6]},
+				Columns: []*schema.Column{AnnouncementsColumns[7]},
 			},
 		},
 	}
@@ -406,6 +408,8 @@ var (
 		{Name: "mcp_xml_inject", Type: field.TypeBool, Default: true},
 		{Name: "supported_model_scopes", Type: field.TypeJSON, SchemaType: map[string]string{"postgres": "jsonb"}},
 		{Name: "sort_order", Type: field.TypeInt, Default: 0},
+		{Name: "allow_messages_dispatch", Type: field.TypeBool, Default: false},
+		{Name: "default_mapped_model", Type: field.TypeString, Size: 100, Default: ""},
 	}
 	// GroupsTable holds the schema information for the "groups" table.
 	GroupsTable = &schema.Table{

backend/ent/mutation.go

@@ -2260,6 +2260,8 @@ type AccountMutation struct {
 	extra                     *map[string]interface{}
 	concurrency               *int
 	addconcurrency            *int
+	load_factor               *int
+	addload_factor            *int
 	priority                  *int
 	addpriority               *int
 	rate_multiplier           *float64
@@ -2845,6 +2847,76 @@ func (m *AccountMutation) ResetConcurrency() {
 	m.addconcurrency = nil
 }
 
+// SetLoadFactor sets the "load_factor" field.
+func (m *AccountMutation) SetLoadFactor(i int) {
+	m.load_factor = &i
+	m.addload_factor = nil
+}
+
+// LoadFactor returns the value of the "load_factor" field in the mutation.
+func (m *AccountMutation) LoadFactor() (r int, exists bool) {
+	v := m.load_factor
+	if v == nil {
+		return
+	}
+	return *v, true
+}
+
+// OldLoadFactor returns the old "load_factor" field's value of the Account entity.
+// If the Account object wasn't provided to the builder, the object is fetched from the database.
+// An error is returned if the mutation operation is not UpdateOne, or the database query fails.
+func (m *AccountMutation) OldLoadFactor(ctx context.Context) (v *int, err error) {
+	if !m.op.Is(OpUpdateOne) {
+		return v, errors.New("OldLoadFactor is only allowed on UpdateOne operations")
+	}
+	if m.id == nil || m.oldValue == nil {
+		return v, errors.New("OldLoadFactor requires an ID field in the mutation")
+	}
+	oldValue, err := m.oldValue(ctx)
+	if err != nil {
+		return v, fmt.Errorf("querying old value for OldLoadFactor: %w", err)
+	}
+	return oldValue.LoadFactor, nil
+}
+
+// AddLoadFactor adds i to the "load_factor" field.
+func (m *AccountMutation) AddLoadFactor(i int) {
+	if m.addload_factor != nil {
+		*m.addload_factor += i
+	} else {
+		m.addload_factor = &i
+	}
+}
+
+// AddedLoadFactor returns the value that was added to the "load_factor" field in this mutation.
+func (m *AccountMutation) AddedLoadFactor() (r int, exists bool) {
+	v := m.addload_factor
+	if v == nil {
+		return
+	}
+	return *v, true
+}
+
+// ClearLoadFactor clears the value of the "load_factor" field.
+func (m *AccountMutation) ClearLoadFactor() {
+	m.load_factor = nil
+	m.addload_factor = nil
+	m.clearedFields[account.FieldLoadFactor] = struct{}{}
+}
+
+// LoadFactorCleared returns if the "load_factor" field was cleared in this mutation.
+func (m *AccountMutation) LoadFactorCleared() bool {
+	_, ok := m.clearedFields[account.FieldLoadFactor]
+	return ok
+}
+
+// ResetLoadFactor resets all changes to the "load_factor" field.
+func (m *AccountMutation) ResetLoadFactor() {
+	m.load_factor = nil
+	m.addload_factor = nil
+	delete(m.clearedFields, account.FieldLoadFactor)
+}
+
 // SetPriority sets the "priority" field.
 func (m *AccountMutation) SetPriority(i int) {
 	m.priority = &i
@@ -3773,7 +3845,7 @@ func (m *AccountMutation) Type() string {
 // order to get all numeric fields that were incremented/decremented, call
 // AddedFields().
 func (m *AccountMutation) Fields() []string {
-	fields := make([]string, 0, 27)
+	fields := make([]string, 0, 28)
 	if m.created_at != nil {
 		fields = append(fields, account.FieldCreatedAt)
 	}
@@ -3807,6 +3879,9 @@ func (m *AccountMutation) Fields() []string {
 	if m.concurrency != nil {
 		fields = append(fields, account.FieldConcurrency)
 	}
+	if m.load_factor != nil {
+		fields = append(fields, account.FieldLoadFactor)
+	}
 	if m.priority != nil {
 		fields = append(fields, account.FieldPriority)
 	}
@@ -3885,6 +3960,8 @@ func (m *AccountMutation) Field(name string) (ent.Value, bool) {
 		return m.ProxyID()
 	case account.FieldConcurrency:
 		return m.Concurrency()
+	case account.FieldLoadFactor:
+		return m.LoadFactor()
 	case account.FieldPriority:
 		return m.Priority()
 	case account.FieldRateMultiplier:
@@ -3948,6 +4025,8 @@ func (m *AccountMutation) OldField(ctx context.Context, name string) (ent.Value,
 		return m.OldProxyID(ctx)
 	case account.FieldConcurrency:
 		return m.OldConcurrency(ctx)
+	case account.FieldLoadFactor:
+		return m.OldLoadFactor(ctx)
 	case account.FieldPriority:
 		return m.OldPriority(ctx)
 	case account.FieldRateMultiplier:
@@ -4066,6 +4145,13 @@ func (m *AccountMutation) SetField(name string, value ent.Value) error {
 		}
 		m.SetConcurrency(v)
 		return nil
+	case account.FieldLoadFactor:
+		v, ok := value.(int)
+		if !ok {
+			return fmt.Errorf("unexpected type %T for field %s", value, name)
+		}
+		m.SetLoadFactor(v)
+		return nil
 	case account.FieldPriority:
 		v, ok := value.(int)
 		if !ok {
@@ -4189,6 +4275,9 @@ func (m *AccountMutation) AddedFields() []string {
 	if m.addconcurrency != nil {
 		fields = append(fields, account.FieldConcurrency)
 	}
+	if m.addload_factor != nil {
+		fields = append(fields, account.FieldLoadFactor)
+	}
 	if m.addpriority != nil {
 		fields = append(fields, account.FieldPriority)
 	}
@@ -4205,6 +4294,8 @@ func (m *AccountMutation) AddedField(name string) (ent.Value, bool) {
 	switch name {
 	case account.FieldConcurrency:
 		return m.AddedConcurrency()
+	case account.FieldLoadFactor:
+		return m.AddedLoadFactor()
 	case account.FieldPriority:
 		return m.AddedPriority()
 	case account.FieldRateMultiplier:
@@ -4225,6 +4316,13 @@ func (m *AccountMutation) AddField(name string, value ent.Value) error {
 		}
 		m.AddConcurrency(v)
 		return nil
+	case account.FieldLoadFactor:
+		v, ok := value.(int)
+		if !ok {
+			return fmt.Errorf("unexpected type %T for field %s", value, name)
+		}
+		m.AddLoadFactor(v)
+		return nil
 	case account.FieldPriority:
 		v, ok := value.(int)
 		if !ok {
@@ -4256,6 +4354,9 @@ func (m *AccountMutation) ClearedFields() []string {
 	if m.FieldCleared(account.FieldProxyID) {
 		fields = append(fields, account.FieldProxyID)
 	}
+	if m.FieldCleared(account.FieldLoadFactor) {
+		fields = append(fields, account.FieldLoadFactor)
+	}
 	if m.FieldCleared(account.FieldErrorMessage) {
 		fields = append(fields, account.FieldErrorMessage)
 	}
@@ -4312,6 +4413,9 @@ func (m *AccountMutation) ClearField(name string) error {
 	case account.FieldProxyID:
 		m.ClearProxyID()
 		return nil
+	case account.FieldLoadFactor:
+		m.ClearLoadFactor()
+		return nil
 	case account.FieldErrorMessage:
 		m.ClearErrorMessage()
 		return nil
@@ -4386,6 +4490,9 @@ func (m *AccountMutation) ResetField(name string) error {
 	case account.FieldConcurrency:
 		m.ResetConcurrency()
 		return nil
+	case account.FieldLoadFactor:
+		m.ResetLoadFactor()
+		return nil
 	case account.FieldPriority:
 		m.ResetPriority()
 		return nil
@@ -5060,6 +5167,7 @@ type AnnouncementMutation struct {
 	title         *string
 	content       *string
 	status        *string
+	notify_mode   *string
 	targeting     *domain.AnnouncementTargeting
 	starts_at     *time.Time
 	ends_at       *time.Time
@@ -5284,6 +5392,42 @@ func (m *AnnouncementMutation) ResetStatus() {
 	m.status = nil
 }
 
+// SetNotifyMode sets the "notify_mode" field.
+func (m *AnnouncementMutation) SetNotifyMode(s string) {
+	m.notify_mode = &s
+}
+
+// NotifyMode returns the value of the "notify_mode" field in the mutation.
+func (m *AnnouncementMutation) NotifyMode() (r string, exists bool) {
+	v := m.notify_mode
+	if v == nil {
+		return
+	}
+	return *v, true
+}
+
+// OldNotifyMode returns the old "notify_mode" field's value of the Announcement entity.
+// If the Announcement object wasn't provided to the builder, the object is fetched from the database.
+// An error is returned if the mutation operation is not UpdateOne, or the database query fails.
+func (m *AnnouncementMutation) OldNotifyMode(ctx context.Context) (v string, err error) {
+	if !m.op.Is(OpUpdateOne) {
+		return v, errors.New("OldNotifyMode is only allowed on UpdateOne operations")
+	}
+	if m.id == nil || m.oldValue == nil {
+		return v, errors.New("OldNotifyMode requires an ID field in the mutation")
+	}
+	oldValue, err := m.oldValue(ctx)
+	if err != nil {
+		return v, fmt.Errorf("querying old value for OldNotifyMode: %w", err)
+	}
+	return oldValue.NotifyMode, nil
+}
+
+// ResetNotifyMode resets all changes to the "notify_mode" field.
+func (m *AnnouncementMutation) ResetNotifyMode() {
+	m.notify_mode = nil
+}
+
 // SetTargeting sets the "targeting" field.
 func (m *AnnouncementMutation) SetTargeting(dt domain.AnnouncementTargeting) {
 	m.targeting = &dt
@@ -5731,7 +5875,7 @@ func (m *AnnouncementMutation) Type() string {
 // order to get all numeric fields that were incremented/decremented, call
 // AddedFields().
 func (m *AnnouncementMutation) Fields() []string {
-	fields := make([]string, 0, 10)
+	fields := make([]string, 0, 11)
 	if m.title != nil {
 		fields = append(fields, announcement.FieldTitle)
 	}
@@ -5741,6 +5885,9 @@ func (m *AnnouncementMutation) Fields() []string {
 	if m.status != nil {
 		fields = append(fields, announcement.FieldStatus)
 	}
+	if m.notify_mode != nil {
+		fields = append(fields, announcement.FieldNotifyMode)
+	}
 	if m.targeting != nil {
 		fields = append(fields, announcement.FieldTargeting)
 	}
@@ -5776,6 +5923,8 @@ func (m *AnnouncementMutation) Field(name string) (ent.Value, bool) {
 		return m.Content()
 	case announcement.FieldStatus:
 		return m.Status()
+	case announcement.FieldNotifyMode:
+		return m.NotifyMode()
 	case announcement.FieldTargeting:
 		return m.Targeting()
 	case announcement.FieldStartsAt:
@@ -5805,6 +5954,8 @@ func (m *AnnouncementMutation) OldField(ctx context.Context, name string) (ent.V
 		return m.OldContent(ctx)
 	case announcement.FieldStatus:
 		return m.OldStatus(ctx)
+	case announcement.FieldNotifyMode:
+		return m.OldNotifyMode(ctx)
 	case announcement.FieldTargeting:
 		return m.OldTargeting(ctx)
 	case announcement.FieldStartsAt:
@@ -5849,6 +6000,13 @@ func (m *AnnouncementMutation) SetField(name string, value ent.Value) error {
 		}
 		m.SetStatus(v)
 		return nil
+	case announcement.FieldNotifyMode:
+		v, ok := value.(string)
+		if !ok {
+			return fmt.Errorf("unexpected type %T for field %s", value, name)
+		}
+		m.SetNotifyMode(v)
+		return nil
 	case announcement.FieldTargeting:
 		v, ok := value.(domain.AnnouncementTargeting)
 		if !ok {
@@ -6016,6 +6174,9 @@ func (m *AnnouncementMutation) ResetField(name string) error {
 	case announcement.FieldStatus:
 		m.ResetStatus()
 		return nil
+	case announcement.FieldNotifyMode:
+		m.ResetNotifyMode()
+		return nil
 	case announcement.FieldTargeting:
 		m.ResetTargeting()
 		return nil
@@ -8089,6 +8250,8 @@ type GroupMutation struct {
 	appendsupported_model_scopes            []string
 	sort_order                              *int
 	addsort_order                           *int
+	allow_messages_dispatch                 *bool
+	default_mapped_model                    *string
 	clearedFields                           map[string]struct{}
 	api_keys                                map[int64]struct{}
 	removedapi_keys                         map[int64]struct{}
@@ -9833,6 +9996,78 @@ func (m *GroupMutation) ResetSortOrder() {
 	m.addsort_order = nil
 }
 
+// SetAllowMessagesDispatch sets the "allow_messages_dispatch" field.
+func (m *GroupMutation) SetAllowMessagesDispatch(b bool) {
+	m.allow_messages_dispatch = &b
+}
+
+// AllowMessagesDispatch returns the value of the "allow_messages_dispatch" field in the mutation.
+func (m *GroupMutation) AllowMessagesDispatch() (r bool, exists bool) {
+	v := m.allow_messages_dispatch
+	if v == nil {
+		return
+	}
+	return *v, true
+}
+
+// OldAllowMessagesDispatch returns the old "allow_messages_dispatch" field's value of the Group entity.
+// If the Group object wasn't provided to the builder, the object is fetched from the database.
+// An error is returned if the mutation operation is not UpdateOne, or the database query fails.
+func (m *GroupMutation) OldAllowMessagesDispatch(ctx context.Context) (v bool, err error) {
+	if !m.op.Is(OpUpdateOne) {
+		return v, errors.New("OldAllowMessagesDispatch is only allowed on UpdateOne operations")
+	}
+	if m.id == nil || m.oldValue == nil {
+		return v, errors.New("OldAllowMessagesDispatch requires an ID field in the mutation")
+	}
+	oldValue, err := m.oldValue(ctx)
+	if err != nil {
+		return v, fmt.Errorf("querying old value for OldAllowMessagesDispatch: %w", err)
+	}
+	return oldValue.AllowMessagesDispatch, nil
+}
+
+// ResetAllowMessagesDispatch resets all changes to the "allow_messages_dispatch" field.
+func (m *GroupMutation) ResetAllowMessagesDispatch() {
+	m.allow_messages_dispatch = nil
+}
+
+// SetDefaultMappedModel sets the "default_mapped_model" field.
+func (m *GroupMutation) SetDefaultMappedModel(s string) {
+	m.default_mapped_model = &s
+}
+
+// DefaultMappedModel returns the value of the "default_mapped_model" field in the mutation.
+func (m *GroupMutation) DefaultMappedModel() (r string, exists bool) {
+	v := m.default_mapped_model
+	if v == nil {
+		return
+	}
+	return *v, true
+}
+
+// OldDefaultMappedModel returns the old "default_mapped_model" field's value of the Group entity.
+// If the Group object wasn't provided to the builder, the object is fetched from the database.
+// An error is returned if the mutation operation is not UpdateOne, or the database query fails.
+func (m *GroupMutation) OldDefaultMappedModel(ctx context.Context) (v string, err error) {
+	if !m.op.Is(OpUpdateOne) {
+		return v, errors.New("OldDefaultMappedModel is only allowed on UpdateOne operations")
+	}
+	if m.id == nil || m.oldValue == nil {
+		return v, errors.New("OldDefaultMappedModel requires an ID field in the mutation")
+	}
+	oldValue, err := m.oldValue(ctx)
+	if err != nil {
+		return v, fmt.Errorf("querying old value for OldDefaultMappedModel: %w", err)
+	}
+	return oldValue.DefaultMappedModel, nil
+}
+
+// ResetDefaultMappedModel resets all changes to the "default_mapped_model" field.
+func (m *GroupMutation) ResetDefaultMappedModel() {
+	m.default_mapped_model = nil
+}
+
 // AddAPIKeyIDs adds the "api_keys" edge to the APIKey entity by ids.
 func (m *GroupMutation) AddAPIKeyIDs(ids ...int64) {
 	if m.api_keys == nil {
@@ -10191,7 +10426,7 @@ func (m *GroupMutation) Type() string {
 // order to get all numeric fields that were incremented/decremented, call
 // AddedFields().
 func (m *GroupMutation) Fields() []string {
-	fields := make([]string, 0, 30)
+	fields := make([]string, 0, 32)
 	if m.created_at != nil {
 		fields = append(fields, group.FieldCreatedAt)
 	}
@@ -10282,6 +10517,12 @@ func (m *GroupMutation) Fields() []string {
 	if m.sort_order != nil {
 		fields = append(fields, group.FieldSortOrder)
 	}
+	if m.allow_messages_dispatch != nil {
+		fields = append(fields, group.FieldAllowMessagesDispatch)
+	}
+	if m.default_mapped_model != nil {
+		fields = append(fields, group.FieldDefaultMappedModel)
+	}
 	return fields
 }
 
@@ -10350,6 +10591,10 @@ func (m *GroupMutation) Field(name string) (ent.Value, bool) {
 		return m.SupportedModelScopes()
 	case group.FieldSortOrder:
 		return m.SortOrder()
+	case group.FieldAllowMessagesDispatch:
+		return m.AllowMessagesDispatch()
+	case group.FieldDefaultMappedModel:
+		return m.DefaultMappedModel()
 	}
 	return nil, false
 }
@@ -10419,6 +10664,10 @@ func (m *GroupMutation) OldField(ctx context.Context, name string) (ent.Value, e
 		return m.OldSupportedModelScopes(ctx)
 	case group.FieldSortOrder:
 		return m.OldSortOrder(ctx)
+	case group.FieldAllowMessagesDispatch:
+		return m.OldAllowMessagesDispatch(ctx)
+	case group.FieldDefaultMappedModel:
+		return m.OldDefaultMappedModel(ctx)
 	}
 	return nil, fmt.Errorf("unknown Group field %s", name)
 }
@@ -10638,6 +10887,20 @@ func (m *GroupMutation) SetField(name string, value ent.Value) error {
 		}
 		m.SetSortOrder(v)
 		return nil
+	case group.FieldAllowMessagesDispatch:
+		v, ok := value.(bool)
+		if !ok {
+			return fmt.Errorf("unexpected type %T for field %s", value, name)
+		}
+		m.SetAllowMessagesDispatch(v)
+		return nil
+	case group.FieldDefaultMappedModel:
+		v, ok := value.(string)
+		if !ok {
+			return fmt.Errorf("unexpected type %T for field %s", value, name)
+		}
+		m.SetDefaultMappedModel(v)
+		return nil
 	}
 	return fmt.Errorf("unknown Group field %s", name)
 }
@@ -11065,6 +11328,12 @@ func (m *GroupMutation) ResetField(name string) error {
 	case group.FieldSortOrder:
 		m.ResetSortOrder()
 		return nil
+	case group.FieldAllowMessagesDispatch:
+		m.ResetAllowMessagesDispatch()
+		return nil
+	case group.FieldDefaultMappedModel:
+		m.ResetDefaultMappedModel()
+		return nil
 	}
 	return fmt.Errorf("unknown Group field %s", name)
 }

backend/ent/runtime/runtime.go

@@ -212,29 +212,29 @@ func init() {
 	// account.DefaultConcurrency holds the default value on creation for the concurrency field.
 	account.DefaultConcurrency = accountDescConcurrency.Default.(int)
 	// accountDescPriority is the schema descriptor for priority field.
-	accountDescPriority := accountFields[8].Descriptor()
+	accountDescPriority := accountFields[9].Descriptor()
 	// account.DefaultPriority holds the default value on creation for the priority field.
 	account.DefaultPriority = accountDescPriority.Default.(int)
 	// accountDescRateMultiplier is the schema descriptor for rate_multiplier field.
-	accountDescRateMultiplier := accountFields[9].Descriptor()
+	accountDescRateMultiplier := accountFields[10].Descriptor()
 	// account.DefaultRateMultiplier holds the default value on creation for the rate_multiplier field.
 	account.DefaultRateMultiplier = accountDescRateMultiplier.Default.(float64)
 	// accountDescStatus is the schema descriptor for status field.
-	accountDescStatus := accountFields[10].Descriptor()
+	accountDescStatus := accountFields[11].Descriptor()
 	// account.DefaultStatus holds the default value on creation for the status field.
 	account.DefaultStatus = accountDescStatus.Default.(string)
 	// account.StatusValidator is a validator for the "status" field. It is called by the builders before save.
 	account.StatusValidator = accountDescStatus.Validators[0].(func(string) error)
 	// accountDescAutoPauseOnExpired is the schema descriptor for auto_pause_on_expired field.
-	accountDescAutoPauseOnExpired := accountFields[14].Descriptor()
+	accountDescAutoPauseOnExpired := accountFields[15].Descriptor()
 	// account.DefaultAutoPauseOnExpired holds the default value on creation for the auto_pause_on_expired field.
 	account.DefaultAutoPauseOnExpired = accountDescAutoPauseOnExpired.Default.(bool)
 	// accountDescSchedulable is the schema descriptor for schedulable field.
-	accountDescSchedulable := accountFields[15].Descriptor()
+	accountDescSchedulable := accountFields[16].Descriptor()
 	// account.DefaultSchedulable holds the default value on creation for the schedulable field.
 	account.DefaultSchedulable = accountDescSchedulable.Default.(bool)
 	// accountDescSessionWindowStatus is the schema descriptor for session_window_status field.
-	accountDescSessionWindowStatus := accountFields[23].Descriptor()
+	accountDescSessionWindowStatus := accountFields[24].Descriptor()
 	// account.SessionWindowStatusValidator is a validator for the "session_window_status" field. It is called by the builders before save.
 	account.SessionWindowStatusValidator = accountDescSessionWindowStatus.Validators[0].(func(string) error)
 	accountgroupFields := schema.AccountGroup{}.Fields()
@@ -277,12 +277,18 @@ func init() {
 	announcement.DefaultStatus = announcementDescStatus.Default.(string)
 	// announcement.StatusValidator is a validator for the "status" field. It is called by the builders before save.
 	announcement.StatusValidator = announcementDescStatus.Validators[0].(func(string) error)
+	// announcementDescNotifyMode is the schema descriptor for notify_mode field.
+	announcementDescNotifyMode := announcementFields[3].Descriptor()
+	// announcement.DefaultNotifyMode holds the default value on creation for the notify_mode field.
+	announcement.DefaultNotifyMode = announcementDescNotifyMode.Default.(string)
+	// announcement.NotifyModeValidator is a validator for the "notify_mode" field. It is called by the builders before save.
+	announcement.NotifyModeValidator = announcementDescNotifyMode.Validators[0].(func(string) error)
 	// announcementDescCreatedAt is the schema descriptor for created_at field.
-	announcementDescCreatedAt := announcementFields[8].Descriptor()
+	announcementDescCreatedAt := announcementFields[9].Descriptor()
 	// announcement.DefaultCreatedAt holds the default value on creation for the created_at field.
 	announcement.DefaultCreatedAt = announcementDescCreatedAt.Default.(func() time.Time)
 	// announcementDescUpdatedAt is the schema descriptor for updated_at field.
-	announcementDescUpdatedAt := announcementFields[9].Descriptor()
+	announcementDescUpdatedAt := announcementFields[10].Descriptor()
 	// announcement.DefaultUpdatedAt holds the default value on creation for the updated_at field.
 	announcement.DefaultUpdatedAt = announcementDescUpdatedAt.Default.(func() time.Time)
 	// announcement.UpdateDefaultUpdatedAt holds the default value on update for the updated_at field.
@@ -447,6 +453,16 @@ func init() {
 	groupDescSortOrder := groupFields[26].Descriptor()
 	// group.DefaultSortOrder holds the default value on creation for the sort_order field.
 	group.DefaultSortOrder = groupDescSortOrder.Default.(int)
+	// groupDescAllowMessagesDispatch is the schema descriptor for allow_messages_dispatch field.
+	groupDescAllowMessagesDispatch := groupFields[27].Descriptor()
+	// group.DefaultAllowMessagesDispatch holds the default value on creation for the allow_messages_dispatch field.
+	group.DefaultAllowMessagesDispatch = groupDescAllowMessagesDispatch.Default.(bool)
+	// groupDescDefaultMappedModel is the schema descriptor for default_mapped_model field.
+	groupDescDefaultMappedModel := groupFields[28].Descriptor()
+	// group.DefaultDefaultMappedModel holds the default value on creation for the default_mapped_model field.
+	group.DefaultDefaultMappedModel = groupDescDefaultMappedModel.Default.(string)
+	// group.DefaultMappedModelValidator is a validator for the "default_mapped_model" field. It is called by the builders before save.
+	group.DefaultMappedModelValidator = groupDescDefaultMappedModel.Validators[0].(func(string) error)
 	idempotencyrecordMixin := schema.IdempotencyRecord{}.Mixin()
 	idempotencyrecordMixinFields0 := idempotencyrecordMixin[0].Fields()
 	_ = idempotencyrecordMixinFields0

backend/ent/schema/account.go

@@ -97,6 +97,8 @@ func (Account) Fields() []ent.Field {
 		field.Int("concurrency").
 			Default(3),
 
+		field.Int("load_factor").Optional().Nillable(),
+
 		// priority: 账户优先级，数值越小优先级越高
 		// 调度器会优先使用高优先级的账户
 		field.Int("priority").

backend/ent/schema/announcement.go

@@ -41,6 +41,10 @@ func (Announcement) Fields() []ent.Field {
 			MaxLen(20).
 			Default(domain.AnnouncementStatusDraft).
 			Comment("状态: draft, active, archived"),
+		field.String("notify_mode").
+			MaxLen(20).
+			Default(domain.AnnouncementNotifyModeSilent).
+			Comment("通知模式: silent(仅铃铛), popup(弹窗提醒)"),
 		field.JSON("targeting", domain.AnnouncementTargeting{}).
 			Optional().
 			SchemaType(map[string]string{dialect.Postgres: "jsonb"}).

backend/ent/schema/group.go

@@ -148,6 +148,15 @@ func (Group) Fields() []ent.Field {
 		field.Int("sort_order").
 			Default(0).
 			Comment("分组显示排序，数值越小越靠前"),
+
+		// OpenAI Messages 调度配置 (added by migration 069)
+		field.Bool("allow_messages_dispatch").
+			Default(false).
+			Comment("是否允许 /v1/messages 调度到此 OpenAI 分组"),
+		field.String("default_mapped_model").
+			MaxLen(100).
+			Default("").
+			Comment("默认映射模型 ID，当账号级映射找不到时使用此值"),
 	}
 }
 

backend/go.mod

@@ -1,12 +1,13 @@
 module github.com/Wei-Shaw/sub2api
 
-go 1.25.7
+go 1.26.1
 
 require (
 	entgo.io/ent v0.14.5
 	github.com/DATA-DOG/go-sqlmock v1.5.2
 	github.com/DouDOU-start/go-sora2api v1.1.0
 	github.com/alitto/pond/v2 v2.6.2
+	github.com/aws/aws-sdk-go-v2 v1.41.3
 	github.com/aws/aws-sdk-go-v2/config v1.32.10
 	github.com/aws/aws-sdk-go-v2/credentials v1.19.10
 	github.com/aws/aws-sdk-go-v2/service/s3 v1.96.2
@@ -38,8 +39,6 @@ require (
 	golang.org/x/net v0.49.0
 	golang.org/x/sync v0.19.0
 	golang.org/x/term v0.40.0
-	google.golang.org/grpc v1.75.1
-	google.golang.org/protobuf v1.36.10
 	gopkg.in/natefinch/lumberjack.v2 v2.2.1
 	gopkg.in/yaml.v3 v3.0.1
 	modernc.org/sqlite v1.44.3
@@ -53,7 +52,6 @@ require (
 	github.com/agext/levenshtein v1.2.3 // indirect
 	github.com/andybalholm/brotli v1.2.0 // indirect
 	github.com/apparentlymart/go-textseg/v15 v15.0.0 // indirect
-	github.com/aws/aws-sdk-go-v2 v1.41.2 // indirect
 	github.com/aws/aws-sdk-go-v2/aws/protocol/eventstream v1.7.5 // indirect
 	github.com/aws/aws-sdk-go-v2/feature/ec2/imds v1.18.18 // indirect
 	github.com/aws/aws-sdk-go-v2/internal/configsources v1.4.18 // indirect
@@ -68,7 +66,7 @@ require (
 	github.com/aws/aws-sdk-go-v2/service/sso v1.30.11 // indirect
 	github.com/aws/aws-sdk-go-v2/service/ssooidc v1.35.15 // indirect
 	github.com/aws/aws-sdk-go-v2/service/sts v1.41.7 // indirect
-	github.com/aws/smithy-go v1.24.1 // indirect
+	github.com/aws/smithy-go v1.24.2 // indirect
 	github.com/bdandy/go-errors v1.2.2 // indirect
 	github.com/bdandy/go-socks4 v1.2.3 // indirect
 	github.com/bmatcuk/doublestar v1.3.4 // indirect
@@ -109,7 +107,6 @@ require (
 	github.com/goccy/go-json v0.10.2 // indirect
 	github.com/google/go-cmp v0.7.0 // indirect
 	github.com/google/go-querystring v1.1.0 // indirect
-	github.com/google/subcommands v1.2.0 // indirect
 	github.com/grpc-ecosystem/grpc-gateway/v2 v2.27.3 // indirect
 	github.com/hashicorp/hcl v1.0.0 // indirect
 	github.com/hashicorp/hcl/v2 v2.18.1 // indirect
@@ -169,6 +166,7 @@ require (
 	go.opentelemetry.io/contrib/instrumentation/net/http/otelhttp v0.49.0 // indirect
 	go.opentelemetry.io/otel v1.37.0 // indirect
 	go.opentelemetry.io/otel/metric v1.37.0 // indirect
+	go.opentelemetry.io/otel/sdk v1.37.0 // indirect
 	go.opentelemetry.io/otel/trace v1.37.0 // indirect
 	go.uber.org/atomic v1.10.0 // indirect
 	go.uber.org/automaxprocs v1.6.0 // indirect
@@ -178,8 +176,8 @@ require (
 	golang.org/x/mod v0.32.0 // indirect
 	golang.org/x/sys v0.41.0 // indirect
 	golang.org/x/text v0.34.0 // indirect
-	golang.org/x/tools v0.41.0 // indirect
-	google.golang.org/genproto/googleapis/rpc v0.0.0-20250929231259-57b25ae835d4 // indirect
+	google.golang.org/grpc v1.75.1 // indirect
+	google.golang.org/protobuf v1.36.10 // indirect
 	gopkg.in/ini.v1 v1.67.0 // indirect
 	modernc.org/libc v1.67.6 // indirect
 	modernc.org/mathutil v1.7.1 // indirect

backend/go.sum

@@ -24,6 +24,8 @@ github.com/apparentlymart/go-textseg/v15 v15.0.0 h1:uYvfpb3DyLSCGWnctWKGj857c6ew
 github.com/apparentlymart/go-textseg/v15 v15.0.0/go.mod h1:K8XmNZdhEBkdlyDdvbmmsvpAG721bKi0joRfFdHIWJ4=
 github.com/aws/aws-sdk-go-v2 v1.41.2 h1:LuT2rzqNQsauaGkPK/7813XxcZ3o3yePY0Iy891T2ls=
 github.com/aws/aws-sdk-go-v2 v1.41.2/go.mod h1:IvvlAZQXvTXznUPfRVfryiG1fbzE2NGK6m9u39YQ+S4=
+github.com/aws/aws-sdk-go-v2 v1.41.3 h1:4kQ/fa22KjDt13QCy1+bYADvdgcxpfH18f0zP542kZA=
+github.com/aws/aws-sdk-go-v2 v1.41.3/go.mod h1:mwsPRE8ceUUpiTgF7QmQIJ7lgsKUPQOUl3o72QBrE1o=
 github.com/aws/aws-sdk-go-v2/aws/protocol/eventstream v1.7.5 h1:zWFmPmgw4sveAYi1mRqG+E/g0461cJ5M4bJ8/nc6d3Q=
 github.com/aws/aws-sdk-go-v2/aws/protocol/eventstream v1.7.5/go.mod h1:nVUlMLVV8ycXSb7mSkcNu9e3v/1TJq2RTlrPwhYWr5c=
 github.com/aws/aws-sdk-go-v2/config v1.32.10 h1:9DMthfO6XWZYLfzZglAgW5Fyou2nRI5CuV44sTedKBI=
@@ -60,6 +62,8 @@ github.com/aws/aws-sdk-go-v2/service/sts v1.41.7 h1:NITQpgo9A5NrDZ57uOWj+abvXSb8
 github.com/aws/aws-sdk-go-v2/service/sts v1.41.7/go.mod h1:sks5UWBhEuWYDPdwlnRFn1w7xWdH29Jcpe+/PJQefEs=
 github.com/aws/smithy-go v1.24.1 h1:VbyeNfmYkWoxMVpGUAbQumkODcYmfMRfZ8yQiH30SK0=
 github.com/aws/smithy-go v1.24.1/go.mod h1:LEj2LM3rBRQJxPZTB4KuzZkaZYnZPnvgIhb4pu07mx0=
+github.com/aws/smithy-go v1.24.2 h1:FzA3bu/nt/vDvmnkg+R8Xl46gmzEDam6mZ1hzmwXFng=
+github.com/aws/smithy-go v1.24.2/go.mod h1:YE2RhdIuDbA5E5bTdciG9KrW3+TiEONeUWCqxX9i1Fc=
 github.com/bdandy/go-errors v1.2.2 h1:WdFv/oukjTJCLa79UfkGmwX7ZxONAihKu4V0mLIs11Q=
 github.com/bdandy/go-errors v1.2.2/go.mod h1:NkYHl4Fey9oRRdbB1CoC6e84tuqQHiqrOcZpqFEkBxM=
 github.com/bdandy/go-socks4 v1.2.3 h1:Q6Y2heY1GRjCtHbmlKfnwrKVU/k81LS8mRGLRlmDlic=
@@ -171,8 +175,6 @@ github.com/goccy/go-json v0.10.2 h1:CrxCmQqYDkv1z7lO7Wbh2HN93uovUHgrECaO5ZrCXAU=
 github.com/goccy/go-json v0.10.2/go.mod h1:6MelG93GURQebXPDq3khkgXZkazVtN9CRI+MGFi0w8I=
 github.com/golang-jwt/jwt/v5 v5.2.2 h1:Rl4B7itRWVtYIHFrSNd7vhTiz9UpLdi6gZhZ3wEeDy8=
 github.com/golang-jwt/jwt/v5 v5.2.2/go.mod h1:pqrtFR0X4osieyHYxtmOUWsAWrfe1Q5UVIyoH402zdk=
-github.com/golang/protobuf v1.5.4 h1:i7eJL8qZTpSEXOPTxNKhASYpMn+8e5Q6AdndVa1dWek=
-github.com/golang/protobuf v1.5.4/go.mod h1:lnTiLA8Wa4RWRcIUkrtSVa5nRhsEGBg48fD6rSs7xps=
 github.com/google/go-cmp v0.5.2/go.mod h1:v8dTdLbMG2kIc/vJvl+f65V22dbkXbowE6jgT/gNBxE=
 github.com/google/go-cmp v0.5.6/go.mod h1:v8dTdLbMG2kIc/vJvl+f65V22dbkXbowE6jgT/gNBxE=
 github.com/google/go-cmp v0.7.0 h1:wk8382ETsv4JYUZwIsn6YpYiWiBsYLSJiTsyBybVuN8=
@@ -182,8 +184,6 @@ github.com/google/go-querystring v1.1.0/go.mod h1:Kcdr2DB4koayq7X8pmAG4sNG59So17
 github.com/google/gofuzz v1.0.0/go.mod h1:dBl0BpW6vV/+mYPU4Po3pmUjxk6FQPldtuIdl/M65Eg=
 github.com/google/pprof v0.0.0-20250317173921-a4b03ec1a45e h1:ijClszYn+mADRFY17kjQEVQ1XRhq2/JR1M3sGqeJoxs=
 github.com/google/pprof v0.0.0-20250317173921-a4b03ec1a45e/go.mod h1:boTsfXsheKC2y+lKOCMpSfarhxDeIzfZG1jqGcPl3cA=
-github.com/google/subcommands v1.2.0 h1:vWQspBTo2nEqTUFita5/KeEWlUL8kQObDFbub/EN9oE=
-github.com/google/subcommands v1.2.0/go.mod h1:ZjhPrFU+Olkh9WazFPsl27BQ4UPiG37m3yTrtFlrHVk=
 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/google/wire v0.7.0 h1:JxUKI6+CVBgCO2WToKy/nQk0sS+amI9z9EjVmdaocj4=
@@ -398,8 +398,6 @@ go.opentelemetry.io/otel/metric v1.37.0 h1:mvwbQS5m0tbmqML4NqK+e3aDiO02vsf/Wgbsd
 go.opentelemetry.io/otel/metric v1.37.0/go.mod h1:04wGrZurHYKOc+RKeye86GwKiTb9FKm1WHtO+4EVr2E=
 go.opentelemetry.io/otel/sdk v1.37.0 h1:ItB0QUqnjesGRvNcmAcU0LyvkVyGJ2xftD29bWdDvKI=
 go.opentelemetry.io/otel/sdk v1.37.0/go.mod h1:VredYzxUvuo2q3WRcDnKDjbdvmO0sCzOvVAiY+yUkAg=
-go.opentelemetry.io/otel/sdk/metric v1.37.0 h1:90lI228XrB9jCMuSdA0673aubgRobVZFhbjxHHspCPc=
-go.opentelemetry.io/otel/sdk/metric v1.37.0/go.mod h1:cNen4ZWfiD37l5NhS+Keb5RXVWZWpRE+9WyVCpbo5ps=
 go.opentelemetry.io/otel/trace v1.37.0 h1:HLdcFNbRQBE2imdSEgm/kwqmQj1Or1l/7bW6mxVK7z4=
 go.opentelemetry.io/otel/trace v1.37.0/go.mod h1:TlgrlQ+PtQO5XFerSPUYG0JSgGyryXewPGyayAWSBS0=
 go.opentelemetry.io/proto/otlp v1.3.1 h1:TrMUixzpM0yuc/znrFTP9MMRh8trP93mkCiDVeXrui0=
@@ -455,8 +453,6 @@ golang.org/x/tools v0.0.0-20180917221912-90fa682c2a6e/go.mod h1:n7NCudcB/nEzxVGm
 golang.org/x/tools v0.41.0 h1:a9b8iMweWG+S0OBnlU36rzLp20z1Rp10w+IY2czHTQc=
 golang.org/x/tools v0.41.0/go.mod h1:XSY6eDqxVNiYgezAVqqCeihT4j1U2CCsqvH3WhQpnlg=
 golang.org/x/xerrors v0.0.0-20191204190536-9bdfabe68543/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
-gonum.org/v1/gonum v0.16.0 h1:5+ul4Swaf3ESvrOnidPp4GZbzf0mxVQpDCYUQE7OJfk=
-gonum.org/v1/gonum v0.16.0/go.mod h1:fef3am4MQ93R2HHpKnLk4/Tbh/s0+wqD5nfa6Pnwy4E=
 google.golang.org/genproto v0.0.0-20231106174013-bbf56f31fb17 h1:wpZ8pe2x1Q3f2KyT5f8oP/fa9rHAKgFPr/HZdNuS+PQ=
 google.golang.org/genproto/googleapis/api v0.0.0-20250929231259-57b25ae835d4 h1:8XJ4pajGwOlasW+L13MnEGA8W4115jJySQtVfS2/IBU=
 google.golang.org/genproto/googleapis/api v0.0.0-20250929231259-57b25ae835d4/go.mod h1:NnuHhy+bxcg30o7FnVAZbXsPHUDQ9qKWAQKCD7VxFtk=

backend/internal/config/config.go

@@ -934,9 +934,10 @@ type DashboardAggregationConfig struct {
 
 // DashboardAggregationRetentionConfig 预聚合保留窗口
 type DashboardAggregationRetentionConfig struct {
-	UsageLogsDays int `mapstructure:"usage_logs_days"`
-	HourlyDays    int `mapstructure:"hourly_days"`
-	DailyDays     int `mapstructure:"daily_days"`
+	UsageLogsDays         int `mapstructure:"usage_logs_days"`
+	UsageBillingDedupDays int `mapstructure:"usage_billing_dedup_days"`
+	HourlyDays            int `mapstructure:"hourly_days"`
+	DailyDays             int `mapstructure:"daily_days"`
 }
 
 // UsageCleanupConfig 使用记录清理任务配置
@@ -1301,6 +1302,7 @@ func setDefaults() {
 	viper.SetDefault("dashboard_aggregation.backfill_enabled", false)
 	viper.SetDefault("dashboard_aggregation.backfill_max_days", 31)
 	viper.SetDefault("dashboard_aggregation.retention.usage_logs_days", 90)
+	viper.SetDefault("dashboard_aggregation.retention.usage_billing_dedup_days", 365)
 	viper.SetDefault("dashboard_aggregation.retention.hourly_days", 180)
 	viper.SetDefault("dashboard_aggregation.retention.daily_days", 730)
 	viper.SetDefault("dashboard_aggregation.recompute_days", 2)
@@ -1402,7 +1404,7 @@ func setDefaults() {
 	viper.SetDefault("gateway.concurrency_slot_ttl_minutes", 30) // 并发槽位过期时间（支持超长请求）
 	viper.SetDefault("gateway.stream_data_interval_timeout", 180)
 	viper.SetDefault("gateway.stream_keepalive_interval", 10)
-	viper.SetDefault("gateway.max_line_size", 40*1024*1024)
+	viper.SetDefault("gateway.max_line_size", 500*1024*1024)
 	viper.SetDefault("gateway.scheduling.sticky_session_max_waiting", 3)
 	viper.SetDefault("gateway.scheduling.sticky_session_wait_timeout", 120*time.Second)
 	viper.SetDefault("gateway.scheduling.fallback_wait_timeout", 30*time.Second)
@@ -1758,6 +1760,12 @@ func (c *Config) Validate() error {
 		if c.DashboardAgg.Retention.UsageLogsDays <= 0 {
 			return fmt.Errorf("dashboard_aggregation.retention.usage_logs_days must be positive")
 		}
+		if c.DashboardAgg.Retention.UsageBillingDedupDays <= 0 {
+			return fmt.Errorf("dashboard_aggregation.retention.usage_billing_dedup_days must be positive")
+		}
+		if c.DashboardAgg.Retention.UsageBillingDedupDays < c.DashboardAgg.Retention.UsageLogsDays {
+			return fmt.Errorf("dashboard_aggregation.retention.usage_billing_dedup_days must be greater than or equal to usage_logs_days")
+		}
 		if c.DashboardAgg.Retention.HourlyDays <= 0 {
 			return fmt.Errorf("dashboard_aggregation.retention.hourly_days must be positive")
 		}
@@ -1780,6 +1788,14 @@ func (c *Config) Validate() error {
 		if c.DashboardAgg.Retention.UsageLogsDays < 0 {
 			return fmt.Errorf("dashboard_aggregation.retention.usage_logs_days must be non-negative")
 		}
+		if c.DashboardAgg.Retention.UsageBillingDedupDays < 0 {
+			return fmt.Errorf("dashboard_aggregation.retention.usage_billing_dedup_days must be non-negative")
+		}
+		if c.DashboardAgg.Retention.UsageBillingDedupDays > 0 &&
+			c.DashboardAgg.Retention.UsageLogsDays > 0 &&
+			c.DashboardAgg.Retention.UsageBillingDedupDays < c.DashboardAgg.Retention.UsageLogsDays {
+			return fmt.Errorf("dashboard_aggregation.retention.usage_billing_dedup_days must be greater than or equal to usage_logs_days")
+		}
 		if c.DashboardAgg.Retention.HourlyDays < 0 {
 			return fmt.Errorf("dashboard_aggregation.retention.hourly_days must be non-negative")
 		}

backend/internal/config/config_test.go

@@ -441,6 +441,9 @@ func TestLoadDefaultDashboardAggregationConfig(t *testing.T) {
 	if cfg.DashboardAgg.Retention.UsageLogsDays != 90 {
 		t.Fatalf("DashboardAgg.Retention.UsageLogsDays = %d, want 90", cfg.DashboardAgg.Retention.UsageLogsDays)
 	}
+	if cfg.DashboardAgg.Retention.UsageBillingDedupDays != 365 {
+		t.Fatalf("DashboardAgg.Retention.UsageBillingDedupDays = %d, want 365", cfg.DashboardAgg.Retention.UsageBillingDedupDays)
+	}
 	if cfg.DashboardAgg.Retention.HourlyDays != 180 {
 		t.Fatalf("DashboardAgg.Retention.HourlyDays = %d, want 180", cfg.DashboardAgg.Retention.HourlyDays)
 	}
@@ -1016,6 +1019,23 @@ func TestValidateConfigErrors(t *testing.T) {
 			mutate:  func(c *Config) { c.DashboardAgg.Enabled = true; c.DashboardAgg.Retention.UsageLogsDays = 0 },
 			wantErr: "dashboard_aggregation.retention.usage_logs_days",
 		},
+		{
+			name: "dashboard aggregation dedup retention",
+			mutate: func(c *Config) {
+				c.DashboardAgg.Enabled = true
+				c.DashboardAgg.Retention.UsageBillingDedupDays = 0
+			},
+			wantErr: "dashboard_aggregation.retention.usage_billing_dedup_days",
+		},
+		{
+			name: "dashboard aggregation dedup retention smaller than usage logs",
+			mutate: func(c *Config) {
+				c.DashboardAgg.Enabled = true
+				c.DashboardAgg.Retention.UsageLogsDays = 30
+				c.DashboardAgg.Retention.UsageBillingDedupDays = 29
+			},
+			wantErr: "dashboard_aggregation.retention.usage_billing_dedup_days",
+		},
 		{
 			name:    "dashboard aggregation disabled interval",
 			mutate:  func(c *Config) { c.DashboardAgg.Enabled = false; c.DashboardAgg.IntervalSeconds = -1 },

backend/internal/domain/announcement.go

@@ -13,6 +13,11 @@ const (
 	AnnouncementStatusArchived = "archived"
 )
 
+const (
+	AnnouncementNotifyModeSilent = "silent"
+	AnnouncementNotifyModePopup  = "popup"
+)
+
 const (
 	AnnouncementConditionTypeSubscription = "subscription"
 	AnnouncementConditionTypeBalance      = "balance"
@@ -195,17 +200,18 @@ func (c AnnouncementCondition) validate() error {
 }
 
 type Announcement struct {
-	ID        int64
-	Title     string
-	Content   string
-	Status    string
-	Targeting AnnouncementTargeting
-	StartsAt  *time.Time
-	EndsAt    *time.Time
-	CreatedBy *int64
-	UpdatedBy *int64
-	CreatedAt time.Time
-	UpdatedAt time.Time
+	ID         int64
+	Title      string
+	Content    string
+	Status     string
+	NotifyMode string
+	Targeting  AnnouncementTargeting
+	StartsAt   *time.Time
+	EndsAt     *time.Time
+	CreatedBy  *int64
+	UpdatedBy  *int64
+	CreatedAt  time.Time
+	UpdatedAt  time.Time
 }
 
 func (a *Announcement) IsActiveAt(now time.Time) bool {

backend/internal/domain/constants.go

@@ -31,6 +31,7 @@ const (
 	AccountTypeSetupToken = "setup-token" // Setup Token类型账号（inference only scope）
 	AccountTypeAPIKey     = "apikey"      // API Key类型账号
 	AccountTypeUpstream   = "upstream"    // 上游透传类型账号（通过 Base URL + API Key 连接上游）
+	AccountTypeBedrock    = "bedrock"     // AWS Bedrock 类型账号（通过 SigV4 签名或 API Key 连接 Bedrock，由 credentials.auth_mode 区分）
 )
 
 // Redeem type constants
@@ -84,10 +85,12 @@ var DefaultAntigravityModelMapping = map[string]string{
 	"claude-haiku-4-5":          "claude-sonnet-4-5",
 	"claude-haiku-4-5-20251001": "claude-sonnet-4-5",
 	// Gemini 2.5 白名单
-	"gemini-2.5-flash":          "gemini-2.5-flash",
-	"gemini-2.5-flash-lite":     "gemini-2.5-flash-lite",
-	"gemini-2.5-flash-thinking": "gemini-2.5-flash-thinking",
-	"gemini-2.5-pro":            "gemini-2.5-pro",
+	"gemini-2.5-flash":               "gemini-2.5-flash",
+	"gemini-2.5-flash-image":         "gemini-2.5-flash-image",
+	"gemini-2.5-flash-image-preview": "gemini-2.5-flash-image",
+	"gemini-2.5-flash-lite":          "gemini-2.5-flash-lite",
+	"gemini-2.5-flash-thinking":      "gemini-2.5-flash-thinking",
+	"gemini-2.5-pro":                 "gemini-2.5-pro",
 	// Gemini 3 白名单
 	"gemini-3-flash":    "gemini-3-flash",
 	"gemini-3-pro-high": "gemini-3-pro-high",
@@ -111,3 +114,27 @@ var DefaultAntigravityModelMapping = map[string]string{
 	"gpt-oss-120b-medium":    "gpt-oss-120b-medium",
 	"tab_flash_lite_preview": "tab_flash_lite_preview",
 }
+
+// DefaultBedrockModelMapping 是 AWS Bedrock 平台的默认模型映射
+// 将 Anthropic 标准模型名映射到 Bedrock 模型 ID
+// 注意：此处的 "us." 前缀仅为默认值，ResolveBedrockModelID 会根据账号配置的
+// aws_region 自动调整为匹配的区域前缀（如 eu.、apac.、jp. 等）
+var DefaultBedrockModelMapping = map[string]string{
+	// Claude Opus
+	"claude-opus-4-6-thinking": "us.anthropic.claude-opus-4-6-v1",
+	"claude-opus-4-6":          "us.anthropic.claude-opus-4-6-v1",
+	"claude-opus-4-5-thinking": "us.anthropic.claude-opus-4-5-20251101-v1:0",
+	"claude-opus-4-5-20251101": "us.anthropic.claude-opus-4-5-20251101-v1:0",
+	"claude-opus-4-1":          "us.anthropic.claude-opus-4-1-20250805-v1:0",
+	"claude-opus-4-20250514":   "us.anthropic.claude-opus-4-20250514-v1:0",
+	// Claude Sonnet
+	"claude-sonnet-4-6-thinking": "us.anthropic.claude-sonnet-4-6",
+	"claude-sonnet-4-6":          "us.anthropic.claude-sonnet-4-6",
+	"claude-sonnet-4-5":          "us.anthropic.claude-sonnet-4-5-20250929-v1:0",
+	"claude-sonnet-4-5-thinking": "us.anthropic.claude-sonnet-4-5-20250929-v1:0",
+	"claude-sonnet-4-5-20250929": "us.anthropic.claude-sonnet-4-5-20250929-v1:0",
+	"claude-sonnet-4-20250514":   "us.anthropic.claude-sonnet-4-20250514-v1:0",
+	// Claude Haiku
+	"claude-haiku-4-5":          "us.anthropic.claude-haiku-4-5-20251001-v1:0",
+	"claude-haiku-4-5-20251001": "us.anthropic.claude-haiku-4-5-20251001-v1:0",
+}

backend/internal/domain/constants_test.go

@@ -6,6 +6,8 @@ func TestDefaultAntigravityModelMapping_ImageCompatibilityAliases(t *testing.T)
 	t.Parallel()
 
 	cases := map[string]string{
+		"gemini-2.5-flash-image":         "gemini-2.5-flash-image",
+		"gemini-2.5-flash-image-preview": "gemini-2.5-flash-image",
 		"gemini-3.1-flash-image":         "gemini-3.1-flash-image",
 		"gemini-3.1-flash-image-preview": "gemini-3.1-flash-image",
 		"gemini-3-pro-image":             "gemini-3.1-flash-image",

backend/internal/handler/admin/account_data.go

@@ -8,6 +8,9 @@ import (
 	"strings"
 	"time"
 
+	"log/slog"
+
+	"github.com/Wei-Shaw/sub2api/internal/pkg/openai"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/response"
 	"github.com/Wei-Shaw/sub2api/internal/service"
 	"github.com/gin-gonic/gin"
@@ -292,6 +295,8 @@ func (h *AccountHandler) importData(ctx context.Context, req DataImportRequest)
 			}
 		}
 
+		enrichCredentialsFromIDToken(&item)
+
 		accountInput := &service.CreateAccountInput{
 			Name:                 item.Name,
 			Notes:                item.Notes,
@@ -535,6 +540,57 @@ func defaultProxyName(name string) string {
 	return name
 }
 
+// enrichCredentialsFromIDToken performs best-effort extraction of user info fields
+// (email, plan_type, chatgpt_account_id, etc.) from id_token in credentials.
+// Only applies to OpenAI/Sora OAuth accounts. Skips expired token errors silently.
+// Existing credential values are never overwritten — only missing fields are filled.
+func enrichCredentialsFromIDToken(item *DataAccount) {
+	if item.Credentials == nil {
+		return
+	}
+	// Only enrich OpenAI/Sora OAuth accounts
+	platform := strings.ToLower(strings.TrimSpace(item.Platform))
+	if platform != service.PlatformOpenAI && platform != service.PlatformSora {
+		return
+	}
+	if strings.ToLower(strings.TrimSpace(item.Type)) != service.AccountTypeOAuth {
+		return
+	}
+
+	idToken, _ := item.Credentials["id_token"].(string)
+	if strings.TrimSpace(idToken) == "" {
+		return
+	}
+
+	// DecodeIDToken skips expiry validation — safe for imported data
+	claims, err := openai.DecodeIDToken(idToken)
+	if err != nil {
+		slog.Debug("import_enrich_id_token_decode_failed", "account", item.Name, "error", err)
+		return
+	}
+
+	userInfo := claims.GetUserInfo()
+	if userInfo == nil {
+		return
+	}
+
+	// Fill missing fields only (never overwrite existing values)
+	setIfMissing := func(key, value string) {
+		if value == "" {
+			return
+		}
+		if existing, _ := item.Credentials[key].(string); existing == "" {
+			item.Credentials[key] = value
+		}
+	}
+
+	setIfMissing("email", userInfo.Email)
+	setIfMissing("plan_type", userInfo.PlanType)
+	setIfMissing("chatgpt_account_id", userInfo.ChatGPTAccountID)
+	setIfMissing("chatgpt_user_id", userInfo.ChatGPTUserID)
+	setIfMissing("organization_id", userInfo.OrganizationID)
+}
+
 func normalizeProxyStatus(status string) string {
 	normalized := strings.TrimSpace(strings.ToLower(status))
 	switch normalized {

backend/internal/handler/admin/account_handler.go

@@ -8,6 +8,7 @@ import (
 	"encoding/json"
 	"errors"
 	"fmt"
+	"log"
 	"net/http"
 	"strconv"
 	"strings"
@@ -18,6 +19,7 @@ import (
 	"github.com/Wei-Shaw/sub2api/internal/handler/dto"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/antigravity"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/claude"
+	infraerrors "github.com/Wei-Shaw/sub2api/internal/pkg/errors"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/geminicli"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/openai"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/response"
@@ -95,13 +97,14 @@ type CreateAccountRequest struct {
 	Name                    string         `json:"name" binding:"required"`
 	Notes                   *string        `json:"notes"`
 	Platform                string         `json:"platform" binding:"required"`
-	Type                    string         `json:"type" binding:"required,oneof=oauth setup-token apikey upstream"`
+	Type                    string         `json:"type" binding:"required,oneof=oauth setup-token apikey upstream bedrock"`
 	Credentials             map[string]any `json:"credentials" binding:"required"`
 	Extra                   map[string]any `json:"extra"`
 	ProxyID                 *int64         `json:"proxy_id"`
 	Concurrency             int            `json:"concurrency"`
 	Priority                int            `json:"priority"`
 	RateMultiplier          *float64       `json:"rate_multiplier"`
+	LoadFactor              *int           `json:"load_factor"`
 	GroupIDs                []int64        `json:"group_ids"`
 	ExpiresAt               *int64         `json:"expires_at"`
 	AutoPauseOnExpired      *bool          `json:"auto_pause_on_expired"`
@@ -113,14 +116,15 @@ type CreateAccountRequest struct {
 type UpdateAccountRequest struct {
 	Name                    string         `json:"name"`
 	Notes                   *string        `json:"notes"`
-	Type                    string         `json:"type" binding:"omitempty,oneof=oauth setup-token apikey upstream"`
+	Type                    string         `json:"type" binding:"omitempty,oneof=oauth setup-token apikey upstream bedrock"`
 	Credentials             map[string]any `json:"credentials"`
 	Extra                   map[string]any `json:"extra"`
 	ProxyID                 *int64         `json:"proxy_id"`
 	Concurrency             *int           `json:"concurrency"`
 	Priority                *int           `json:"priority"`
 	RateMultiplier          *float64       `json:"rate_multiplier"`
-	Status                  string         `json:"status" binding:"omitempty,oneof=active inactive"`
+	LoadFactor              *int           `json:"load_factor"`
+	Status                  string         `json:"status" binding:"omitempty,oneof=active inactive error"`
 	GroupIDs                *[]int64       `json:"group_ids"`
 	ExpiresAt               *int64         `json:"expires_at"`
 	AutoPauseOnExpired      *bool          `json:"auto_pause_on_expired"`
@@ -135,6 +139,7 @@ type BulkUpdateAccountsRequest struct {
 	Concurrency             *int           `json:"concurrency"`
 	Priority                *int           `json:"priority"`
 	RateMultiplier          *float64       `json:"rate_multiplier"`
+	LoadFactor              *int           `json:"load_factor"`
 	Status                  string         `json:"status" binding:"omitempty,oneof=active inactive error"`
 	Schedulable             *bool          `json:"schedulable"`
 	GroupIDs                *[]int64       `json:"group_ids"`
@@ -285,8 +290,8 @@ func (h *AccountHandler) List(c *gin.Context) {
 		}
 	}
 
-	// 仅非 lite 模式获取窗口费用（PostgreSQL 聚合查询，高开销）
-	if !lite && len(windowCostAccountIDs) > 0 {
+	// 始终获取窗口费用（PostgreSQL 聚合查询）
+	if len(windowCostAccountIDs) > 0 {
 		windowCosts = make(map[int64]float64)
 		var mu sync.Mutex
 		g, gctx := errgroup.WithContext(c.Request.Context())
@@ -506,6 +511,7 @@ func (h *AccountHandler) Create(c *gin.Context) {
 			Concurrency:           req.Concurrency,
 			Priority:              req.Priority,
 			RateMultiplier:        req.RateMultiplier,
+			LoadFactor:            req.LoadFactor,
 			GroupIDs:              req.GroupIDs,
 			ExpiresAt:             req.ExpiresAt,
 			AutoPauseOnExpired:    req.AutoPauseOnExpired,
@@ -575,6 +581,7 @@ func (h *AccountHandler) Update(c *gin.Context) {
 		Concurrency:           req.Concurrency, // 指针类型，nil 表示未提供
 		Priority:              req.Priority,    // 指针类型，nil 表示未提供
 		RateMultiplier:        req.RateMultiplier,
+		LoadFactor:            req.LoadFactor,
 		Status:                req.Status,
 		GroupIDs:              req.GroupIDs,
 		ExpiresAt:             req.ExpiresAt,
@@ -621,6 +628,7 @@ func (h *AccountHandler) Delete(c *gin.Context) {
 // TestAccountRequest represents the request body for testing an account
 type TestAccountRequest struct {
 	ModelID string `json:"model_id"`
+	Prompt  string `json:"prompt"`
 }
 
 type SyncFromCRSRequest struct {
@@ -651,10 +659,46 @@ func (h *AccountHandler) Test(c *gin.Context) {
 	_ = c.ShouldBindJSON(&req)
 
 	// Use AccountTestService to test the account with SSE streaming
-	if err := h.accountTestService.TestAccountConnection(c, accountID, req.ModelID); err != nil {
+	if err := h.accountTestService.TestAccountConnection(c, accountID, req.ModelID, req.Prompt); err != nil {
 		// Error already sent via SSE, just log
 		return
 	}
+
+	if h.rateLimitService != nil {
+		if _, err := h.rateLimitService.RecoverAccountAfterSuccessfulTest(c.Request.Context(), accountID); err != nil {
+			_ = c.Error(err)
+		}
+	}
+}
+
+// RecoverState handles unified recovery of recoverable account runtime state.
+// POST /api/v1/admin/accounts/:id/recover-state
+func (h *AccountHandler) RecoverState(c *gin.Context) {
+	accountID, err := strconv.ParseInt(c.Param("id"), 10, 64)
+	if err != nil {
+		response.BadRequest(c, "Invalid account ID")
+		return
+	}
+
+	if h.rateLimitService == nil {
+		response.Error(c, http.StatusServiceUnavailable, "Rate limit service unavailable")
+		return
+	}
+
+	if _, err := h.rateLimitService.RecoverAccountState(c.Request.Context(), accountID, service.AccountRecoveryOptions{
+		InvalidateToken: true,
+	}); err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	account, err := h.adminService.GetAccount(c.Request.Context(), accountID)
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	response.Success(c, h.buildAccountResponseWithRuntime(c.Request.Context(), account))
 }
 
 // SyncFromCRS handles syncing accounts from claude-relay-service (CRS)
@@ -710,52 +754,31 @@ func (h *AccountHandler) PreviewFromCRS(c *gin.Context) {
 	response.Success(c, result)
 }
 
-// Refresh handles refreshing account credentials
-// POST /api/v1/admin/accounts/:id/refresh
-func (h *AccountHandler) Refresh(c *gin.Context) {
-	accountID, err := strconv.ParseInt(c.Param("id"), 10, 64)
-	if err != nil {
-		response.BadRequest(c, "Invalid account ID")
-		return
-	}
-
-	// Get account
-	account, err := h.adminService.GetAccount(c.Request.Context(), accountID)
-	if err != nil {
-		response.NotFound(c, "Account not found")
-		return
-	}
-
-	// Only refresh OAuth-based accounts (oauth and setup-token)
+// refreshSingleAccount refreshes credentials for a single OAuth account.
+// Returns (updatedAccount, warning, error) where warning is used for Antigravity ProjectIDMissing scenario.
+func (h *AccountHandler) refreshSingleAccount(ctx context.Context, account *service.Account) (*service.Account, string, error) {
 	if !account.IsOAuth() {
-		response.BadRequest(c, "Cannot refresh non-OAuth account credentials")
-		return
+		return nil, "", infraerrors.BadRequest("NOT_OAUTH", "cannot refresh non-OAuth account")
 	}
 
 	var newCredentials map[string]any
 
 	if account.IsOpenAI() {
-		// Use OpenAI OAuth service to refresh token
-		tokenInfo, err := h.openaiOAuthService.RefreshAccountToken(c.Request.Context(), account)
+		tokenInfo, err := h.openaiOAuthService.RefreshAccountToken(ctx, account)
 		if err != nil {
-			response.ErrorFrom(c, err)
-			return
+			return nil, "", err
 		}
 
-		// Build new credentials from token info
 		newCredentials = h.openaiOAuthService.BuildAccountCredentials(tokenInfo)
-
-		// Preserve non-token settings from existing credentials
 		for k, v := range account.Credentials {
 			if _, exists := newCredentials[k]; !exists {
 				newCredentials[k] = v
 			}
 		}
 	} else if account.Platform == service.PlatformGemini {
-		tokenInfo, err := h.geminiOAuthService.RefreshAccountToken(c.Request.Context(), account)
+		tokenInfo, err := h.geminiOAuthService.RefreshAccountToken(ctx, account)
 		if err != nil {
-			response.InternalError(c, "Failed to refresh credentials: "+err.Error())
-			return
+			return nil, "", fmt.Errorf("failed to refresh credentials: %w", err)
 		}
 
 		newCredentials = h.geminiOAuthService.BuildAccountCredentials(tokenInfo)
@@ -765,10 +788,9 @@ func (h *AccountHandler) Refresh(c *gin.Context) {
 			}
 		}
 	} else if account.Platform == service.PlatformAntigravity {
-		tokenInfo, err := h.antigravityOAuthService.RefreshAccountToken(c.Request.Context(), account)
+		tokenInfo, err := h.antigravityOAuthService.RefreshAccountToken(ctx, account)
 		if err != nil {
-			response.ErrorFrom(c, err)
-			return
+			return nil, "", err
 		}
 
 		newCredentials = h.antigravityOAuthService.BuildAccountCredentials(tokenInfo)
@@ -787,37 +809,27 @@ func (h *AccountHandler) Refresh(c *gin.Context) {
 		}
 
 		// 如果 project_id 获取失败，更新凭证但不标记为 error
-		// LoadCodeAssist 失败可能是临时网络问题，给它机会在下次自动刷新时重试
 		if tokenInfo.ProjectIDMissing {
-			// 先更新凭证（token 本身刷新成功了）
-			_, updateErr := h.adminService.UpdateAccount(c.Request.Context(), accountID, &service.UpdateAccountInput{
+			updatedAccount, updateErr := h.adminService.UpdateAccount(ctx, account.ID, &service.UpdateAccountInput{
 				Credentials: newCredentials,
 			})
 			if updateErr != nil {
-				response.InternalError(c, "Failed to update credentials: "+updateErr.Error())
-				return
+				return nil, "", fmt.Errorf("failed to update credentials: %w", updateErr)
 			}
-			// 不标记为 error，只返回警告信息
-			response.Success(c, gin.H{
-				"message": "Token refreshed successfully, but project_id could not be retrieved (will retry automatically)",
-				"warning": "missing_project_id_temporary",
-			})
-			return
+			return updatedAccount, "missing_project_id_temporary", nil
 		}
 
 		// 成功获取到 project_id，如果之前是 missing_project_id 错误则清除
 		if account.Status == service.StatusError && strings.Contains(account.ErrorMessage, "missing_project_id:") {
-			if _, clearErr := h.adminService.ClearAccountError(c.Request.Context(), accountID); clearErr != nil {
-				response.InternalError(c, "Failed to clear account error: "+clearErr.Error())
-				return
+			if _, clearErr := h.adminService.ClearAccountError(ctx, account.ID); clearErr != nil {
+				return nil, "", fmt.Errorf("failed to clear account error: %w", clearErr)
 			}
 		}
 	} else {
 		// Use Anthropic/Claude OAuth service to refresh token
-		tokenInfo, err := h.oauthService.RefreshAccountToken(c.Request.Context(), account)
+		tokenInfo, err := h.oauthService.RefreshAccountToken(ctx, account)
 		if err != nil {
-			response.ErrorFrom(c, err)
-			return
+			return nil, "", err
 		}
 
 		// Copy existing credentials to preserve non-token settings (e.g., intercept_warmup_requests)
@@ -839,20 +851,54 @@ func (h *AccountHandler) Refresh(c *gin.Context) {
 		}
 	}
 
-	updatedAccount, err := h.adminService.UpdateAccount(c.Request.Context(), accountID, &service.UpdateAccountInput{
+	updatedAccount, err := h.adminService.UpdateAccount(ctx, account.ID, &service.UpdateAccountInput{
 		Credentials: newCredentials,
 	})
+	if err != nil {
+		return nil, "", err
+	}
+
+	// 刷新成功后，清除 token 缓存，确保下次请求使用新 token
+	if h.tokenCacheInvalidator != nil {
+		if invalidateErr := h.tokenCacheInvalidator.InvalidateToken(ctx, updatedAccount); invalidateErr != nil {
+			log.Printf("[WARN] Failed to invalidate token cache for account %d: %v", updatedAccount.ID, invalidateErr)
+		}
+	}
+
+	// OpenAI OAuth: 刷新成功后检查并设置 privacy_mode
+	h.adminService.EnsureOpenAIPrivacy(ctx, updatedAccount)
+
+	return updatedAccount, "", nil
+}
+
+// Refresh handles refreshing account credentials
+// POST /api/v1/admin/accounts/:id/refresh
+func (h *AccountHandler) Refresh(c *gin.Context) {
+	accountID, err := strconv.ParseInt(c.Param("id"), 10, 64)
+	if err != nil {
+		response.BadRequest(c, "Invalid account ID")
+		return
+	}
+
+	// Get account
+	account, err := h.adminService.GetAccount(c.Request.Context(), accountID)
+	if err != nil {
+		response.NotFound(c, "Account not found")
+		return
+	}
+
+	updatedAccount, warning, err := h.refreshSingleAccount(c.Request.Context(), account)
 	if err != nil {
 		response.ErrorFrom(c, err)
 		return
 	}
 
-	// 刷新成功后，清除 token 缓存，确保下次请求使用新 token
-	if h.tokenCacheInvalidator != nil {
-		if invalidateErr := h.tokenCacheInvalidator.InvalidateToken(c.Request.Context(), updatedAccount); invalidateErr != nil {
-			// 缓存失效失败只记录日志，不影响主流程
-			_ = c.Error(invalidateErr)
-		}
+	if warning == "missing_project_id_temporary" {
+		response.Success(c, gin.H{
+			"message": "Token refreshed successfully, but project_id could not be retrieved (will retry automatically)",
+			"warning": "missing_project_id_temporary",
+		})
+		return
 	}
 
 	response.Success(c, h.buildAccountResponseWithRuntime(c.Request.Context(), updatedAccount))
@@ -908,14 +954,175 @@ func (h *AccountHandler) ClearError(c *gin.Context) {
 	// 这解决了管理员重置账号状态后，旧的失效 token 仍在缓存中导致立即再次 401 的问题
 	if h.tokenCacheInvalidator != nil && account.IsOAuth() {
 		if invalidateErr := h.tokenCacheInvalidator.InvalidateToken(c.Request.Context(), account); invalidateErr != nil {
-			// 缓存失效失败只记录日志，不影响主流程
-			_ = c.Error(invalidateErr)
+			log.Printf("[WARN] Failed to invalidate token cache for account %d: %v", accountID, invalidateErr)
 		}
 	}
 
 	response.Success(c, h.buildAccountResponseWithRuntime(c.Request.Context(), account))
 }
 
+// BatchClearError handles batch clearing account errors
+// POST /api/v1/admin/accounts/batch-clear-error
+func (h *AccountHandler) BatchClearError(c *gin.Context) {
+	var req struct {
+		AccountIDs []int64 `json:"account_ids"`
+	}
+	if err := c.ShouldBindJSON(&req); err != nil {
+		response.BadRequest(c, "Invalid request: "+err.Error())
+		return
+	}
+	if len(req.AccountIDs) == 0 {
+		response.BadRequest(c, "account_ids is required")
+		return
+	}
+
+	ctx := c.Request.Context()
+
+	const maxConcurrency = 10
+	g, gctx := errgroup.WithContext(ctx)
+	g.SetLimit(maxConcurrency)
+
+	var mu sync.Mutex
+	var successCount, failedCount int
+	var errors []gin.H
+
+	// 注意：所有 goroutine 必须 return nil，避免 errgroup cancel 其他并发任务
+	for _, id := range req.AccountIDs {
+		accountID := id // 闭包捕获
+		g.Go(func() error {
+			account, err := h.adminService.ClearAccountError(gctx, accountID)
+			if err != nil {
+				mu.Lock()
+				failedCount++
+				errors = append(errors, gin.H{
+					"account_id": accountID,
+					"error":      err.Error(),
+				})
+				mu.Unlock()
+				return nil
+			}
+
+			// 清除错误后，同时清除 token 缓存
+			if h.tokenCacheInvalidator != nil && account.IsOAuth() {
+				if invalidateErr := h.tokenCacheInvalidator.InvalidateToken(gctx, account); invalidateErr != nil {
+					log.Printf("[WARN] Failed to invalidate token cache for account %d: %v", accountID, invalidateErr)
+				}
+			}
+
+			mu.Lock()
+			successCount++
+			mu.Unlock()
+			return nil
+		})
+	}
+
+	if err := g.Wait(); err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	response.Success(c, gin.H{
+		"total":   len(req.AccountIDs),
+		"success": successCount,
+		"failed":  failedCount,
+		"errors":  errors,
+	})
+}
+
+// BatchRefresh handles batch refreshing account credentials
+// POST /api/v1/admin/accounts/batch-refresh
+func (h *AccountHandler) BatchRefresh(c *gin.Context) {
+	var req struct {
+		AccountIDs []int64 `json:"account_ids"`
+	}
+	if err := c.ShouldBindJSON(&req); err != nil {
+		response.BadRequest(c, "Invalid request: "+err.Error())
+		return
+	}
+	if len(req.AccountIDs) == 0 {
+		response.BadRequest(c, "account_ids is required")
+		return
+	}
+
+	ctx := c.Request.Context()
+
+	accounts, err := h.adminService.GetAccountsByIDs(ctx, req.AccountIDs)
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	// 建立已获取账号的 ID 集合，检测缺失的 ID
+	foundIDs := make(map[int64]bool, len(accounts))
+	for _, acc := range accounts {
+		if acc != nil {
+			foundIDs[acc.ID] = true
+		}
+	}
+
+	const maxConcurrency = 10
+	g, gctx := errgroup.WithContext(ctx)
+	g.SetLimit(maxConcurrency)
+
+	var mu sync.Mutex
+	var successCount, failedCount int
+	var errors []gin.H
+	var warnings []gin.H
+
+	// 将不存在的账号 ID 标记为失败
+	for _, id := range req.AccountIDs {
+		if !foundIDs[id] {
+			failedCount++
+			errors = append(errors, gin.H{
+				"account_id": id,
+				"error":      "account not found",
+			})
+		}
+	}
+
+	// 注意：所有 goroutine 必须 return nil，避免 errgroup cancel 其他并发任务
+	for _, account := range accounts {
+		acc := account // 闭包捕获
+		if acc == nil {
+			continue
+		}
+		g.Go(func() error {
+			_, warning, err := h.refreshSingleAccount(gctx, acc)
+			mu.Lock()
+			if err != nil {
+				failedCount++
+				errors = append(errors, gin.H{
+					"account_id": acc.ID,
+					"error":      err.Error(),
+				})
+			} else {
+				successCount++
+				if warning != "" {
+					warnings = append(warnings, gin.H{
+						"account_id": acc.ID,
+						"warning":    warning,
+					})
+				}
+			}
+			mu.Unlock()
+			return nil
+		})
+	}
+
+	if err := g.Wait(); err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	response.Success(c, gin.H{
+		"total":    len(req.AccountIDs),
+		"success":  successCount,
+		"failed":   failedCount,
+		"errors":   errors,
+		"warnings": warnings,
+	})
+}
+
 // BatchCreate handles batch creating accounts
 // POST /api/v1/admin/accounts/batch
 func (h *AccountHandler) BatchCreate(c *gin.Context) {
@@ -1101,6 +1308,7 @@ func (h *AccountHandler) BulkUpdate(c *gin.Context) {
 		req.Concurrency != nil ||
 		req.Priority != nil ||
 		req.RateMultiplier != nil ||
+		req.LoadFactor != nil ||
 		req.Status != "" ||
 		req.Schedulable != nil ||
 		req.GroupIDs != nil ||
@@ -1119,6 +1327,7 @@ func (h *AccountHandler) BulkUpdate(c *gin.Context) {
 		Concurrency:           req.Concurrency,
 		Priority:              req.Priority,
 		RateMultiplier:        req.RateMultiplier,
+		LoadFactor:            req.LoadFactor,
 		Status:                req.Status,
 		Schedulable:           req.Schedulable,
 		GroupIDs:              req.GroupIDs,
@@ -1328,6 +1537,29 @@ func (h *AccountHandler) ClearRateLimit(c *gin.Context) {
 	response.Success(c, h.buildAccountResponseWithRuntime(c.Request.Context(), account))
 }
 
+// ResetQuota handles resetting account quota usage
+// POST /api/v1/admin/accounts/:id/reset-quota
+func (h *AccountHandler) ResetQuota(c *gin.Context) {
+	accountID, err := strconv.ParseInt(c.Param("id"), 10, 64)
+	if err != nil {
+		response.BadRequest(c, "Invalid account ID")
+		return
+	}
+
+	if err := h.adminService.ResetAccountQuota(c.Request.Context(), accountID); err != nil {
+		response.InternalError(c, "Failed to reset account quota: "+err.Error())
+		return
+	}
+
+	account, err := h.adminService.GetAccount(c.Request.Context(), accountID)
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	response.Success(c, h.buildAccountResponseWithRuntime(c.Request.Context(), account))
+}
+
 // GetTempUnschedulable handles getting temporary unschedulable status
 // GET /api/v1/admin/accounts/:id/temp-unschedulable
 func (h *AccountHandler) GetTempUnschedulable(c *gin.Context) {
@@ -1486,13 +1718,12 @@ func (h *AccountHandler) GetAvailableModels(c *gin.Context) {
 
 	// Handle OpenAI accounts
 	if account.IsOpenAI() {
-		// For OAuth accounts: return default OpenAI models
-		if account.IsOAuth() {
+		// OpenAI 自动透传会绕过常规模型改写，测试/模型列表也应回落到默认模型集。
+		if account.IsOpenAIPassthroughEnabled() {
 			response.Success(c, openai.DefaultModels)
 			return
 		}
 
-		// For API Key accounts: check model_mapping
 		mapping := account.GetModelMapping()
 		if len(mapping) == 0 {
 			response.Success(c, openai.DefaultModels)

backend/internal/handler/admin/account_handler_available_models_test.go

@@ -0,0 +1,105 @@
+package admin
+
+import (
+	"context"
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"testing"
+
+	"github.com/Wei-Shaw/sub2api/internal/service"
+	"github.com/gin-gonic/gin"
+	"github.com/stretchr/testify/require"
+)
+
+type availableModelsAdminService struct {
+	*stubAdminService
+	account service.Account
+}
+
+func (s *availableModelsAdminService) GetAccount(_ context.Context, id int64) (*service.Account, error) {
+	if s.account.ID == id {
+		acc := s.account
+		return &acc, nil
+	}
+	return s.stubAdminService.GetAccount(context.Background(), id)
+}
+
+func setupAvailableModelsRouter(adminSvc service.AdminService) *gin.Engine {
+	gin.SetMode(gin.TestMode)
+	router := gin.New()
+	handler := NewAccountHandler(adminSvc, nil, nil, nil, nil, nil, nil, nil, nil, nil, nil, nil, nil)
+	router.GET("/api/v1/admin/accounts/:id/models", handler.GetAvailableModels)
+	return router
+}
+
+func TestAccountHandlerGetAvailableModels_OpenAIOAuthUsesExplicitModelMapping(t *testing.T) {
+	svc := &availableModelsAdminService{
+		stubAdminService: newStubAdminService(),
+		account: service.Account{
+			ID:       42,
+			Name:     "openai-oauth",
+			Platform: service.PlatformOpenAI,
+			Type:     service.AccountTypeOAuth,
+			Status:   service.StatusActive,
+			Credentials: map[string]any{
+				"model_mapping": map[string]any{
+					"gpt-5": "gpt-5.1",
+				},
+			},
+		},
+	}
+	router := setupAvailableModelsRouter(svc)
+
+	rec := httptest.NewRecorder()
+	req := httptest.NewRequest(http.MethodGet, "/api/v1/admin/accounts/42/models", nil)
+	router.ServeHTTP(rec, req)
+
+	require.Equal(t, http.StatusOK, rec.Code)
+
+	var resp struct {
+		Data []struct {
+			ID string `json:"id"`
+		} `json:"data"`
+	}
+	require.NoError(t, json.Unmarshal(rec.Body.Bytes(), &resp))
+	require.Len(t, resp.Data, 1)
+	require.Equal(t, "gpt-5", resp.Data[0].ID)
+}
+
+func TestAccountHandlerGetAvailableModels_OpenAIOAuthPassthroughFallsBackToDefaults(t *testing.T) {
+	svc := &availableModelsAdminService{
+		stubAdminService: newStubAdminService(),
+		account: service.Account{
+			ID:       43,
+			Name:     "openai-oauth-passthrough",
+			Platform: service.PlatformOpenAI,
+			Type:     service.AccountTypeOAuth,
+			Status:   service.StatusActive,
+			Credentials: map[string]any{
+				"model_mapping": map[string]any{
+					"gpt-5": "gpt-5.1",
+				},
+			},
+			Extra: map[string]any{
+				"openai_passthrough": true,
+			},
+		},
+	}
+	router := setupAvailableModelsRouter(svc)
+
+	rec := httptest.NewRecorder()
+	req := httptest.NewRequest(http.MethodGet, "/api/v1/admin/accounts/43/models", nil)
+	router.ServeHTTP(rec, req)
+
+	require.Equal(t, http.StatusOK, rec.Code)
+
+	var resp struct {
+		Data []struct {
+			ID string `json:"id"`
+		} `json:"data"`
+	}
+	require.NoError(t, json.Unmarshal(rec.Body.Bytes(), &resp))
+	require.NotEmpty(t, resp.Data)
+	require.NotEqual(t, "gpt-5", resp.Data[0].ID)
+}

backend/internal/handler/admin/admin_service_stub_test.go

@@ -175,6 +175,18 @@ func (s *stubAdminService) GetGroupAPIKeys(ctx context.Context, groupID int64, p
 	return s.apiKeys, int64(len(s.apiKeys)), nil
 }
 
+func (s *stubAdminService) GetGroupRateMultipliers(_ context.Context, _ int64) ([]service.UserGroupRateEntry, error) {
+	return nil, nil
+}
+
+func (s *stubAdminService) ClearGroupRateMultipliers(_ context.Context, _ int64) error {
+	return nil
+}
+
+func (s *stubAdminService) BatchSetGroupRateMultipliers(_ context.Context, _ int64, _ []service.GroupRateMultiplierInput) error {
+	return nil
+}
+
 func (s *stubAdminService) ListAccounts(ctx context.Context, page, pageSize int, platform, accountType, status, search string, groupID int64) ([]service.Account, int64, error) {
 	return s.accounts, int64(len(s.accounts)), nil
 }
@@ -425,5 +437,13 @@ func (s *stubAdminService) AdminUpdateAPIKeyGroupID(ctx context.Context, keyID i
 	return nil, service.ErrAPIKeyNotFound
 }
 
+func (s *stubAdminService) ResetAccountQuota(ctx context.Context, id int64) error {
+	return nil
+}
+
+func (s *stubAdminService) EnsureOpenAIPrivacy(ctx context.Context, account *service.Account) string {
+	return ""
+}
+
 // Ensure stub implements interface.
 var _ service.AdminService = (*stubAdminService)(nil)

backend/internal/handler/admin/announcement_handler.go

@@ -27,21 +27,23 @@ func NewAnnouncementHandler(announcementService *service.AnnouncementService) *A
 }
 
 type CreateAnnouncementRequest struct {
-	Title     string                        `json:"title" binding:"required"`
-	Content   string                        `json:"content" binding:"required"`
-	Status    string                        `json:"status" binding:"omitempty,oneof=draft active archived"`
-	Targeting service.AnnouncementTargeting `json:"targeting"`
-	StartsAt  *int64                        `json:"starts_at"` // Unix seconds, 0/empty = immediate
-	EndsAt    *int64                        `json:"ends_at"`   // Unix seconds, 0/empty = never
+	Title      string                        `json:"title" binding:"required"`
+	Content    string                        `json:"content" binding:"required"`
+	Status     string                        `json:"status" binding:"omitempty,oneof=draft active archived"`
+	NotifyMode string                        `json:"notify_mode" binding:"omitempty,oneof=silent popup"`
+	Targeting  service.AnnouncementTargeting `json:"targeting"`
+	StartsAt   *int64                        `json:"starts_at"` // Unix seconds, 0/empty = immediate
+	EndsAt     *int64                        `json:"ends_at"`   // Unix seconds, 0/empty = never
 }
 
 type UpdateAnnouncementRequest struct {
-	Title     *string                        `json:"title"`
-	Content   *string                        `json:"content"`
-	Status    *string                        `json:"status" binding:"omitempty,oneof=draft active archived"`
-	Targeting *service.AnnouncementTargeting `json:"targeting"`
-	StartsAt  *int64                         `json:"starts_at"` // Unix seconds, 0 = clear
-	EndsAt    *int64                         `json:"ends_at"`   // Unix seconds, 0 = clear
+	Title      *string                        `json:"title"`
+	Content    *string                        `json:"content"`
+	Status     *string                        `json:"status" binding:"omitempty,oneof=draft active archived"`
+	NotifyMode *string                        `json:"notify_mode" binding:"omitempty,oneof=silent popup"`
+	Targeting  *service.AnnouncementTargeting `json:"targeting"`
+	StartsAt   *int64                         `json:"starts_at"` // Unix seconds, 0 = clear
+	EndsAt     *int64                         `json:"ends_at"`   // Unix seconds, 0 = clear
 }
 
 // List handles listing announcements with filters
@@ -110,11 +112,12 @@ func (h *AnnouncementHandler) Create(c *gin.Context) {
 	}
 
 	input := &service.CreateAnnouncementInput{
-		Title:     req.Title,
-		Content:   req.Content,
-		Status:    req.Status,
-		Targeting: req.Targeting,
-		ActorID:   &subject.UserID,
+		Title:      req.Title,
+		Content:    req.Content,
+		Status:     req.Status,
+		NotifyMode: req.NotifyMode,
+		Targeting:  req.Targeting,
+		ActorID:    &subject.UserID,
 	}
 
 	if req.StartsAt != nil && *req.StartsAt > 0 {
@@ -157,11 +160,12 @@ func (h *AnnouncementHandler) Update(c *gin.Context) {
 	}
 
 	input := &service.UpdateAnnouncementInput{
-		Title:     req.Title,
-		Content:   req.Content,
-		Status:    req.Status,
-		Targeting: req.Targeting,
-		ActorID:   &subject.UserID,
+		Title:      req.Title,
+		Content:    req.Content,
+		Status:     req.Status,
+		NotifyMode: req.NotifyMode,
+		Targeting:  req.Targeting,
+		ActorID:    &subject.UserID,
 	}
 
 	if req.StartsAt != nil {

backend/internal/handler/admin/backup_handler.go

@@ -0,0 +1,204 @@
+package admin
+
+import (
+	"github.com/Wei-Shaw/sub2api/internal/pkg/response"
+	"github.com/Wei-Shaw/sub2api/internal/server/middleware"
+	"github.com/Wei-Shaw/sub2api/internal/service"
+	"github.com/gin-gonic/gin"
+)
+
+type BackupHandler struct {
+	backupService *service.BackupService
+	userService   *service.UserService
+}
+
+func NewBackupHandler(backupService *service.BackupService, userService *service.UserService) *BackupHandler {
+	return &BackupHandler{
+		backupService: backupService,
+		userService:   userService,
+	}
+}
+
+// ─── S3 配置 ───
+
+func (h *BackupHandler) GetS3Config(c *gin.Context) {
+	cfg, err := h.backupService.GetS3Config(c.Request.Context())
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+	response.Success(c, cfg)
+}
+
+func (h *BackupHandler) UpdateS3Config(c *gin.Context) {
+	var req service.BackupS3Config
+	if err := c.ShouldBindJSON(&req); err != nil {
+		response.BadRequest(c, "Invalid request: "+err.Error())
+		return
+	}
+	cfg, err := h.backupService.UpdateS3Config(c.Request.Context(), req)
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+	response.Success(c, cfg)
+}
+
+func (h *BackupHandler) TestS3Connection(c *gin.Context) {
+	var req service.BackupS3Config
+	if err := c.ShouldBindJSON(&req); err != nil {
+		response.BadRequest(c, "Invalid request: "+err.Error())
+		return
+	}
+	err := h.backupService.TestS3Connection(c.Request.Context(), req)
+	if err != nil {
+		response.Success(c, gin.H{"ok": false, "message": err.Error()})
+		return
+	}
+	response.Success(c, gin.H{"ok": true, "message": "connection successful"})
+}
+
+// ─── 定时备份 ───
+
+func (h *BackupHandler) GetSchedule(c *gin.Context) {
+	cfg, err := h.backupService.GetSchedule(c.Request.Context())
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+	response.Success(c, cfg)
+}
+
+func (h *BackupHandler) UpdateSchedule(c *gin.Context) {
+	var req service.BackupScheduleConfig
+	if err := c.ShouldBindJSON(&req); err != nil {
+		response.BadRequest(c, "Invalid request: "+err.Error())
+		return
+	}
+	cfg, err := h.backupService.UpdateSchedule(c.Request.Context(), req)
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+	response.Success(c, cfg)
+}
+
+// ─── 备份操作 ───
+
+type CreateBackupRequest struct {
+	ExpireDays *int `json:"expire_days"` // nil=使用默认值14，0=永不过期
+}
+
+func (h *BackupHandler) CreateBackup(c *gin.Context) {
+	var req CreateBackupRequest
+	_ = c.ShouldBindJSON(&req) // 允许空 body
+
+	expireDays := 14 // 默认14天过期
+	if req.ExpireDays != nil {
+		expireDays = *req.ExpireDays
+	}
+
+	record, err := h.backupService.CreateBackup(c.Request.Context(), "manual", expireDays)
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+	response.Success(c, record)
+}
+
+func (h *BackupHandler) ListBackups(c *gin.Context) {
+	records, err := h.backupService.ListBackups(c.Request.Context())
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+	if records == nil {
+		records = []service.BackupRecord{}
+	}
+	response.Success(c, gin.H{"items": records})
+}
+
+func (h *BackupHandler) GetBackup(c *gin.Context) {
+	backupID := c.Param("id")
+	if backupID == "" {
+		response.BadRequest(c, "backup ID is required")
+		return
+	}
+	record, err := h.backupService.GetBackupRecord(c.Request.Context(), backupID)
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+	response.Success(c, record)
+}
+
+func (h *BackupHandler) DeleteBackup(c *gin.Context) {
+	backupID := c.Param("id")
+	if backupID == "" {
+		response.BadRequest(c, "backup ID is required")
+		return
+	}
+	if err := h.backupService.DeleteBackup(c.Request.Context(), backupID); err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+	response.Success(c, gin.H{"deleted": true})
+}
+
+func (h *BackupHandler) GetDownloadURL(c *gin.Context) {
+	backupID := c.Param("id")
+	if backupID == "" {
+		response.BadRequest(c, "backup ID is required")
+		return
+	}
+	url, err := h.backupService.GetBackupDownloadURL(c.Request.Context(), backupID)
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+	response.Success(c, gin.H{"url": url})
+}
+
+// ─── 恢复操作（需要重新输入管理员密码） ───
+
+type RestoreBackupRequest struct {
+	Password string `json:"password" binding:"required"`
+}
+
+func (h *BackupHandler) RestoreBackup(c *gin.Context) {
+	backupID := c.Param("id")
+	if backupID == "" {
+		response.BadRequest(c, "backup ID is required")
+		return
+	}
+
+	var req RestoreBackupRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		response.BadRequest(c, "password is required for restore operation")
+		return
+	}
+
+	// 从上下文获取当前管理员用户 ID
+	sub, ok := middleware.GetAuthSubjectFromContext(c)
+	if !ok {
+		response.Unauthorized(c, "unauthorized")
+		return
+	}
+
+	// 获取管理员用户并验证密码
+	user, err := h.userService.GetByID(c.Request.Context(), sub.UserID)
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+	if !user.CheckPassword(req.Password) {
+		response.BadRequest(c, "incorrect admin password")
+		return
+	}
+
+	if err := h.backupService.RestoreBackup(c.Request.Context(), backupID); err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+	response.Success(c, gin.H{"restored": true})
+}

backend/internal/handler/admin/dashboard_handler.go

@@ -249,11 +249,12 @@ func (h *DashboardHandler) GetUsageTrend(c *gin.Context) {
 		}
 	}
 
-	trend, err := h.dashboardService.GetUsageTrendWithFilters(c.Request.Context(), startTime, endTime, granularity, userID, apiKeyID, accountID, groupID, model, requestType, stream, billingType)
+	trend, hit, err := h.getUsageTrendCached(c.Request.Context(), startTime, endTime, granularity, userID, apiKeyID, accountID, groupID, model, requestType, stream, billingType)
 	if err != nil {
 		response.Error(c, 500, "Failed to get usage trend")
 		return
 	}
+	c.Header("X-Snapshot-Cache", cacheStatusValue(hit))
 
 	response.Success(c, gin.H{
 		"trend":       trend,
@@ -321,11 +322,12 @@ func (h *DashboardHandler) GetModelStats(c *gin.Context) {
 		}
 	}
 
-	stats, err := h.dashboardService.GetModelStatsWithFilters(c.Request.Context(), startTime, endTime, userID, apiKeyID, accountID, groupID, requestType, stream, billingType)
+	stats, hit, err := h.getModelStatsCached(c.Request.Context(), startTime, endTime, userID, apiKeyID, accountID, groupID, requestType, stream, billingType)
 	if err != nil {
 		response.Error(c, 500, "Failed to get model statistics")
 		return
 	}
+	c.Header("X-Snapshot-Cache", cacheStatusValue(hit))
 
 	response.Success(c, gin.H{
 		"models":     stats,
@@ -391,11 +393,12 @@ func (h *DashboardHandler) GetGroupStats(c *gin.Context) {
 		}
 	}
 
-	stats, err := h.dashboardService.GetGroupStatsWithFilters(c.Request.Context(), startTime, endTime, userID, apiKeyID, accountID, groupID, requestType, stream, billingType)
+	stats, hit, err := h.getGroupStatsCached(c.Request.Context(), startTime, endTime, userID, apiKeyID, accountID, groupID, requestType, stream, billingType)
 	if err != nil {
 		response.Error(c, 500, "Failed to get group statistics")
 		return
 	}
+	c.Header("X-Snapshot-Cache", cacheStatusValue(hit))
 
 	response.Success(c, gin.H{
 		"groups":     stats,
@@ -416,11 +419,12 @@ func (h *DashboardHandler) GetAPIKeyUsageTrend(c *gin.Context) {
 		limit = 5
 	}
 
-	trend, err := h.dashboardService.GetAPIKeyUsageTrend(c.Request.Context(), startTime, endTime, granularity, limit)
+	trend, hit, err := h.getAPIKeyUsageTrendCached(c.Request.Context(), startTime, endTime, granularity, limit)
 	if err != nil {
 		response.Error(c, 500, "Failed to get API key usage trend")
 		return
 	}
+	c.Header("X-Snapshot-Cache", cacheStatusValue(hit))
 
 	response.Success(c, gin.H{
 		"trend":       trend,
@@ -442,11 +446,12 @@ func (h *DashboardHandler) GetUserUsageTrend(c *gin.Context) {
 		limit = 12
 	}
 
-	trend, err := h.dashboardService.GetUserUsageTrend(c.Request.Context(), startTime, endTime, granularity, limit)
+	trend, hit, err := h.getUserUsageTrendCached(c.Request.Context(), startTime, endTime, granularity, limit)
 	if err != nil {
 		response.Error(c, 500, "Failed to get user usage trend")
 		return
 	}
+	c.Header("X-Snapshot-Cache", cacheStatusValue(hit))
 
 	response.Success(c, gin.H{
 		"trend":       trend,
@@ -461,9 +466,60 @@ type BatchUsersUsageRequest struct {
 	UserIDs []int64 `json:"user_ids" binding:"required"`
 }
 
+var dashboardUsersRankingCache = newSnapshotCache(5 * time.Minute)
 var dashboardBatchUsersUsageCache = newSnapshotCache(30 * time.Second)
 var dashboardBatchAPIKeysUsageCache = newSnapshotCache(30 * time.Second)
 
+func parseRankingLimit(raw string) int {
+	limit, err := strconv.Atoi(strings.TrimSpace(raw))
+	if err != nil || limit <= 0 {
+		return 12
+	}
+	if limit > 50 {
+		return 50
+	}
+	return limit
+}
+
+// GetUserSpendingRanking handles getting user spending ranking data.
+// GET /api/v1/admin/dashboard/users-ranking
+func (h *DashboardHandler) GetUserSpendingRanking(c *gin.Context) {
+	startTime, endTime := parseTimeRange(c)
+	limit := parseRankingLimit(c.DefaultQuery("limit", "12"))
+
+	keyRaw, _ := json.Marshal(struct {
+		Start string `json:"start"`
+		End   string `json:"end"`
+		Limit int    `json:"limit"`
+	}{
+		Start: startTime.UTC().Format(time.RFC3339),
+		End:   endTime.UTC().Format(time.RFC3339),
+		Limit: limit,
+	})
+	cacheKey := string(keyRaw)
+	if cached, ok := dashboardUsersRankingCache.Get(cacheKey); ok {
+		c.Header("X-Snapshot-Cache", "hit")
+		response.Success(c, cached.Payload)
+		return
+	}
+
+	ranking, err := h.dashboardService.GetUserSpendingRanking(c.Request.Context(), startTime, endTime, limit)
+	if err != nil {
+		response.Error(c, 500, "Failed to get user spending ranking")
+		return
+	}
+
+	payload := gin.H{
+		"ranking":           ranking.Ranking,
+		"total_actual_cost": ranking.TotalActualCost,
+		"start_date":        startTime.Format("2006-01-02"),
+		"end_date":          endTime.Add(-24 * time.Hour).Format("2006-01-02"),
+	}
+	dashboardUsersRankingCache.Set(cacheKey, payload)
+	c.Header("X-Snapshot-Cache", "miss")
+	response.Success(c, payload)
+}
+
 // GetBatchUsersUsage handles getting usage stats for multiple users
 // POST /api/v1/admin/dashboard/users-usage
 func (h *DashboardHandler) GetBatchUsersUsage(c *gin.Context) {

backend/internal/handler/admin/dashboard_handler_cache_test.go

@@ -0,0 +1,118 @@
+package admin
+
+import (
+	"context"
+	"net/http"
+	"net/http/httptest"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/Wei-Shaw/sub2api/internal/pkg/usagestats"
+	"github.com/Wei-Shaw/sub2api/internal/service"
+	"github.com/gin-gonic/gin"
+	"github.com/stretchr/testify/require"
+)
+
+type dashboardUsageRepoCacheProbe struct {
+	service.UsageLogRepository
+	trendCalls      atomic.Int32
+	usersTrendCalls atomic.Int32
+}
+
+func (r *dashboardUsageRepoCacheProbe) GetUsageTrendWithFilters(
+	ctx context.Context,
+	startTime, endTime time.Time,
+	granularity string,
+	userID, apiKeyID, accountID, groupID int64,
+	model string,
+	requestType *int16,
+	stream *bool,
+	billingType *int8,
+) ([]usagestats.TrendDataPoint, error) {
+	r.trendCalls.Add(1)
+	return []usagestats.TrendDataPoint{{
+		Date:        "2026-03-11",
+		Requests:    1,
+		TotalTokens: 2,
+		Cost:        3,
+		ActualCost:  4,
+	}}, nil
+}
+
+func (r *dashboardUsageRepoCacheProbe) GetUserUsageTrend(
+	ctx context.Context,
+	startTime, endTime time.Time,
+	granularity string,
+	limit int,
+) ([]usagestats.UserUsageTrendPoint, error) {
+	r.usersTrendCalls.Add(1)
+	return []usagestats.UserUsageTrendPoint{{
+		Date:       "2026-03-11",
+		UserID:     1,
+		Email:      "cache@test.dev",
+		Requests:   2,
+		Tokens:     20,
+		Cost:       2,
+		ActualCost: 1,
+	}}, nil
+}
+
+func resetDashboardReadCachesForTest() {
+	dashboardTrendCache = newSnapshotCache(30 * time.Second)
+	dashboardUsersTrendCache = newSnapshotCache(30 * time.Second)
+	dashboardAPIKeysTrendCache = newSnapshotCache(30 * time.Second)
+	dashboardModelStatsCache = newSnapshotCache(30 * time.Second)
+	dashboardGroupStatsCache = newSnapshotCache(30 * time.Second)
+	dashboardSnapshotV2Cache = newSnapshotCache(30 * time.Second)
+}
+
+func TestDashboardHandler_GetUsageTrend_UsesCache(t *testing.T) {
+	t.Cleanup(resetDashboardReadCachesForTest)
+	resetDashboardReadCachesForTest()
+
+	gin.SetMode(gin.TestMode)
+	repo := &dashboardUsageRepoCacheProbe{}
+	dashboardSvc := service.NewDashboardService(repo, nil, nil, nil)
+	handler := NewDashboardHandler(dashboardSvc, nil)
+	router := gin.New()
+	router.GET("/admin/dashboard/trend", handler.GetUsageTrend)
+
+	req1 := httptest.NewRequest(http.MethodGet, "/admin/dashboard/trend?start_date=2026-03-01&end_date=2026-03-07&granularity=day", nil)
+	rec1 := httptest.NewRecorder()
+	router.ServeHTTP(rec1, req1)
+	require.Equal(t, http.StatusOK, rec1.Code)
+	require.Equal(t, "miss", rec1.Header().Get("X-Snapshot-Cache"))
+
+	req2 := httptest.NewRequest(http.MethodGet, "/admin/dashboard/trend?start_date=2026-03-01&end_date=2026-03-07&granularity=day", nil)
+	rec2 := httptest.NewRecorder()
+	router.ServeHTTP(rec2, req2)
+	require.Equal(t, http.StatusOK, rec2.Code)
+	require.Equal(t, "hit", rec2.Header().Get("X-Snapshot-Cache"))
+	require.Equal(t, int32(1), repo.trendCalls.Load())
+}
+
+func TestDashboardHandler_GetUserUsageTrend_UsesCache(t *testing.T) {
+	t.Cleanup(resetDashboardReadCachesForTest)
+	resetDashboardReadCachesForTest()
+
+	gin.SetMode(gin.TestMode)
+	repo := &dashboardUsageRepoCacheProbe{}
+	dashboardSvc := service.NewDashboardService(repo, nil, nil, nil)
+	handler := NewDashboardHandler(dashboardSvc, nil)
+	router := gin.New()
+	router.GET("/admin/dashboard/users-trend", handler.GetUserUsageTrend)
+
+	req1 := httptest.NewRequest(http.MethodGet, "/admin/dashboard/users-trend?start_date=2026-03-01&end_date=2026-03-07&granularity=day&limit=8", nil)
+	rec1 := httptest.NewRecorder()
+	router.ServeHTTP(rec1, req1)
+	require.Equal(t, http.StatusOK, rec1.Code)
+	require.Equal(t, "miss", rec1.Header().Get("X-Snapshot-Cache"))
+
+	req2 := httptest.NewRequest(http.MethodGet, "/admin/dashboard/users-trend?start_date=2026-03-01&end_date=2026-03-07&granularity=day&limit=8", nil)
+	rec2 := httptest.NewRecorder()
+	router.ServeHTTP(rec2, req2)
+	require.Equal(t, http.StatusOK, rec2.Code)
+	require.Equal(t, "hit", rec2.Header().Get("X-Snapshot-Cache"))
+	require.Equal(t, int32(1), repo.usersTrendCalls.Load())
+}

backend/internal/handler/admin/dashboard_handler_request_type_test.go

@@ -19,6 +19,9 @@ type dashboardUsageRepoCapture struct {
 	trendStream      *bool
 	modelRequestType *int16
 	modelStream      *bool
+	rankingLimit     int
+	ranking          []usagestats.UserSpendingRankingItem
+	rankingTotal     float64
 }
 
 func (s *dashboardUsageRepoCapture) GetUsageTrendWithFilters(
@@ -49,6 +52,18 @@ func (s *dashboardUsageRepoCapture) GetModelStatsWithFilters(
 	return []usagestats.ModelStat{}, nil
 }
 
+func (s *dashboardUsageRepoCapture) GetUserSpendingRanking(
+	ctx context.Context,
+	startTime, endTime time.Time,
+	limit int,
+) (*usagestats.UserSpendingRankingResponse, error) {
+	s.rankingLimit = limit
+	return &usagestats.UserSpendingRankingResponse{
+		Ranking:         s.ranking,
+		TotalActualCost: s.rankingTotal,
+	}, nil
+}
+
 func newDashboardRequestTypeTestRouter(repo *dashboardUsageRepoCapture) *gin.Engine {
 	gin.SetMode(gin.TestMode)
 	dashboardSvc := service.NewDashboardService(repo, nil, nil, nil)
@@ -56,6 +71,7 @@ func newDashboardRequestTypeTestRouter(repo *dashboardUsageRepoCapture) *gin.Eng
 	router := gin.New()
 	router.GET("/admin/dashboard/trend", handler.GetUsageTrend)
 	router.GET("/admin/dashboard/models", handler.GetModelStats)
+	router.GET("/admin/dashboard/users-ranking", handler.GetUserSpendingRanking)
 	return router
 }
 
@@ -130,3 +146,30 @@ func TestDashboardModelStatsInvalidStream(t *testing.T) {
 
 	require.Equal(t, http.StatusBadRequest, rec.Code)
 }
+
+func TestDashboardUsersRankingLimitAndCache(t *testing.T) {
+	dashboardUsersRankingCache = newSnapshotCache(5 * time.Minute)
+	repo := &dashboardUsageRepoCapture{
+		ranking: []usagestats.UserSpendingRankingItem{
+			{UserID: 7, Email: "rank@example.com", ActualCost: 10.5, Requests: 3, Tokens: 300},
+		},
+		rankingTotal: 88.8,
+	}
+	router := newDashboardRequestTypeTestRouter(repo)
+
+	req := httptest.NewRequest(http.MethodGet, "/admin/dashboard/users-ranking?limit=100&start_date=2025-01-01&end_date=2025-01-02", nil)
+	rec := httptest.NewRecorder()
+	router.ServeHTTP(rec, req)
+
+	require.Equal(t, http.StatusOK, rec.Code)
+	require.Equal(t, 50, repo.rankingLimit)
+	require.Contains(t, rec.Body.String(), "\"total_actual_cost\":88.8")
+	require.Equal(t, "miss", rec.Header().Get("X-Snapshot-Cache"))
+
+	req2 := httptest.NewRequest(http.MethodGet, "/admin/dashboard/users-ranking?limit=100&start_date=2025-01-01&end_date=2025-01-02", nil)
+	rec2 := httptest.NewRecorder()
+	router.ServeHTTP(rec2, req2)
+
+	require.Equal(t, http.StatusOK, rec2.Code)
+	require.Equal(t, "hit", rec2.Header().Get("X-Snapshot-Cache"))
+}

backend/internal/handler/admin/dashboard_query_cache.go

@@ -0,0 +1,200 @@
+package admin
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"time"
+
+	"github.com/Wei-Shaw/sub2api/internal/pkg/usagestats"
+)
+
+var (
+	dashboardTrendCache        = newSnapshotCache(30 * time.Second)
+	dashboardModelStatsCache   = newSnapshotCache(30 * time.Second)
+	dashboardGroupStatsCache   = newSnapshotCache(30 * time.Second)
+	dashboardUsersTrendCache   = newSnapshotCache(30 * time.Second)
+	dashboardAPIKeysTrendCache = newSnapshotCache(30 * time.Second)
+)
+
+type dashboardTrendCacheKey struct {
+	StartTime   string `json:"start_time"`
+	EndTime     string `json:"end_time"`
+	Granularity string `json:"granularity"`
+	UserID      int64  `json:"user_id"`
+	APIKeyID    int64  `json:"api_key_id"`
+	AccountID   int64  `json:"account_id"`
+	GroupID     int64  `json:"group_id"`
+	Model       string `json:"model"`
+	RequestType *int16 `json:"request_type"`
+	Stream      *bool  `json:"stream"`
+	BillingType *int8  `json:"billing_type"`
+}
+
+type dashboardModelGroupCacheKey struct {
+	StartTime   string `json:"start_time"`
+	EndTime     string `json:"end_time"`
+	UserID      int64  `json:"user_id"`
+	APIKeyID    int64  `json:"api_key_id"`
+	AccountID   int64  `json:"account_id"`
+	GroupID     int64  `json:"group_id"`
+	RequestType *int16 `json:"request_type"`
+	Stream      *bool  `json:"stream"`
+	BillingType *int8  `json:"billing_type"`
+}
+
+type dashboardEntityTrendCacheKey struct {
+	StartTime   string `json:"start_time"`
+	EndTime     string `json:"end_time"`
+	Granularity string `json:"granularity"`
+	Limit       int    `json:"limit"`
+}
+
+func cacheStatusValue(hit bool) string {
+	if hit {
+		return "hit"
+	}
+	return "miss"
+}
+
+func mustMarshalDashboardCacheKey(value any) string {
+	raw, err := json.Marshal(value)
+	if err != nil {
+		return ""
+	}
+	return string(raw)
+}
+
+func snapshotPayloadAs[T any](payload any) (T, error) {
+	typed, ok := payload.(T)
+	if !ok {
+		var zero T
+		return zero, fmt.Errorf("unexpected cache payload type %T", payload)
+	}
+	return typed, nil
+}
+
+func (h *DashboardHandler) getUsageTrendCached(
+	ctx context.Context,
+	startTime, endTime time.Time,
+	granularity string,
+	userID, apiKeyID, accountID, groupID int64,
+	model string,
+	requestType *int16,
+	stream *bool,
+	billingType *int8,
+) ([]usagestats.TrendDataPoint, bool, error) {
+	key := mustMarshalDashboardCacheKey(dashboardTrendCacheKey{
+		StartTime:   startTime.UTC().Format(time.RFC3339),
+		EndTime:     endTime.UTC().Format(time.RFC3339),
+		Granularity: granularity,
+		UserID:      userID,
+		APIKeyID:    apiKeyID,
+		AccountID:   accountID,
+		GroupID:     groupID,
+		Model:       model,
+		RequestType: requestType,
+		Stream:      stream,
+		BillingType: billingType,
+	})
+	entry, hit, err := dashboardTrendCache.GetOrLoad(key, func() (any, error) {
+		return h.dashboardService.GetUsageTrendWithFilters(ctx, startTime, endTime, granularity, userID, apiKeyID, accountID, groupID, model, requestType, stream, billingType)
+	})
+	if err != nil {
+		return nil, hit, err
+	}
+	trend, err := snapshotPayloadAs[[]usagestats.TrendDataPoint](entry.Payload)
+	return trend, hit, err
+}
+
+func (h *DashboardHandler) getModelStatsCached(
+	ctx context.Context,
+	startTime, endTime time.Time,
+	userID, apiKeyID, accountID, groupID int64,
+	requestType *int16,
+	stream *bool,
+	billingType *int8,
+) ([]usagestats.ModelStat, bool, error) {
+	key := mustMarshalDashboardCacheKey(dashboardModelGroupCacheKey{
+		StartTime:   startTime.UTC().Format(time.RFC3339),
+		EndTime:     endTime.UTC().Format(time.RFC3339),
+		UserID:      userID,
+		APIKeyID:    apiKeyID,
+		AccountID:   accountID,
+		GroupID:     groupID,
+		RequestType: requestType,
+		Stream:      stream,
+		BillingType: billingType,
+	})
+	entry, hit, err := dashboardModelStatsCache.GetOrLoad(key, func() (any, error) {
+		return h.dashboardService.GetModelStatsWithFilters(ctx, startTime, endTime, userID, apiKeyID, accountID, groupID, requestType, stream, billingType)
+	})
+	if err != nil {
+		return nil, hit, err
+	}
+	stats, err := snapshotPayloadAs[[]usagestats.ModelStat](entry.Payload)
+	return stats, hit, err
+}
+
+func (h *DashboardHandler) getGroupStatsCached(
+	ctx context.Context,
+	startTime, endTime time.Time,
+	userID, apiKeyID, accountID, groupID int64,
+	requestType *int16,
+	stream *bool,
+	billingType *int8,
+) ([]usagestats.GroupStat, bool, error) {
+	key := mustMarshalDashboardCacheKey(dashboardModelGroupCacheKey{
+		StartTime:   startTime.UTC().Format(time.RFC3339),
+		EndTime:     endTime.UTC().Format(time.RFC3339),
+		UserID:      userID,
+		APIKeyID:    apiKeyID,
+		AccountID:   accountID,
+		GroupID:     groupID,
+		RequestType: requestType,
+		Stream:      stream,
+		BillingType: billingType,
+	})
+	entry, hit, err := dashboardGroupStatsCache.GetOrLoad(key, func() (any, error) {
+		return h.dashboardService.GetGroupStatsWithFilters(ctx, startTime, endTime, userID, apiKeyID, accountID, groupID, requestType, stream, billingType)
+	})
+	if err != nil {
+		return nil, hit, err
+	}
+	stats, err := snapshotPayloadAs[[]usagestats.GroupStat](entry.Payload)
+	return stats, hit, err
+}
+
+func (h *DashboardHandler) getAPIKeyUsageTrendCached(ctx context.Context, startTime, endTime time.Time, granularity string, limit int) ([]usagestats.APIKeyUsageTrendPoint, bool, error) {
+	key := mustMarshalDashboardCacheKey(dashboardEntityTrendCacheKey{
+		StartTime:   startTime.UTC().Format(time.RFC3339),
+		EndTime:     endTime.UTC().Format(time.RFC3339),
+		Granularity: granularity,
+		Limit:       limit,
+	})
+	entry, hit, err := dashboardAPIKeysTrendCache.GetOrLoad(key, func() (any, error) {
+		return h.dashboardService.GetAPIKeyUsageTrend(ctx, startTime, endTime, granularity, limit)
+	})
+	if err != nil {
+		return nil, hit, err
+	}
+	trend, err := snapshotPayloadAs[[]usagestats.APIKeyUsageTrendPoint](entry.Payload)
+	return trend, hit, err
+}
+
+func (h *DashboardHandler) getUserUsageTrendCached(ctx context.Context, startTime, endTime time.Time, granularity string, limit int) ([]usagestats.UserUsageTrendPoint, bool, error) {
+	key := mustMarshalDashboardCacheKey(dashboardEntityTrendCacheKey{
+		StartTime:   startTime.UTC().Format(time.RFC3339),
+		EndTime:     endTime.UTC().Format(time.RFC3339),
+		Granularity: granularity,
+		Limit:       limit,
+	})
+	entry, hit, err := dashboardUsersTrendCache.GetOrLoad(key, func() (any, error) {
+		return h.dashboardService.GetUserUsageTrend(ctx, startTime, endTime, granularity, limit)
+	})
+	if err != nil {
+		return nil, hit, err
+	}
+	trend, err := snapshotPayloadAs[[]usagestats.UserUsageTrendPoint](entry.Payload)
+	return trend, hit, err
+}

backend/internal/handler/admin/dashboard_snapshot_v2_handler.go

@@ -1,7 +1,9 @@
 package admin
 
 import (
+	"context"
 	"encoding/json"
+	"errors"
 	"net/http"
 	"strconv"
 	"strings"
@@ -111,20 +113,45 @@ func (h *DashboardHandler) GetSnapshotV2(c *gin.Context) {
 	})
 	cacheKey := string(keyRaw)
 
-	if cached, ok := dashboardSnapshotV2Cache.Get(cacheKey); ok {
-		if cached.ETag != "" {
-			c.Header("ETag", cached.ETag)
-			c.Header("Vary", "If-None-Match")
-			if ifNoneMatchMatched(c.GetHeader("If-None-Match"), cached.ETag) {
-				c.Status(http.StatusNotModified)
-				return
-			}
-		}
-		c.Header("X-Snapshot-Cache", "hit")
-		response.Success(c, cached.Payload)
+	cached, hit, err := dashboardSnapshotV2Cache.GetOrLoad(cacheKey, func() (any, error) {
+		return h.buildSnapshotV2Response(
+			c.Request.Context(),
+			startTime,
+			endTime,
+			granularity,
+			filters,
+			includeStats,
+			includeTrend,
+			includeModels,
+			includeGroups,
+			includeUsersTrend,
+			usersTrendLimit,
+		)
+	})
+	if err != nil {
+		response.Error(c, 500, err.Error())
 		return
 	}
+	if cached.ETag != "" {
+		c.Header("ETag", cached.ETag)
+		c.Header("Vary", "If-None-Match")
+		if ifNoneMatchMatched(c.GetHeader("If-None-Match"), cached.ETag) {
+			c.Status(http.StatusNotModified)
+			return
+		}
+	}
+	c.Header("X-Snapshot-Cache", cacheStatusValue(hit))
+	response.Success(c, cached.Payload)
+}
 
+func (h *DashboardHandler) buildSnapshotV2Response(
+	ctx context.Context,
+	startTime, endTime time.Time,
+	granularity string,
+	filters *dashboardSnapshotV2Filters,
+	includeStats, includeTrend, includeModels, includeGroups, includeUsersTrend bool,
+	usersTrendLimit int,
+) (*dashboardSnapshotV2Response, error) {
 	resp := &dashboardSnapshotV2Response{
 		GeneratedAt: time.Now().UTC().Format(time.RFC3339),
 		StartDate:   startTime.Format("2006-01-02"),
@@ -133,10 +160,9 @@ func (h *DashboardHandler) GetSnapshotV2(c *gin.Context) {
 	}
 
 	if includeStats {
-		stats, err := h.dashboardService.GetDashboardStats(c.Request.Context())
+		stats, err := h.dashboardService.GetDashboardStats(ctx)
 		if err != nil {
-			response.Error(c, 500, "Failed to get dashboard statistics")
-			return
+			return nil, errors.New("failed to get dashboard statistics")
 		}
 		resp.Stats = &dashboardSnapshotV2Stats{
 			DashboardStats: *stats,
@@ -145,8 +171,8 @@ func (h *DashboardHandler) GetSnapshotV2(c *gin.Context) {
 	}
 
 	if includeTrend {
-		trend, err := h.dashboardService.GetUsageTrendWithFilters(
-			c.Request.Context(),
+		trend, _, err := h.getUsageTrendCached(
+			ctx,
 			startTime,
 			endTime,
 			granularity,
@@ -160,15 +186,14 @@ func (h *DashboardHandler) GetSnapshotV2(c *gin.Context) {
 			filters.BillingType,
 		)
 		if err != nil {
-			response.Error(c, 500, "Failed to get usage trend")
-			return
+			return nil, errors.New("failed to get usage trend")
 		}
 		resp.Trend = trend
 	}
 
 	if includeModels {
-		models, err := h.dashboardService.GetModelStatsWithFilters(
-			c.Request.Context(),
+		models, _, err := h.getModelStatsCached(
+			ctx,
 			startTime,
 			endTime,
 			filters.UserID,
@@ -180,15 +205,14 @@ func (h *DashboardHandler) GetSnapshotV2(c *gin.Context) {
 			filters.BillingType,
 		)
 		if err != nil {
-			response.Error(c, 500, "Failed to get model statistics")
-			return
+			return nil, errors.New("failed to get model statistics")
 		}
 		resp.Models = models
 	}
 
 	if includeGroups {
-		groups, err := h.dashboardService.GetGroupStatsWithFilters(
-			c.Request.Context(),
+		groups, _, err := h.getGroupStatsCached(
+			ctx,
 			startTime,
 			endTime,
 			filters.UserID,
@@ -200,34 +224,20 @@ func (h *DashboardHandler) GetSnapshotV2(c *gin.Context) {
 			filters.BillingType,
 		)
 		if err != nil {
-			response.Error(c, 500, "Failed to get group statistics")
-			return
+			return nil, errors.New("failed to get group statistics")
 		}
 		resp.Groups = groups
 	}
 
 	if includeUsersTrend {
-		usersTrend, err := h.dashboardService.GetUserUsageTrend(
-			c.Request.Context(),
-			startTime,
-			endTime,
-			granularity,
-			usersTrendLimit,
-		)
+		usersTrend, _, err := h.getUserUsageTrendCached(ctx, startTime, endTime, granularity, usersTrendLimit)
 		if err != nil {
-			response.Error(c, 500, "Failed to get user usage trend")
-			return
+			return nil, errors.New("failed to get user usage trend")
 		}
 		resp.UsersTrend = usersTrend
 	}
 
-	cached := dashboardSnapshotV2Cache.Set(cacheKey, resp)
-	if cached.ETag != "" {
-		c.Header("ETag", cached.ETag)
-		c.Header("Vary", "If-None-Match")
-	}
-	c.Header("X-Snapshot-Cache", "miss")
-	response.Success(c, resp)
+	return resp, nil
 }
 
 func parseDashboardSnapshotV2Filters(c *gin.Context) (*dashboardSnapshotV2Filters, error) {

backend/internal/handler/admin/group_handler.go

@@ -53,6 +53,9 @@ type CreateGroupRequest struct {
 	SupportedModelScopes []string `json:"supported_model_scopes"`
 	// Sora 存储配额
 	SoraStorageQuotaBytes int64 `json:"sora_storage_quota_bytes"`
+	// OpenAI Messages 调度配置（仅 openai 平台使用）
+	AllowMessagesDispatch bool   `json:"allow_messages_dispatch"`
+	DefaultMappedModel    string `json:"default_mapped_model"`
 	// 从指定分组复制账号（创建后自动绑定）
 	CopyAccountsFromGroupIDs []int64 `json:"copy_accounts_from_group_ids"`
 }
@@ -88,6 +91,9 @@ type UpdateGroupRequest struct {
 	SupportedModelScopes *[]string `json:"supported_model_scopes"`
 	// Sora 存储配额
 	SoraStorageQuotaBytes *int64 `json:"sora_storage_quota_bytes"`
+	// OpenAI Messages 调度配置（仅 openai 平台使用）
+	AllowMessagesDispatch *bool   `json:"allow_messages_dispatch"`
+	DefaultMappedModel    *string `json:"default_mapped_model"`
 	// 从指定分组复制账号（同步操作：先清空当前分组的账号绑定，再绑定源分组的账号）
 	CopyAccountsFromGroupIDs []int64 `json:"copy_accounts_from_group_ids"`
 }
@@ -203,6 +209,8 @@ func (h *GroupHandler) Create(c *gin.Context) {
 		MCPXMLInject:                    req.MCPXMLInject,
 		SupportedModelScopes:            req.SupportedModelScopes,
 		SoraStorageQuotaBytes:           req.SoraStorageQuotaBytes,
+		AllowMessagesDispatch:           req.AllowMessagesDispatch,
+		DefaultMappedModel:              req.DefaultMappedModel,
 		CopyAccountsFromGroupIDs:        req.CopyAccountsFromGroupIDs,
 	})
 	if err != nil {
@@ -254,6 +262,8 @@ func (h *GroupHandler) Update(c *gin.Context) {
 		MCPXMLInject:                    req.MCPXMLInject,
 		SupportedModelScopes:            req.SupportedModelScopes,
 		SoraStorageQuotaBytes:           req.SoraStorageQuotaBytes,
+		AllowMessagesDispatch:           req.AllowMessagesDispatch,
+		DefaultMappedModel:              req.DefaultMappedModel,
 		CopyAccountsFromGroupIDs:        req.CopyAccountsFromGroupIDs,
 	})
 	if err != nil {
@@ -325,6 +335,72 @@ func (h *GroupHandler) GetGroupAPIKeys(c *gin.Context) {
 	response.Paginated(c, outKeys, total, page, pageSize)
 }
 
+// GetGroupRateMultipliers handles getting rate multipliers for users in a group
+// GET /api/v1/admin/groups/:id/rate-multipliers
+func (h *GroupHandler) GetGroupRateMultipliers(c *gin.Context) {
+	groupID, err := strconv.ParseInt(c.Param("id"), 10, 64)
+	if err != nil {
+		response.BadRequest(c, "Invalid group ID")
+		return
+	}
+
+	entries, err := h.adminService.GetGroupRateMultipliers(c.Request.Context(), groupID)
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	if entries == nil {
+		entries = []service.UserGroupRateEntry{}
+	}
+	response.Success(c, entries)
+}
+
+// ClearGroupRateMultipliers handles clearing all rate multipliers for a group
+// DELETE /api/v1/admin/groups/:id/rate-multipliers
+func (h *GroupHandler) ClearGroupRateMultipliers(c *gin.Context) {
+	groupID, err := strconv.ParseInt(c.Param("id"), 10, 64)
+	if err != nil {
+		response.BadRequest(c, "Invalid group ID")
+		return
+	}
+
+	if err := h.adminService.ClearGroupRateMultipliers(c.Request.Context(), groupID); err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	response.Success(c, gin.H{"message": "Rate multipliers cleared successfully"})
+}
+
+// BatchSetGroupRateMultipliersRequest represents batch set rate multipliers request
+type BatchSetGroupRateMultipliersRequest struct {
+	Entries []service.GroupRateMultiplierInput `json:"entries" binding:"required"`
+}
+
+// BatchSetGroupRateMultipliers handles batch setting rate multipliers for a group
+// PUT /api/v1/admin/groups/:id/rate-multipliers
+func (h *GroupHandler) BatchSetGroupRateMultipliers(c *gin.Context) {
+	groupID, err := strconv.ParseInt(c.Param("id"), 10, 64)
+	if err != nil {
+		response.BadRequest(c, "Invalid group ID")
+		return
+	}
+
+	var req BatchSetGroupRateMultipliersRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		response.BadRequest(c, "Invalid request: "+err.Error())
+		return
+	}
+
+	if err := h.adminService.BatchSetGroupRateMultipliers(c.Request.Context(), groupID, req.Entries); err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	response.Success(c, gin.H{"message": "Rate multipliers updated successfully"})
+}
+
 // UpdateSortOrderRequest represents the request to update group sort orders
 type UpdateSortOrderRequest struct {
 	Updates []struct {

backend/internal/handler/admin/openai_oauth_handler.go

@@ -289,6 +289,7 @@ func (h *OpenAIOAuthHandler) CreateAccountFromOAuth(c *gin.Context) {
 		Platform:    platform,
 		Type:        "oauth",
 		Credentials: credentials,
+		Extra:       nil,
 		ProxyID:     req.ProxyID,
 		Concurrency: req.Concurrency,
 		Priority:    req.Priority,

backend/internal/handler/admin/ops_alerts_handler.go

@@ -23,6 +23,13 @@ var validOpsAlertMetricTypes = []string{
 	"cpu_usage_percent",
 	"memory_usage_percent",
 	"concurrency_queue_depth",
+	"group_available_accounts",
+	"group_available_ratio",
+	"group_rate_limit_ratio",
+	"account_rate_limited_count",
+	"account_error_count",
+	"account_error_ratio",
+	"overload_account_count",
 }
 
 var validOpsAlertMetricTypeSet = func() map[string]struct{} {
@@ -82,7 +89,10 @@ func isPercentOrRateMetric(metricType string) bool {
 		"error_rate",
 		"upstream_error_rate",
 		"cpu_usage_percent",
-		"memory_usage_percent":
+		"memory_usage_percent",
+		"group_available_ratio",
+		"group_rate_limit_ratio",
+		"account_error_ratio":
 		return true
 	default:
 		return false

backend/internal/handler/admin/redeem_handler.go

@@ -41,12 +41,15 @@ type GenerateRedeemCodesRequest struct {
 }
 
 // CreateAndRedeemCodeRequest represents creating a fixed code and redeeming it for a target user.
+// Type 为 omitempty 而非 required 是为了向后兼容旧版调用方（不传 type 时默认 balance）。
 type CreateAndRedeemCodeRequest struct {
-	Code   string  `json:"code" binding:"required,min=3,max=128"`
-	Type   string  `json:"type" binding:"required,oneof=balance concurrency subscription invitation"`
-	Value  float64 `json:"value" binding:"required,gt=0"`
-	UserID int64   `json:"user_id" binding:"required,gt=0"`
-	Notes  string  `json:"notes"`
+	Code         string  `json:"code" binding:"required,min=3,max=128"`
+	Type         string  `json:"type" binding:"omitempty,oneof=balance concurrency subscription invitation"` // 不传时默认 balance（向后兼容）
+	Value        float64 `json:"value" binding:"required,gt=0"`
+	UserID       int64   `json:"user_id" binding:"required,gt=0"`
+	GroupID      *int64  `json:"group_id"`                                    // subscription 类型必填
+	ValidityDays int     `json:"validity_days" binding:"omitempty,max=36500"` // subscription 类型必填，>0
+	Notes        string  `json:"notes"`
 }
 
 // List handles listing all redeem codes with pagination
@@ -136,6 +139,22 @@ func (h *RedeemHandler) CreateAndRedeem(c *gin.Context) {
 		return
 	}
 	req.Code = strings.TrimSpace(req.Code)
+	// 向后兼容：旧版调用方（如 Sub2ApiPay）不传 type 字段，默认当作 balance 充值处理。
+	// 请勿删除此默认值逻辑，否则会导致旧版调用方 400 报错。
+	if req.Type == "" {
+		req.Type = "balance"
+	}
+
+	if req.Type == "subscription" {
+		if req.GroupID == nil {
+			response.BadRequest(c, "group_id is required for subscription type")
+			return
+		}
+		if req.ValidityDays <= 0 {
+			response.BadRequest(c, "validity_days must be greater than 0 for subscription type")
+			return
+		}
+	}
 
 	executeAdminIdempotentJSON(c, "admin.redeem_codes.create_and_redeem", req, service.DefaultWriteIdempotencyTTL(), func(ctx context.Context) (any, error) {
 		existing, err := h.redeemService.GetByCode(ctx, req.Code)
@@ -147,11 +166,13 @@ func (h *RedeemHandler) CreateAndRedeem(c *gin.Context) {
 		}
 
 		createErr := h.redeemService.CreateCode(ctx, &service.RedeemCode{
-			Code:   req.Code,
-			Type:   req.Type,
-			Value:  req.Value,
-			Status: service.StatusUnused,
-			Notes:  req.Notes,
+			Code:         req.Code,
+			Type:         req.Type,
+			Value:        req.Value,
+			Status:       service.StatusUnused,
+			Notes:        req.Notes,
+			GroupID:      req.GroupID,
+			ValidityDays: req.ValidityDays,
 		})
 		if createErr != nil {
 			// Unique code race: if code now exists, use idempotent semantics by used_by.

backend/internal/handler/admin/redeem_handler_test.go

@@ -0,0 +1,135 @@
+package admin
+
+import (
+	"bytes"
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"testing"
+
+	"github.com/Wei-Shaw/sub2api/internal/service"
+	"github.com/gin-gonic/gin"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// newCreateAndRedeemHandler creates a RedeemHandler with a non-nil (but minimal)
+// RedeemService so that CreateAndRedeem's nil guard passes and we can test the
+// parameter-validation layer that runs before any service call.
+func newCreateAndRedeemHandler() *RedeemHandler {
+	return &RedeemHandler{
+		adminService:  newStubAdminService(),
+		redeemService: &service.RedeemService{}, // non-nil to pass nil guard
+	}
+}
+
+// postCreateAndRedeemValidation calls CreateAndRedeem and returns the response
+// status code. For cases that pass validation and proceed into the service layer,
+// a panic may occur (because RedeemService internals are nil); this is expected
+// and treated as "validation passed" (returns 0 to indicate panic).
+func postCreateAndRedeemValidation(t *testing.T, handler *RedeemHandler, body any) (code int) {
+	t.Helper()
+	gin.SetMode(gin.TestMode)
+	w := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(w)
+
+	jsonBytes, err := json.Marshal(body)
+	require.NoError(t, err)
+	c.Request, _ = http.NewRequest(http.MethodPost, "/api/v1/admin/redeem-codes/create-and-redeem", bytes.NewReader(jsonBytes))
+	c.Request.Header.Set("Content-Type", "application/json")
+
+	defer func() {
+		if r := recover(); r != nil {
+			// Panic means we passed validation and entered service layer (expected for minimal stub).
+			code = 0
+		}
+	}()
+	handler.CreateAndRedeem(c)
+	return w.Code
+}
+
+func TestCreateAndRedeem_TypeDefaultsToBalance(t *testing.T) {
+	// 不传 type 字段时应默认 balance，不触发 subscription 校验。
+	// 验证通过后进入 service 层会 panic（返回 0），说明默认值生效。
+	h := newCreateAndRedeemHandler()
+	code := postCreateAndRedeemValidation(t, h, map[string]any{
+		"code":    "test-balance-default",
+		"value":   10.0,
+		"user_id": 1,
+	})
+
+	assert.NotEqual(t, http.StatusBadRequest, code,
+		"omitting type should default to balance and pass validation")
+}
+
+func TestCreateAndRedeem_SubscriptionRequiresGroupID(t *testing.T) {
+	h := newCreateAndRedeemHandler()
+	code := postCreateAndRedeemValidation(t, h, map[string]any{
+		"code":          "test-sub-no-group",
+		"type":          "subscription",
+		"value":         29.9,
+		"user_id":       1,
+		"validity_days": 30,
+		// group_id 缺失
+	})
+
+	assert.Equal(t, http.StatusBadRequest, code)
+}
+
+func TestCreateAndRedeem_SubscriptionRequiresPositiveValidityDays(t *testing.T) {
+	groupID := int64(5)
+	h := newCreateAndRedeemHandler()
+
+	cases := []struct {
+		name         string
+		validityDays int
+	}{
+		{"zero", 0},
+		{"negative", -1},
+	}
+
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			code := postCreateAndRedeemValidation(t, h, map[string]any{
+				"code":          "test-sub-bad-days-" + tc.name,
+				"type":          "subscription",
+				"value":         29.9,
+				"user_id":       1,
+				"group_id":      groupID,
+				"validity_days": tc.validityDays,
+			})
+
+			assert.Equal(t, http.StatusBadRequest, code)
+		})
+	}
+}
+
+func TestCreateAndRedeem_SubscriptionValidParamsPassValidation(t *testing.T) {
+	groupID := int64(5)
+	h := newCreateAndRedeemHandler()
+	code := postCreateAndRedeemValidation(t, h, map[string]any{
+		"code":          "test-sub-valid",
+		"type":          "subscription",
+		"value":         29.9,
+		"user_id":       1,
+		"group_id":      groupID,
+		"validity_days": 31,
+	})
+
+	assert.NotEqual(t, http.StatusBadRequest, code,
+		"valid subscription params should pass validation")
+}
+
+func TestCreateAndRedeem_BalanceIgnoresSubscriptionFields(t *testing.T) {
+	h := newCreateAndRedeemHandler()
+	// balance 类型不传 group_id 和 validity_days，不应报 400
+	code := postCreateAndRedeemValidation(t, h, map[string]any{
+		"code":    "test-balance-no-extras",
+		"type":    "balance",
+		"value":   50.0,
+		"user_id": 1,
+	})
+
+	assert.NotEqual(t, http.StatusBadRequest, code,
+		"balance type should not require group_id or validity_days")
+}

backend/internal/handler/admin/scheduled_test_handler.go

@@ -25,6 +25,7 @@ type createScheduledTestPlanRequest struct {
 	CronExpression string `json:"cron_expression" binding:"required"`
 	Enabled        *bool  `json:"enabled"`
 	MaxResults     int    `json:"max_results"`
+	AutoRecover    *bool  `json:"auto_recover"`
 }
 
 type updateScheduledTestPlanRequest struct {
@@ -32,6 +33,7 @@ type updateScheduledTestPlanRequest struct {
 	CronExpression string `json:"cron_expression"`
 	Enabled        *bool  `json:"enabled"`
 	MaxResults     int    `json:"max_results"`
+	AutoRecover    *bool  `json:"auto_recover"`
 }
 
 // ListByAccount GET /admin/accounts/:id/scheduled-test-plans
@@ -68,6 +70,9 @@ func (h *ScheduledTestHandler) Create(c *gin.Context) {
 	if req.Enabled != nil {
 		plan.Enabled = *req.Enabled
 	}
+	if req.AutoRecover != nil {
+		plan.AutoRecover = *req.AutoRecover
+	}
 
 	created, err := h.scheduledTestSvc.CreatePlan(c.Request.Context(), plan)
 	if err != nil {
@@ -109,6 +114,9 @@ func (h *ScheduledTestHandler) Update(c *gin.Context) {
 	if req.MaxResults > 0 {
 		existing.MaxResults = req.MaxResults
 	}
+	if req.AutoRecover != nil {
+		existing.AutoRecover = *req.AutoRecover
+	}
 
 	updated, err := h.scheduledTestSvc.UpdatePlan(c.Request.Context(), existing)
 	if err != nil {

backend/internal/handler/admin/setting_handler.go

@@ -125,6 +125,7 @@ func (h *SettingHandler) GetSettings(c *gin.Context) {
 		OpsMetricsIntervalSeconds:            settings.OpsMetricsIntervalSeconds,
 		MinClaudeCodeVersion:                 settings.MinClaudeCodeVersion,
 		AllowUngroupedKeyScheduling:          settings.AllowUngroupedKeyScheduling,
+		BackendModeEnabled:                   settings.BackendModeEnabled,
 	})
 }
 
@@ -199,6 +200,9 @@ type UpdateSettingsRequest struct {
 
 	// 分组隔离
 	AllowUngroupedKeyScheduling bool `json:"allow_ungrouped_key_scheduling"`
+
+	// Backend Mode
+	BackendModeEnabled bool `json:"backend_mode_enabled"`
 }
 
 // UpdateSettings 更新系统设置
@@ -473,6 +477,7 @@ func (h *SettingHandler) UpdateSettings(c *gin.Context) {
 		IdentityPatchPrompt:              req.IdentityPatchPrompt,
 		MinClaudeCodeVersion:             req.MinClaudeCodeVersion,
 		AllowUngroupedKeyScheduling:      req.AllowUngroupedKeyScheduling,
+		BackendModeEnabled:               req.BackendModeEnabled,
 		OpsMonitoringEnabled: func() bool {
 			if req.OpsMonitoringEnabled != nil {
 				return *req.OpsMonitoringEnabled
@@ -571,6 +576,7 @@ func (h *SettingHandler) UpdateSettings(c *gin.Context) {
 		OpsMetricsIntervalSeconds:            updatedSettings.OpsMetricsIntervalSeconds,
 		MinClaudeCodeVersion:                 updatedSettings.MinClaudeCodeVersion,
 		AllowUngroupedKeyScheduling:          updatedSettings.AllowUngroupedKeyScheduling,
+		BackendModeEnabled:                   updatedSettings.BackendModeEnabled,
 	})
 }
 
@@ -725,6 +731,9 @@ func diffSettings(before *service.SystemSettings, after *service.SystemSettings,
 	if before.AllowUngroupedKeyScheduling != after.AllowUngroupedKeyScheduling {
 		changed = append(changed, "allow_ungrouped_key_scheduling")
 	}
+	if before.BackendModeEnabled != after.BackendModeEnabled {
+		changed = append(changed, "backend_mode_enabled")
+	}
 	if before.PurchaseSubscriptionEnabled != after.PurchaseSubscriptionEnabled {
 		changed = append(changed, "purchase_subscription_enabled")
 	}
@@ -1348,6 +1357,118 @@ func (h *SettingHandler) TestSoraS3Connection(c *gin.Context) {
 	response.Success(c, gin.H{"message": "S3 连接成功"})
 }
 
+// GetRectifierSettings 获取请求整流器配置
+// GET /api/v1/admin/settings/rectifier
+func (h *SettingHandler) GetRectifierSettings(c *gin.Context) {
+	settings, err := h.settingService.GetRectifierSettings(c.Request.Context())
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	response.Success(c, dto.RectifierSettings{
+		Enabled:                  settings.Enabled,
+		ThinkingSignatureEnabled: settings.ThinkingSignatureEnabled,
+		ThinkingBudgetEnabled:    settings.ThinkingBudgetEnabled,
+	})
+}
+
+// UpdateRectifierSettingsRequest 更新整流器配置请求
+type UpdateRectifierSettingsRequest struct {
+	Enabled                  bool `json:"enabled"`
+	ThinkingSignatureEnabled bool `json:"thinking_signature_enabled"`
+	ThinkingBudgetEnabled    bool `json:"thinking_budget_enabled"`
+}
+
+// UpdateRectifierSettings 更新请求整流器配置
+// PUT /api/v1/admin/settings/rectifier
+func (h *SettingHandler) UpdateRectifierSettings(c *gin.Context) {
+	var req UpdateRectifierSettingsRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		response.BadRequest(c, "Invalid request: "+err.Error())
+		return
+	}
+
+	settings := &service.RectifierSettings{
+		Enabled:                  req.Enabled,
+		ThinkingSignatureEnabled: req.ThinkingSignatureEnabled,
+		ThinkingBudgetEnabled:    req.ThinkingBudgetEnabled,
+	}
+
+	if err := h.settingService.SetRectifierSettings(c.Request.Context(), settings); err != nil {
+		response.BadRequest(c, err.Error())
+		return
+	}
+
+	// 重新获取设置返回
+	updatedSettings, err := h.settingService.GetRectifierSettings(c.Request.Context())
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	response.Success(c, dto.RectifierSettings{
+		Enabled:                  updatedSettings.Enabled,
+		ThinkingSignatureEnabled: updatedSettings.ThinkingSignatureEnabled,
+		ThinkingBudgetEnabled:    updatedSettings.ThinkingBudgetEnabled,
+	})
+}
+
+// GetBetaPolicySettings 获取 Beta 策略配置
+// GET /api/v1/admin/settings/beta-policy
+func (h *SettingHandler) GetBetaPolicySettings(c *gin.Context) {
+	settings, err := h.settingService.GetBetaPolicySettings(c.Request.Context())
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	rules := make([]dto.BetaPolicyRule, len(settings.Rules))
+	for i, r := range settings.Rules {
+		rules[i] = dto.BetaPolicyRule(r)
+	}
+	response.Success(c, dto.BetaPolicySettings{Rules: rules})
+}
+
+// UpdateBetaPolicySettingsRequest 更新 Beta 策略配置请求
+type UpdateBetaPolicySettingsRequest struct {
+	Rules []dto.BetaPolicyRule `json:"rules"`
+}
+
+// UpdateBetaPolicySettings 更新 Beta 策略配置
+// PUT /api/v1/admin/settings/beta-policy
+func (h *SettingHandler) UpdateBetaPolicySettings(c *gin.Context) {
+	var req UpdateBetaPolicySettingsRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		response.BadRequest(c, "Invalid request: "+err.Error())
+		return
+	}
+
+	rules := make([]service.BetaPolicyRule, len(req.Rules))
+	for i, r := range req.Rules {
+		rules[i] = service.BetaPolicyRule(r)
+	}
+
+	settings := &service.BetaPolicySettings{Rules: rules}
+	if err := h.settingService.SetBetaPolicySettings(c.Request.Context(), settings); err != nil {
+		response.BadRequest(c, err.Error())
+		return
+	}
+
+	// Re-fetch to return updated settings
+	updated, err := h.settingService.GetBetaPolicySettings(c.Request.Context())
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	outRules := make([]dto.BetaPolicyRule, len(updated.Rules))
+	for i, r := range updated.Rules {
+		outRules[i] = dto.BetaPolicyRule(r)
+	}
+	response.Success(c, dto.BetaPolicySettings{Rules: outRules})
+}
+
 // UpdateStreamTimeoutSettingsRequest 更新流超时配置请求
 type UpdateStreamTimeoutSettingsRequest struct {
 	Enabled                bool   `json:"enabled"`

backend/internal/handler/admin/snapshot_cache.go

@@ -7,6 +7,8 @@ import (
 	"strings"
 	"sync"
 	"time"
+
+	"golang.org/x/sync/singleflight"
 )
 
 type snapshotCacheEntry struct {
@@ -19,6 +21,12 @@ type snapshotCache struct {
 	mu    sync.RWMutex
 	ttl   time.Duration
 	items map[string]snapshotCacheEntry
+	sf    singleflight.Group
+}
+
+type snapshotCacheLoadResult struct {
+	Entry snapshotCacheEntry
+	Hit   bool
 }
 
 func newSnapshotCache(ttl time.Duration) *snapshotCache {
@@ -70,6 +78,41 @@ func (c *snapshotCache) Set(key string, payload any) snapshotCacheEntry {
 	return entry
 }
 
+func (c *snapshotCache) GetOrLoad(key string, load func() (any, error)) (snapshotCacheEntry, bool, error) {
+	if load == nil {
+		return snapshotCacheEntry{}, false, nil
+	}
+	if entry, ok := c.Get(key); ok {
+		return entry, true, nil
+	}
+	if c == nil || key == "" {
+		payload, err := load()
+		if err != nil {
+			return snapshotCacheEntry{}, false, err
+		}
+		return c.Set(key, payload), false, nil
+	}
+
+	value, err, _ := c.sf.Do(key, func() (any, error) {
+		if entry, ok := c.Get(key); ok {
+			return snapshotCacheLoadResult{Entry: entry, Hit: true}, nil
+		}
+		payload, err := load()
+		if err != nil {
+			return nil, err
+		}
+		return snapshotCacheLoadResult{Entry: c.Set(key, payload), Hit: false}, nil
+	})
+	if err != nil {
+		return snapshotCacheEntry{}, false, err
+	}
+	result, ok := value.(snapshotCacheLoadResult)
+	if !ok {
+		return snapshotCacheEntry{}, false, nil
+	}
+	return result.Entry, result.Hit, nil
+}
+
 func buildETagFromAny(payload any) string {
 	raw, err := json.Marshal(payload)
 	if err != nil {

backend/internal/handler/admin/snapshot_cache_test.go

@@ -3,6 +3,8 @@
 package admin
 
 import (
+	"sync"
+	"sync/atomic"
 	"testing"
 	"time"
 
@@ -95,6 +97,61 @@ func TestBuildETagFromAny_UnmarshalablePayload(t *testing.T) {
 	require.Empty(t, etag)
 }
 
+func TestSnapshotCache_GetOrLoad_MissThenHit(t *testing.T) {
+	c := newSnapshotCache(5 * time.Second)
+	var loads atomic.Int32
+
+	entry, hit, err := c.GetOrLoad("key1", func() (any, error) {
+		loads.Add(1)
+		return map[string]string{"hello": "world"}, nil
+	})
+	require.NoError(t, err)
+	require.False(t, hit)
+	require.NotEmpty(t, entry.ETag)
+	require.Equal(t, int32(1), loads.Load())
+
+	entry2, hit, err := c.GetOrLoad("key1", func() (any, error) {
+		loads.Add(1)
+		return map[string]string{"unexpected": "value"}, nil
+	})
+	require.NoError(t, err)
+	require.True(t, hit)
+	require.Equal(t, entry.ETag, entry2.ETag)
+	require.Equal(t, int32(1), loads.Load())
+}
+
+func TestSnapshotCache_GetOrLoad_ConcurrentSingleflight(t *testing.T) {
+	c := newSnapshotCache(5 * time.Second)
+	var loads atomic.Int32
+	start := make(chan struct{})
+	const callers = 8
+	errCh := make(chan error, callers)
+
+	var wg sync.WaitGroup
+	wg.Add(callers)
+	for range callers {
+		go func() {
+			defer wg.Done()
+			<-start
+			_, _, err := c.GetOrLoad("shared", func() (any, error) {
+				loads.Add(1)
+				time.Sleep(20 * time.Millisecond)
+				return "value", nil
+			})
+			errCh <- err
+		}()
+	}
+	close(start)
+	wg.Wait()
+	close(errCh)
+
+	for err := range errCh {
+		require.NoError(t, err)
+	}
+
+	require.Equal(t, int32(1), loads.Load())
+}
+
 func TestParseBoolQueryWithDefault(t *testing.T) {
 	tests := []struct {
 		name string

backend/internal/handler/admin/subscription_handler.go

@@ -216,6 +216,38 @@ func (h *SubscriptionHandler) Extend(c *gin.Context) {
 	})
 }
 
+// ResetSubscriptionQuotaRequest represents the reset quota request
+type ResetSubscriptionQuotaRequest struct {
+	Daily   bool `json:"daily"`
+	Weekly  bool `json:"weekly"`
+	Monthly bool `json:"monthly"`
+}
+
+// ResetQuota resets daily, weekly, and/or monthly usage for a subscription.
+// POST /api/v1/admin/subscriptions/:id/reset-quota
+func (h *SubscriptionHandler) ResetQuota(c *gin.Context) {
+	subscriptionID, err := strconv.ParseInt(c.Param("id"), 10, 64)
+	if err != nil {
+		response.BadRequest(c, "Invalid subscription ID")
+		return
+	}
+	var req ResetSubscriptionQuotaRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		response.BadRequest(c, "Invalid request: "+err.Error())
+		return
+	}
+	if !req.Daily && !req.Weekly && !req.Monthly {
+		response.BadRequest(c, "At least one of 'daily', 'weekly', or 'monthly' must be true")
+		return
+	}
+	sub, err := h.subscriptionService.AdminResetQuota(c.Request.Context(), subscriptionID, req.Daily, req.Weekly, req.Monthly)
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+	response.Success(c, dto.UserSubscriptionFromServiceAdmin(sub))
+}
+
 // Revoke handles revoking a subscription
 // DELETE /api/v1/admin/subscriptions/:id
 func (h *SubscriptionHandler) Revoke(c *gin.Context) {

backend/internal/handler/auth_handler.go

@@ -194,6 +194,12 @@ func (h *AuthHandler) Login(c *gin.Context) {
 		return
 	}
 
+	// Backend mode: only admin can login
+	if h.settingSvc.IsBackendModeEnabled(c.Request.Context()) && !user.IsAdmin() {
+		response.Forbidden(c, "Backend mode is active. Only admin login is allowed.")
+		return
+	}
+
 	h.respondWithTokenPair(c, user)
 }
 
@@ -250,16 +256,22 @@ func (h *AuthHandler) Login2FA(c *gin.Context) {
 		return
 	}
 
-	// Delete the login session
-	_ = h.totpService.DeleteLoginSession(c.Request.Context(), req.TempToken)
-
-	// Get the user
+	// Get the user (before session deletion so we can check backend mode)
 	user, err := h.userService.GetByID(c.Request.Context(), session.UserID)
 	if err != nil {
 		response.ErrorFrom(c, err)
 		return
 	}
 
+	// Backend mode: only admin can login (check BEFORE deleting session)
+	if h.settingSvc.IsBackendModeEnabled(c.Request.Context()) && !user.IsAdmin() {
+		response.Forbidden(c, "Backend mode is active. Only admin login is allowed.")
+		return
+	}
+
+	// Delete the login session (only after all checks pass)
+	_ = h.totpService.DeleteLoginSession(c.Request.Context(), req.TempToken)
+
 	h.respondWithTokenPair(c, user)
 }
 
@@ -522,16 +534,22 @@ func (h *AuthHandler) RefreshToken(c *gin.Context) {
 		return
 	}
 
-	tokenPair, err := h.authService.RefreshTokenPair(c.Request.Context(), req.RefreshToken)
+	result, err := h.authService.RefreshTokenPair(c.Request.Context(), req.RefreshToken)
 	if err != nil {
 		response.ErrorFrom(c, err)
 		return
 	}
 
+	// Backend mode: block non-admin token refresh
+	if h.settingSvc.IsBackendModeEnabled(c.Request.Context()) && result.UserRole != "admin" {
+		response.Forbidden(c, "Backend mode is active. Only admin login is allowed.")
+		return
+	}
+
 	response.Success(c, RefreshTokenResponse{
-		AccessToken:  tokenPair.AccessToken,
-		RefreshToken: tokenPair.RefreshToken,
-		ExpiresIn:    tokenPair.ExpiresIn,
+		AccessToken:  result.AccessToken,
+		RefreshToken: result.RefreshToken,
+		ExpiresIn:    result.ExpiresIn,
 		TokenType:    "Bearer",
 	})
 }

backend/internal/handler/auth_linuxdo_oauth.go

@@ -211,8 +211,22 @@ func (h *AuthHandler) LinuxDoOAuthCallback(c *gin.Context) {
 		email = linuxDoSyntheticEmail(subject)
 	}
 
-	tokenPair, _, err := h.authService.LoginOrRegisterOAuthWithTokenPair(c.Request.Context(), email, username)
+	// 传入空邀请码；如果需要邀请码，服务层返回 ErrOAuthInvitationRequired
+	tokenPair, _, err := h.authService.LoginOrRegisterOAuthWithTokenPair(c.Request.Context(), email, username, "")
 	if err != nil {
+		if errors.Is(err, service.ErrOAuthInvitationRequired) {
+			pendingToken, tokenErr := h.authService.CreatePendingOAuthToken(email, username)
+			if tokenErr != nil {
+				redirectOAuthError(c, frontendCallback, "login_failed", "service_error", "")
+				return
+			}
+			fragment := url.Values{}
+			fragment.Set("error", "invitation_required")
+			fragment.Set("pending_oauth_token", pendingToken)
+			fragment.Set("redirect", redirectTo)
+			redirectWithFragment(c, frontendCallback, fragment)
+			return
+		}
 		// 避免把内部细节泄露给客户端；给前端保留结构化原因与提示信息即可。
 		redirectOAuthError(c, frontendCallback, "login_failed", infraerrors.Reason(err), infraerrors.Message(err))
 		return
@@ -227,6 +241,41 @@ func (h *AuthHandler) LinuxDoOAuthCallback(c *gin.Context) {
 	redirectWithFragment(c, frontendCallback, fragment)
 }
 
+type completeLinuxDoOAuthRequest struct {
+	PendingOAuthToken string `json:"pending_oauth_token" binding:"required"`
+	InvitationCode    string `json:"invitation_code"     binding:"required"`
+}
+
+// CompleteLinuxDoOAuthRegistration completes a pending OAuth registration by validating
+// the invitation code and creating the user account.
+// POST /api/v1/auth/oauth/linuxdo/complete-registration
+func (h *AuthHandler) CompleteLinuxDoOAuthRegistration(c *gin.Context) {
+	var req completeLinuxDoOAuthRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "INVALID_REQUEST", "message": err.Error()})
+		return
+	}
+
+	email, username, err := h.authService.VerifyPendingOAuthToken(req.PendingOAuthToken)
+	if err != nil {
+		c.JSON(http.StatusUnauthorized, gin.H{"error": "INVALID_TOKEN", "message": "invalid or expired registration token"})
+		return
+	}
+
+	tokenPair, _, err := h.authService.LoginOrRegisterOAuthWithTokenPair(c.Request.Context(), email, username, req.InvitationCode)
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	c.JSON(http.StatusOK, gin.H{
+		"access_token":  tokenPair.AccessToken,
+		"refresh_token": tokenPair.RefreshToken,
+		"expires_in":    tokenPair.ExpiresIn,
+		"token_type":    "Bearer",
+	})
+}
+
 func (h *AuthHandler) getLinuxDoOAuthConfig(ctx context.Context) (config.LinuxDoConnectConfig, error) {
 	if h != nil && h.settingSvc != nil {
 		return h.settingSvc.GetLinuxDoConnectOAuthConfig(ctx)

backend/internal/handler/dto/announcement.go

@@ -7,10 +7,11 @@ import (
 )
 
 type Announcement struct {
-	ID      int64  `json:"id"`
-	Title   string `json:"title"`
-	Content string `json:"content"`
-	Status  string `json:"status"`
+	ID         int64  `json:"id"`
+	Title      string `json:"title"`
+	Content    string `json:"content"`
+	Status     string `json:"status"`
+	NotifyMode string `json:"notify_mode"`
 
 	Targeting service.AnnouncementTargeting `json:"targeting"`
 
@@ -25,9 +26,10 @@ type Announcement struct {
 }
 
 type UserAnnouncement struct {
-	ID      int64  `json:"id"`
-	Title   string `json:"title"`
-	Content string `json:"content"`
+	ID         int64  `json:"id"`
+	Title      string `json:"title"`
+	Content    string `json:"content"`
+	NotifyMode string `json:"notify_mode"`
 
 	StartsAt *time.Time `json:"starts_at,omitempty"`
 	EndsAt   *time.Time `json:"ends_at,omitempty"`
@@ -43,17 +45,18 @@ func AnnouncementFromService(a *service.Announcement) *Announcement {
 		return nil
 	}
 	return &Announcement{
-		ID:        a.ID,
-		Title:     a.Title,
-		Content:   a.Content,
-		Status:    a.Status,
-		Targeting: a.Targeting,
-		StartsAt:  a.StartsAt,
-		EndsAt:    a.EndsAt,
-		CreatedBy: a.CreatedBy,
-		UpdatedBy: a.UpdatedBy,
-		CreatedAt: a.CreatedAt,
-		UpdatedAt: a.UpdatedAt,
+		ID:         a.ID,
+		Title:      a.Title,
+		Content:    a.Content,
+		Status:     a.Status,
+		NotifyMode: a.NotifyMode,
+		Targeting:  a.Targeting,
+		StartsAt:   a.StartsAt,
+		EndsAt:     a.EndsAt,
+		CreatedBy:  a.CreatedBy,
+		UpdatedBy:  a.UpdatedBy,
+		CreatedAt:  a.CreatedAt,
+		UpdatedAt:  a.UpdatedAt,
 	}
 }
 
@@ -62,13 +65,14 @@ func UserAnnouncementFromService(a *service.UserAnnouncement) *UserAnnouncement
 		return nil
 	}
 	return &UserAnnouncement{
-		ID:        a.Announcement.ID,
-		Title:     a.Announcement.Title,
-		Content:   a.Announcement.Content,
-		StartsAt:  a.Announcement.StartsAt,
-		EndsAt:    a.Announcement.EndsAt,
-		ReadAt:    a.ReadAt,
-		CreatedAt: a.Announcement.CreatedAt,
-		UpdatedAt: a.Announcement.UpdatedAt,
+		ID:         a.Announcement.ID,
+		Title:      a.Announcement.Title,
+		Content:    a.Announcement.Content,
+		NotifyMode: a.Announcement.NotifyMode,
+		StartsAt:   a.Announcement.StartsAt,
+		EndsAt:     a.Announcement.EndsAt,
+		ReadAt:     a.ReadAt,
+		CreatedAt:  a.Announcement.CreatedAt,
+		UpdatedAt:  a.Announcement.UpdatedAt,
 	}
 }

backend/internal/handler/dto/mappers.go

@@ -71,7 +71,7 @@ func APIKeyFromService(k *service.APIKey) *APIKey {
 	if k == nil {
 		return nil
 	}
-	return &APIKey{
+	out := &APIKey{
 		ID:            k.ID,
 		UserID:        k.UserID,
 		Key:           k.Key,
@@ -89,15 +89,28 @@ func APIKeyFromService(k *service.APIKey) *APIKey {
 		RateLimit5h:   k.RateLimit5h,
 		RateLimit1d:   k.RateLimit1d,
 		RateLimit7d:   k.RateLimit7d,
-		Usage5h:       k.Usage5h,
-		Usage1d:       k.Usage1d,
-		Usage7d:       k.Usage7d,
+		Usage5h:       k.EffectiveUsage5h(),
+		Usage1d:       k.EffectiveUsage1d(),
+		Usage7d:       k.EffectiveUsage7d(),
 		Window5hStart: k.Window5hStart,
 		Window1dStart: k.Window1dStart,
 		Window7dStart: k.Window7dStart,
 		User:          UserFromServiceShallow(k.User),
 		Group:         GroupFromServiceShallow(k.Group),
 	}
+	if k.Window5hStart != nil && !service.IsWindowExpired(k.Window5hStart, service.RateLimitWindow5h) {
+		t := k.Window5hStart.Add(service.RateLimitWindow5h)
+		out.Reset5hAt = &t
+	}
+	if k.Window1dStart != nil && !service.IsWindowExpired(k.Window1dStart, service.RateLimitWindow1d) {
+		t := k.Window1dStart.Add(service.RateLimitWindow1d)
+		out.Reset1dAt = &t
+	}
+	if k.Window7dStart != nil && !service.IsWindowExpired(k.Window7dStart, service.RateLimitWindow7d) {
+		t := k.Window7dStart.Add(service.RateLimitWindow7d)
+		out.Reset7dAt = &t
+	}
+	return out
 }
 
 func GroupFromServiceShallow(g *service.Group) *Group {
@@ -126,6 +139,7 @@ func GroupFromServiceAdmin(g *service.Group) *AdminGroup {
 		ModelRouting:         g.ModelRouting,
 		ModelRoutingEnabled:  g.ModelRoutingEnabled,
 		MCPXMLInject:         g.MCPXMLInject,
+		DefaultMappedModel:   g.DefaultMappedModel,
 		SupportedModelScopes: g.SupportedModelScopes,
 		AccountCount:         g.AccountCount,
 		SortOrder:            g.SortOrder,
@@ -164,6 +178,7 @@ func groupFromServiceBase(g *service.Group) Group {
 		FallbackGroupID:                 g.FallbackGroupID,
 		FallbackGroupIDOnInvalidRequest: g.FallbackGroupIDOnInvalidRequest,
 		SoraStorageQuotaBytes:           g.SoraStorageQuotaBytes,
+		AllowMessagesDispatch:           g.AllowMessagesDispatch,
 		CreatedAt:                       g.CreatedAt,
 		UpdatedAt:                       g.UpdatedAt,
 	}
@@ -183,6 +198,7 @@ func AccountFromServiceShallow(a *service.Account) *Account {
 		Extra:                   a.Extra,
 		ProxyID:                 a.ProxyID,
 		Concurrency:             a.Concurrency,
+		LoadFactor:              a.LoadFactor,
 		Priority:                a.Priority,
 		RateMultiplier:          a.BillingRateMultiplier(),
 		Status:                  a.Status,
@@ -248,6 +264,50 @@ func AccountFromServiceShallow(a *service.Account) *Account {
 		}
 	}
 
+	// 提取账号配额限制（apikey / bedrock 类型有效）
+	if a.IsAPIKeyOrBedrock() {
+		if limit := a.GetQuotaLimit(); limit > 0 {
+			out.QuotaLimit = &limit
+			used := a.GetQuotaUsed()
+			out.QuotaUsed = &used
+		}
+		if limit := a.GetQuotaDailyLimit(); limit > 0 {
+			out.QuotaDailyLimit = &limit
+			used := a.GetQuotaDailyUsed()
+			out.QuotaDailyUsed = &used
+		}
+		if limit := a.GetQuotaWeeklyLimit(); limit > 0 {
+			out.QuotaWeeklyLimit = &limit
+			used := a.GetQuotaWeeklyUsed()
+			out.QuotaWeeklyUsed = &used
+		}
+		// 固定时间重置配置
+		if mode := a.GetQuotaDailyResetMode(); mode == "fixed" {
+			out.QuotaDailyResetMode = &mode
+			hour := a.GetQuotaDailyResetHour()
+			out.QuotaDailyResetHour = &hour
+		}
+		if mode := a.GetQuotaWeeklyResetMode(); mode == "fixed" {
+			out.QuotaWeeklyResetMode = &mode
+			day := a.GetQuotaWeeklyResetDay()
+			out.QuotaWeeklyResetDay = &day
+			hour := a.GetQuotaWeeklyResetHour()
+			out.QuotaWeeklyResetHour = &hour
+		}
+		if a.GetQuotaDailyResetMode() == "fixed" || a.GetQuotaWeeklyResetMode() == "fixed" {
+			tz := a.GetQuotaResetTimezone()
+			out.QuotaResetTimezone = &tz
+		}
+		if a.Extra != nil {
+			if v, ok := a.Extra["quota_daily_reset_at"].(string); ok && v != "" {
+				out.QuotaDailyResetAt = &v
+			}
+			if v, ok := a.Extra["quota_weekly_reset_at"].(string); ok && v != "" {
+				out.QuotaWeeklyResetAt = &v
+			}
+		}
+	}
+
 	return out
 }
 
@@ -461,6 +521,7 @@ func usageLogFromServiceUser(l *service.UsageLog) UsageLog {
 		AccountID:             l.AccountID,
 		RequestID:             l.RequestID,
 		Model:                 l.Model,
+		ServiceTier:           l.ServiceTier,
 		ReasoningEffort:       l.ReasoningEffort,
 		GroupID:               l.GroupID,
 		SubscriptionID:        l.SubscriptionID,

backend/internal/handler/dto/mappers_usage_test.go

@@ -71,3 +71,29 @@ func TestRequestTypeStringPtrNil(t *testing.T) {
 	t.Parallel()
 	require.Nil(t, requestTypeStringPtr(nil))
 }
+
+func TestUsageLogFromService_IncludesServiceTierForUserAndAdmin(t *testing.T) {
+	t.Parallel()
+
+	serviceTier := "priority"
+	log := &service.UsageLog{
+		RequestID:             "req_3",
+		Model:                 "gpt-5.4",
+		ServiceTier:           &serviceTier,
+		AccountRateMultiplier: f64Ptr(1.5),
+	}
+
+	userDTO := UsageLogFromService(log)
+	adminDTO := UsageLogFromServiceAdmin(log)
+
+	require.NotNil(t, userDTO.ServiceTier)
+	require.Equal(t, serviceTier, *userDTO.ServiceTier)
+	require.NotNil(t, adminDTO.ServiceTier)
+	require.Equal(t, serviceTier, *adminDTO.ServiceTier)
+	require.NotNil(t, adminDTO.AccountRateMultiplier)
+	require.InDelta(t, 1.5, *adminDTO.AccountRateMultiplier, 1e-12)
+}
+
+func f64Ptr(value float64) *float64 {
+	return &value
+}

backend/internal/handler/dto/settings.go

@@ -81,6 +81,9 @@ type SystemSettings struct {
 
 	// 分组隔离
 	AllowUngroupedKeyScheduling bool `json:"allow_ungrouped_key_scheduling"`
+
+	// Backend Mode
+	BackendModeEnabled bool `json:"backend_mode_enabled"`
 }
 
 type DefaultSubscriptionSetting struct {
@@ -111,6 +114,7 @@ type PublicSettings struct {
 	CustomMenuItems                  []CustomMenuItem `json:"custom_menu_items"`
 	LinuxDoOAuthEnabled              bool             `json:"linuxdo_oauth_enabled"`
 	SoraClientEnabled                bool             `json:"sora_client_enabled"`
+	BackendModeEnabled               bool             `json:"backend_mode_enabled"`
 	Version                          string           `json:"version"`
 }
 
@@ -161,6 +165,26 @@ type StreamTimeoutSettings struct {
 	ThresholdWindowMinutes int    `json:"threshold_window_minutes"`
 }
 
+// RectifierSettings 请求整流器配置 DTO
+type RectifierSettings struct {
+	Enabled                  bool `json:"enabled"`
+	ThinkingSignatureEnabled bool `json:"thinking_signature_enabled"`
+	ThinkingBudgetEnabled    bool `json:"thinking_budget_enabled"`
+}
+
+// BetaPolicyRule Beta 策略规则 DTO
+type BetaPolicyRule struct {
+	BetaToken    string `json:"beta_token"`
+	Action       string `json:"action"`
+	Scope        string `json:"scope"`
+	ErrorMessage string `json:"error_message,omitempty"`
+}
+
+// BetaPolicySettings Beta 策略配置 DTO
+type BetaPolicySettings struct {
+	Rules []BetaPolicyRule `json:"rules"`
+}
+
 // ParseCustomMenuItems parses a JSON string into a slice of CustomMenuItem.
 // Returns empty slice on empty/invalid input.
 func ParseCustomMenuItems(raw string) []CustomMenuItem {

backend/internal/handler/dto/types.go

@@ -57,6 +57,9 @@ type APIKey struct {
 	Window5hStart *time.Time `json:"window_5h_start"`
 	Window1dStart *time.Time `json:"window_1d_start"`
 	Window7dStart *time.Time `json:"window_7d_start"`
+	Reset5hAt     *time.Time `json:"reset_5h_at,omitempty"`
+	Reset1dAt     *time.Time `json:"reset_1d_at,omitempty"`
+	Reset7dAt     *time.Time `json:"reset_7d_at,omitempty"`
 
 	User  *User  `json:"user,omitempty"`
 	Group *Group `json:"group,omitempty"`
@@ -96,6 +99,9 @@ type Group struct {
 	// Sora 存储配额
 	SoraStorageQuotaBytes int64 `json:"sora_storage_quota_bytes"`
 
+	// OpenAI Messages 调度开关（用户侧需要此字段判断是否展示 Claude Code 教程）
+	AllowMessagesDispatch bool `json:"allow_messages_dispatch"`
+
 	CreatedAt time.Time `json:"created_at"`
 	UpdatedAt time.Time `json:"updated_at"`
 }
@@ -112,6 +118,9 @@ type AdminGroup struct {
 	// MCP XML 协议注入（仅 antigravity 平台使用）
 	MCPXMLInject bool `json:"mcp_xml_inject"`
 
+	// OpenAI Messages 调度配置（仅 openai 平台使用）
+	DefaultMappedModel string `json:"default_mapped_model"`
+
 	// 支持的模型系列（仅 antigravity 平台使用）
 	SupportedModelScopes []string       `json:"supported_model_scopes"`
 	AccountGroups        []AccountGroup `json:"account_groups,omitempty"`
@@ -131,6 +140,7 @@ type Account struct {
 	Extra              map[string]any `json:"extra"`
 	ProxyID            *int64         `json:"proxy_id"`
 	Concurrency        int            `json:"concurrency"`
+	LoadFactor         *int           `json:"load_factor,omitempty"`
 	Priority           int            `json:"priority"`
 	RateMultiplier     float64        `json:"rate_multiplier"`
 	Status             string         `json:"status"`
@@ -185,6 +195,24 @@ type Account struct {
 	CacheTTLOverrideEnabled *bool   `json:"cache_ttl_override_enabled,omitempty"`
 	CacheTTLOverrideTarget  *string `json:"cache_ttl_override_target,omitempty"`
 
+	// API Key 账号配额限制
+	QuotaLimit       *float64 `json:"quota_limit,omitempty"`
+	QuotaUsed        *float64 `json:"quota_used,omitempty"`
+	QuotaDailyLimit  *float64 `json:"quota_daily_limit,omitempty"`
+	QuotaDailyUsed   *float64 `json:"quota_daily_used,omitempty"`
+	QuotaWeeklyLimit *float64 `json:"quota_weekly_limit,omitempty"`
+	QuotaWeeklyUsed  *float64 `json:"quota_weekly_used,omitempty"`
+
+	// 配额固定时间重置配置
+	QuotaDailyResetMode  *string `json:"quota_daily_reset_mode,omitempty"`
+	QuotaDailyResetHour  *int    `json:"quota_daily_reset_hour,omitempty"`
+	QuotaWeeklyResetMode *string `json:"quota_weekly_reset_mode,omitempty"`
+	QuotaWeeklyResetDay  *int    `json:"quota_weekly_reset_day,omitempty"`
+	QuotaWeeklyResetHour *int    `json:"quota_weekly_reset_hour,omitempty"`
+	QuotaResetTimezone   *string `json:"quota_reset_timezone,omitempty"`
+	QuotaDailyResetAt    *string `json:"quota_daily_reset_at,omitempty"`
+	QuotaWeeklyResetAt   *string `json:"quota_weekly_reset_at,omitempty"`
+
 	Proxy         *Proxy         `json:"proxy,omitempty"`
 	AccountGroups []AccountGroup `json:"account_groups,omitempty"`
 
@@ -304,6 +332,8 @@ type UsageLog struct {
 	AccountID int64  `json:"account_id"`
 	RequestID string `json:"request_id"`
 	Model     string `json:"model"`
+	// ServiceTier records the OpenAI service tier used for billing, e.g. "priority" / "flex".
+	ServiceTier *string `json:"service_tier,omitempty"`
 	// ReasoningEffort is the request's reasoning effort level (OpenAI Responses API).
 	// nil means not provided / not applicable.
 	ReasoningEffort *string `json:"reasoning_effort,omitempty"`

backend/internal/handler/failover_loop.go

@@ -30,7 +30,7 @@ const (
 
 const (
 	// maxSameAccountRetries 同账号重试次数上限（针对 RetryableOnSameAccount 错误）
-	maxSameAccountRetries = 2
+	maxSameAccountRetries = 3
 	// sameAccountRetryDelay 同账号重试间隔
 	sameAccountRetryDelay = 500 * time.Millisecond
 	// singleAccountBackoffDelay 单账号分组 503 退避重试固定延时。

backend/internal/handler/failover_loop_test.go

@@ -291,35 +291,31 @@ func TestHandleFailoverError_SameAccountRetry(t *testing.T) {
 		require.Less(t, elapsed, 2*time.Second)
 	})
 
-	t.Run("第二次重试仍返回FailoverContinue", func(t *testing.T) {
+	t.Run("达到最大重试次数前均返回FailoverContinue", func(t *testing.T) {
 		mock := &mockTempUnscheduler{}
 		fs := NewFailoverState(3, false)
 		err := newTestFailoverErr(400, true, false)
 
-		// 第一次
-		action := fs.HandleFailoverError(context.Background(), mock, 100, "openai", err)
-		require.Equal(t, FailoverContinue, action)
-		require.Equal(t, 1, fs.SameAccountRetryCount[100])
+		for i := 1; i <= maxSameAccountRetries; i++ {
+			action := fs.HandleFailoverError(context.Background(), mock, 100, "openai", err)
+			require.Equal(t, FailoverContinue, action)
+			require.Equal(t, i, fs.SameAccountRetryCount[100])
+		}
 
-		// 第二次
-		action = fs.HandleFailoverError(context.Background(), mock, 100, "openai", err)
-		require.Equal(t, FailoverContinue, action)
-		require.Equal(t, 2, fs.SameAccountRetryCount[100])
-
-		require.Empty(t, mock.calls, "两次重试期间均不应调用 TempUnschedule")
+		require.Empty(t, mock.calls, "达到最大重试次数前均不应调用 TempUnschedule")
 	})
 
-	t.Run("第三次重试耗尽_触发TempUnschedule并切换", func(t *testing.T) {
+	t.Run("超过最大重试次数后触发TempUnschedule并切换", func(t *testing.T) {
 		mock := &mockTempUnscheduler{}
 		fs := NewFailoverState(3, false)
 		err := newTestFailoverErr(400, true, false)
 
-		// 第一次、第二次重试
-		fs.HandleFailoverError(context.Background(), mock, 100, "openai", err)
-		fs.HandleFailoverError(context.Background(), mock, 100, "openai", err)
-		require.Equal(t, 2, fs.SameAccountRetryCount[100])
+		for i := 0; i < maxSameAccountRetries; i++ {
+			fs.HandleFailoverError(context.Background(), mock, 100, "openai", err)
+		}
+		require.Equal(t, maxSameAccountRetries, fs.SameAccountRetryCount[100])
 
-		// 第三次：重试已达到 maxSameAccountRetries(2)，应切换账号
+		// 第 maxSameAccountRetries+1 次：重试耗尽，应切换账号
 		action := fs.HandleFailoverError(context.Background(), mock, 100, "openai", err)
 		require.Equal(t, FailoverContinue, action)
 		require.Equal(t, 1, fs.SwitchCount)
@@ -354,13 +350,14 @@ func TestHandleFailoverError_SameAccountRetry(t *testing.T) {
 		err := newTestFailoverErr(400, true, false)
 
 		// 耗尽账号 100 的重试
-		fs.HandleFailoverError(context.Background(), mock, 100, "openai", err)
-		fs.HandleFailoverError(context.Background(), mock, 100, "openai", err)
-		// 第三次: 重试耗尽 → 切换
+		for i := 0; i < maxSameAccountRetries; i++ {
+			fs.HandleFailoverError(context.Background(), mock, 100, "openai", err)
+		}
+		// 第 maxSameAccountRetries+1 次: 重试耗尽 → 切换
 		action := fs.HandleFailoverError(context.Background(), mock, 100, "openai", err)
 		require.Equal(t, FailoverContinue, action)
 
-		// 再次遇到账号 100，计数仍为 2，条件不满足 → 直接切换
+		// 再次遇到账号 100，计数仍为 maxSameAccountRetries，条件不满足 → 直接切换
 		action = fs.HandleFailoverError(context.Background(), mock, 100, "openai", err)
 		require.Equal(t, FailoverContinue, action)
 		require.Len(t, mock.calls, 2, "第二次耗尽也应调用 TempUnschedule")
@@ -386,9 +383,10 @@ func TestHandleFailoverError_TempUnschedule(t *testing.T) {
 		fs := NewFailoverState(3, false)
 		err := newTestFailoverErr(502, true, false)
 
-		// 耗尽重试
-		fs.HandleFailoverError(context.Background(), mock, 42, "openai", err)
-		fs.HandleFailoverError(context.Background(), mock, 42, "openai", err)
+		for i := 0; i < maxSameAccountRetries; i++ {
+			fs.HandleFailoverError(context.Background(), mock, 42, "openai", err)
+		}
+		// 再次触发时才会执行 TempUnschedule + 切换
 		fs.HandleFailoverError(context.Background(), mock, 42, "openai", err)
 
 		require.Len(t, mock.calls, 1)
@@ -521,17 +519,16 @@ func TestHandleFailoverError_IntegrationScenario(t *testing.T) {
 		mock := &mockTempUnscheduler{}
 		fs := NewFailoverState(3, true) // hasBoundSession=true
 
-		// 1. 账号 100 遇到可重试错误，同账号重试 2 次
+		// 1. 账号 100 遇到可重试错误，同账号重试 maxSameAccountRetries 次
 		retryErr := newTestFailoverErr(400, true, false)
-		action := fs.HandleFailoverError(context.Background(), mock, 100, "openai", retryErr)
-		require.Equal(t, FailoverContinue, action)
+		for i := 0; i < maxSameAccountRetries; i++ {
+			action := fs.HandleFailoverError(context.Background(), mock, 100, "openai", retryErr)
+			require.Equal(t, FailoverContinue, action)
+		}
 		require.True(t, fs.ForceCacheBilling, "hasBoundSession=true 应设置 ForceCacheBilling")
 
-		action = fs.HandleFailoverError(context.Background(), mock, 100, "openai", retryErr)
-		require.Equal(t, FailoverContinue, action)
-
-		// 2. 账号 100 重试耗尽 → TempUnschedule + 切换
-		action = fs.HandleFailoverError(context.Background(), mock, 100, "openai", retryErr)
+		// 2. 账号 100 超过重试上限 → TempUnschedule + 切换
+		action := fs.HandleFailoverError(context.Background(), mock, 100, "openai", retryErr)
 		require.Equal(t, FailoverContinue, action)
 		require.Equal(t, 1, fs.SwitchCount)
 		require.Len(t, mock.calls, 1)

backend/internal/handler/gateway_handler.go

@@ -434,19 +434,21 @@ func (h *GatewayHandler) Messages(c *gin.Context) {
 			// 捕获请求信息（用于异步记录，避免在 goroutine 中访问 gin.Context）
 			userAgent := c.GetHeader("User-Agent")
 			clientIP := ip.GetClientIP(c)
+			requestPayloadHash := service.HashUsageRequestPayload(body)
 
 			// 使用量记录通过有界 worker 池提交，避免请求热路径创建无界 goroutine。
 			h.submitUsageRecordTask(func(ctx context.Context) {
 				if err := h.gatewayService.RecordUsage(ctx, &service.RecordUsageInput{
-					Result:            result,
-					APIKey:            apiKey,
-					User:              apiKey.User,
-					Account:           account,
-					Subscription:      subscription,
-					UserAgent:         userAgent,
-					IPAddress:         clientIP,
-					ForceCacheBilling: fs.ForceCacheBilling,
-					APIKeyService:     h.apiKeyService,
+					Result:             result,
+					APIKey:             apiKey,
+					User:               apiKey.User,
+					Account:            account,
+					Subscription:       subscription,
+					UserAgent:          userAgent,
+					IPAddress:          clientIP,
+					RequestPayloadHash: requestPayloadHash,
+					ForceCacheBilling:  fs.ForceCacheBilling,
+					APIKeyService:      h.apiKeyService,
 				}); err != nil {
 					logger.L().With(
 						zap.String("component", "handler.gateway.messages"),
@@ -652,6 +654,13 @@ func (h *GatewayHandler) Messages(c *gin.Context) {
 				accountReleaseFunc()
 			}
 			if err != nil {
+				// Beta policy block: return 400 immediately, no failover
+				var betaBlockedErr *service.BetaBlockedError
+				if errors.As(err, &betaBlockedErr) {
+					h.errorResponse(c, http.StatusBadRequest, "invalid_request_error", betaBlockedErr.Message)
+					return
+				}
+
 				var promptTooLongErr *service.PromptTooLongError
 				if errors.As(err, &promptTooLongErr) {
 					reqLog.Warn("gateway.prompt_too_long_from_antigravity",
@@ -729,19 +738,21 @@ func (h *GatewayHandler) Messages(c *gin.Context) {
 			// 捕获请求信息（用于异步记录，避免在 goroutine 中访问 gin.Context）
 			userAgent := c.GetHeader("User-Agent")
 			clientIP := ip.GetClientIP(c)
+			requestPayloadHash := service.HashUsageRequestPayload(body)
 
 			// 使用量记录通过有界 worker 池提交，避免请求热路径创建无界 goroutine。
 			h.submitUsageRecordTask(func(ctx context.Context) {
 				if err := h.gatewayService.RecordUsage(ctx, &service.RecordUsageInput{
-					Result:            result,
-					APIKey:            currentAPIKey,
-					User:              currentAPIKey.User,
-					Account:           account,
-					Subscription:      currentSubscription,
-					UserAgent:         userAgent,
-					IPAddress:         clientIP,
-					ForceCacheBilling: fs.ForceCacheBilling,
-					APIKeyService:     h.apiKeyService,
+					Result:             result,
+					APIKey:             currentAPIKey,
+					User:               currentAPIKey.User,
+					Account:            account,
+					Subscription:       currentSubscription,
+					UserAgent:          userAgent,
+					IPAddress:          clientIP,
+					RequestPayloadHash: requestPayloadHash,
+					ForceCacheBilling:  fs.ForceCacheBilling,
+					APIKeyService:      h.apiKeyService,
 				}); err != nil {
 					logger.L().With(
 						zap.String("component", "handler.gateway.messages"),
@@ -971,34 +982,46 @@ func (h *GatewayHandler) usageQuotaLimited(c *gin.Context, ctx context.Context,
 		if err == nil && rateLimitData != nil {
 			var rateLimits []gin.H
 			if apiKey.RateLimit5h > 0 {
-				used := rateLimitData.Usage5h
-				rateLimits = append(rateLimits, gin.H{
+				used := rateLimitData.EffectiveUsage5h()
+				entry := gin.H{
 					"window":       "5h",
 					"limit":        apiKey.RateLimit5h,
 					"used":         used,
 					"remaining":    max(0, apiKey.RateLimit5h-used),
 					"window_start": rateLimitData.Window5hStart,
-				})
+				}
+				if rateLimitData.Window5hStart != nil && !service.IsWindowExpired(rateLimitData.Window5hStart, service.RateLimitWindow5h) {
+					entry["reset_at"] = rateLimitData.Window5hStart.Add(service.RateLimitWindow5h)
+				}
+				rateLimits = append(rateLimits, entry)
 			}
 			if apiKey.RateLimit1d > 0 {
-				used := rateLimitData.Usage1d
-				rateLimits = append(rateLimits, gin.H{
+				used := rateLimitData.EffectiveUsage1d()
+				entry := gin.H{
 					"window":       "1d",
 					"limit":        apiKey.RateLimit1d,
 					"used":         used,
 					"remaining":    max(0, apiKey.RateLimit1d-used),
 					"window_start": rateLimitData.Window1dStart,
-				})
+				}
+				if rateLimitData.Window1dStart != nil && !service.IsWindowExpired(rateLimitData.Window1dStart, service.RateLimitWindow1d) {
+					entry["reset_at"] = rateLimitData.Window1dStart.Add(service.RateLimitWindow1d)
+				}
+				rateLimits = append(rateLimits, entry)
 			}
 			if apiKey.RateLimit7d > 0 {
-				used := rateLimitData.Usage7d
-				rateLimits = append(rateLimits, gin.H{
+				used := rateLimitData.EffectiveUsage7d()
+				entry := gin.H{
 					"window":       "7d",
 					"limit":        apiKey.RateLimit7d,
 					"used":         used,
 					"remaining":    max(0, apiKey.RateLimit7d-used),
 					"window_start": rateLimitData.Window7dStart,
-				})
+				}
+				if rateLimitData.Window7dStart != nil && !service.IsWindowExpired(rateLimitData.Window7dStart, service.RateLimitWindow7d) {
+					entry["reset_at"] = rateLimitData.Window7dStart.Add(service.RateLimitWindow7d)
+				}
+				rateLimits = append(rateLimits, entry)
 			}
 			if len(rateLimits) > 0 {
 				resp["rate_limits"] = rateLimits

backend/internal/handler/gateway_handler_warmup_intercept_unit_test.go

@@ -127,6 +127,7 @@ func (f *fakeConcurrencyCache) GetAccountConcurrencyBatch(_ context.Context, acc
 	return result, nil
 }
 func (f *fakeConcurrencyCache) CleanupExpiredAccountSlots(context.Context, int64) error { return nil }
+func (f *fakeConcurrencyCache) CleanupStaleProcessSlots(context.Context, string) error  { return nil }
 
 func newTestGatewayHandler(t *testing.T, group *service.Group, accounts []*service.Account) (*GatewayHandler, func()) {
 	t.Helper()
@@ -138,6 +139,7 @@ func newTestGatewayHandler(t *testing.T, group *service.Group, accounts []*servi
 		nil, // accountRepo (not used: scheduler snapshot hit)
 		&fakeGroupRepo{group: group},
 		nil, // usageLogRepo
+		nil, // usageBillingRepo
 		nil, // userRepo
 		nil, // userSubRepo
 		nil, // userGroupRateRepo
@@ -155,6 +157,7 @@ func newTestGatewayHandler(t *testing.T, group *service.Group, accounts []*servi
 		nil, // sessionLimitCache
 		nil, // rpmCache
 		nil, // digestStore
+		nil, // settingService
 	)
 
 	// RunModeSimple：跳过计费检查，避免引入 repo/cache 依赖。

backend/internal/handler/gateway_helper_fastpath_test.go

@@ -89,6 +89,10 @@ func (m *concurrencyCacheMock) CleanupExpiredAccountSlots(ctx context.Context, a
 	return nil
 }
 
+func (m *concurrencyCacheMock) CleanupStaleProcessSlots(ctx context.Context, activeRequestPrefix string) error {
+	return nil
+}
+
 func TestConcurrencyHelper_TryAcquireUserSlot(t *testing.T) {
 	cache := &concurrencyCacheMock{
 		acquireUserSlotFn: func(ctx context.Context, userID int64, maxConcurrency int, requestID string) (bool, error) {

backend/internal/handler/gateway_helper_hotpath_test.go

@@ -120,6 +120,10 @@ func (s *helperConcurrencyCacheStub) CleanupExpiredAccountSlots(ctx context.Cont
 	return nil
 }
 
+func (s *helperConcurrencyCacheStub) CleanupStaleProcessSlots(ctx context.Context, activeRequestPrefix string) error {
+	return nil
+}
+
 func newHelperTestContext(method, path string) (*gin.Context, *httptest.ResponseRecorder) {
 	gin.SetMode(gin.TestMode)
 	rec := httptest.NewRecorder()

backend/internal/handler/gemini_v1beta_handler.go

@@ -503,6 +503,7 @@ func (h *GatewayHandler) GeminiV1BetaModels(c *gin.Context) {
 		}
 
 		// 使用量记录通过有界 worker 池提交，避免请求热路径创建无界 goroutine。
+		requestPayloadHash := service.HashUsageRequestPayload(body)
 		h.submitUsageRecordTask(func(ctx context.Context) {
 			if err := h.gatewayService.RecordUsageWithLongContext(ctx, &service.RecordUsageLongContextInput{
 				Result:                result,
@@ -512,6 +513,7 @@ func (h *GatewayHandler) GeminiV1BetaModels(c *gin.Context) {
 				Subscription:          subscription,
 				UserAgent:             userAgent,
 				IPAddress:             clientIP,
+				RequestPayloadHash:    requestPayloadHash,
 				LongContextThreshold:  200000, // Gemini 200K 阈值
 				LongContextMultiplier: 2.0,    // 超出部分双倍计费
 				ForceCacheBilling:     fs.ForceCacheBilling,

backend/internal/handler/handler.go

@@ -12,6 +12,7 @@ type AdminHandlers struct {
 	Account          *admin.AccountHandler
 	Announcement     *admin.AnnouncementHandler
 	DataManagement   *admin.DataManagementHandler
+	Backup           *admin.BackupHandler
 	OAuth            *admin.OAuthHandler
 	OpenAIOAuth      *admin.OpenAIOAuthHandler
 	GeminiOAuth      *admin.GeminiOAuthHandler

backend/internal/handler/openai_chat_completions.go

@@ -0,0 +1,284 @@
+package handler
+
+import (
+	"context"
+	"errors"
+	"net/http"
+	"time"
+
+	pkghttputil "github.com/Wei-Shaw/sub2api/internal/pkg/httputil"
+	"github.com/Wei-Shaw/sub2api/internal/pkg/ip"
+	"github.com/Wei-Shaw/sub2api/internal/pkg/logger"
+	middleware2 "github.com/Wei-Shaw/sub2api/internal/server/middleware"
+	"github.com/Wei-Shaw/sub2api/internal/service"
+	"github.com/gin-gonic/gin"
+	"github.com/tidwall/gjson"
+	"go.uber.org/zap"
+)
+
+// ChatCompletions handles OpenAI Chat Completions API requests.
+// POST /v1/chat/completions
+func (h *OpenAIGatewayHandler) ChatCompletions(c *gin.Context) {
+	streamStarted := false
+	defer h.recoverResponsesPanic(c, &streamStarted)
+
+	requestStart := time.Now()
+
+	apiKey, ok := middleware2.GetAPIKeyFromContext(c)
+	if !ok {
+		h.errorResponse(c, http.StatusUnauthorized, "authentication_error", "Invalid API key")
+		return
+	}
+
+	subject, ok := middleware2.GetAuthSubjectFromContext(c)
+	if !ok {
+		h.errorResponse(c, http.StatusInternalServerError, "api_error", "User context not found")
+		return
+	}
+	reqLog := requestLogger(
+		c,
+		"handler.openai_gateway.chat_completions",
+		zap.Int64("user_id", subject.UserID),
+		zap.Int64("api_key_id", apiKey.ID),
+		zap.Any("group_id", apiKey.GroupID),
+	)
+
+	if !h.ensureResponsesDependencies(c, reqLog) {
+		return
+	}
+
+	body, err := pkghttputil.ReadRequestBodyWithPrealloc(c.Request)
+	if err != nil {
+		if maxErr, ok := extractMaxBytesError(err); ok {
+			h.errorResponse(c, http.StatusRequestEntityTooLarge, "invalid_request_error", buildBodyTooLargeMessage(maxErr.Limit))
+			return
+		}
+		h.errorResponse(c, http.StatusBadRequest, "invalid_request_error", "Failed to read request body")
+		return
+	}
+	if len(body) == 0 {
+		h.errorResponse(c, http.StatusBadRequest, "invalid_request_error", "Request body is empty")
+		return
+	}
+
+	if !gjson.ValidBytes(body) {
+		h.errorResponse(c, http.StatusBadRequest, "invalid_request_error", "Failed to parse request body")
+		return
+	}
+
+	modelResult := gjson.GetBytes(body, "model")
+	if !modelResult.Exists() || modelResult.Type != gjson.String || modelResult.String() == "" {
+		h.errorResponse(c, http.StatusBadRequest, "invalid_request_error", "model is required")
+		return
+	}
+	reqModel := modelResult.String()
+	reqStream := gjson.GetBytes(body, "stream").Bool()
+
+	reqLog = reqLog.With(zap.String("model", reqModel), zap.Bool("stream", reqStream))
+
+	setOpsRequestContext(c, reqModel, reqStream, body)
+
+	if h.errorPassthroughService != nil {
+		service.BindErrorPassthroughService(c, h.errorPassthroughService)
+	}
+
+	subscription, _ := middleware2.GetSubscriptionFromContext(c)
+
+	service.SetOpsLatencyMs(c, service.OpsAuthLatencyMsKey, time.Since(requestStart).Milliseconds())
+	routingStart := time.Now()
+
+	userReleaseFunc, acquired := h.acquireResponsesUserSlot(c, subject.UserID, subject.Concurrency, reqStream, &streamStarted, reqLog)
+	if !acquired {
+		return
+	}
+	if userReleaseFunc != nil {
+		defer userReleaseFunc()
+	}
+
+	if err := h.billingCacheService.CheckBillingEligibility(c.Request.Context(), apiKey.User, apiKey, apiKey.Group, subscription); err != nil {
+		reqLog.Info("openai_chat_completions.billing_eligibility_check_failed", zap.Error(err))
+		status, code, message := billingErrorDetails(err)
+		h.handleStreamingAwareError(c, status, code, message, streamStarted)
+		return
+	}
+
+	sessionHash := h.gatewayService.GenerateSessionHash(c, body)
+	promptCacheKey := h.gatewayService.ExtractSessionID(c, body)
+
+	maxAccountSwitches := h.maxAccountSwitches
+	switchCount := 0
+	failedAccountIDs := make(map[int64]struct{})
+	sameAccountRetryCount := make(map[int64]int)
+	var lastFailoverErr *service.UpstreamFailoverError
+
+	for {
+		c.Set("openai_chat_completions_fallback_model", "")
+		reqLog.Debug("openai_chat_completions.account_selecting", zap.Int("excluded_account_count", len(failedAccountIDs)))
+		selection, scheduleDecision, err := h.gatewayService.SelectAccountWithScheduler(
+			c.Request.Context(),
+			apiKey.GroupID,
+			"",
+			sessionHash,
+			reqModel,
+			failedAccountIDs,
+			service.OpenAIUpstreamTransportAny,
+		)
+		if err != nil {
+			reqLog.Warn("openai_chat_completions.account_select_failed",
+				zap.Error(err),
+				zap.Int("excluded_account_count", len(failedAccountIDs)),
+			)
+			if len(failedAccountIDs) == 0 {
+				defaultModel := ""
+				if apiKey.Group != nil {
+					defaultModel = apiKey.Group.DefaultMappedModel
+				}
+				if defaultModel != "" && defaultModel != reqModel {
+					reqLog.Info("openai_chat_completions.fallback_to_default_model",
+						zap.String("default_mapped_model", defaultModel),
+					)
+					selection, scheduleDecision, err = h.gatewayService.SelectAccountWithScheduler(
+						c.Request.Context(),
+						apiKey.GroupID,
+						"",
+						sessionHash,
+						defaultModel,
+						failedAccountIDs,
+						service.OpenAIUpstreamTransportAny,
+					)
+					if err == nil && selection != nil {
+						c.Set("openai_chat_completions_fallback_model", defaultModel)
+					}
+				}
+				if err != nil {
+					h.handleStreamingAwareError(c, http.StatusServiceUnavailable, "api_error", "Service temporarily unavailable", streamStarted)
+					return
+				}
+			} else {
+				if lastFailoverErr != nil {
+					h.handleFailoverExhausted(c, lastFailoverErr, streamStarted)
+				} else {
+					h.handleStreamingAwareError(c, http.StatusBadGateway, "api_error", "Upstream request failed", streamStarted)
+				}
+				return
+			}
+		}
+		if selection == nil || selection.Account == nil {
+			h.handleStreamingAwareError(c, http.StatusServiceUnavailable, "api_error", "No available accounts", streamStarted)
+			return
+		}
+		account := selection.Account
+		sessionHash = ensureOpenAIPoolModeSessionHash(sessionHash, account)
+		reqLog.Debug("openai_chat_completions.account_selected", zap.Int64("account_id", account.ID), zap.String("account_name", account.Name))
+		_ = scheduleDecision
+		setOpsSelectedAccount(c, account.ID, account.Platform)
+
+		accountReleaseFunc, acquired := h.acquireResponsesAccountSlot(c, apiKey.GroupID, sessionHash, selection, reqStream, &streamStarted, reqLog)
+		if !acquired {
+			return
+		}
+
+		service.SetOpsLatencyMs(c, service.OpsRoutingLatencyMsKey, time.Since(routingStart).Milliseconds())
+		forwardStart := time.Now()
+
+		defaultMappedModel := c.GetString("openai_chat_completions_fallback_model")
+		result, err := h.gatewayService.ForwardAsChatCompletions(c.Request.Context(), c, account, body, promptCacheKey, defaultMappedModel)
+
+		forwardDurationMs := time.Since(forwardStart).Milliseconds()
+		if accountReleaseFunc != nil {
+			accountReleaseFunc()
+		}
+		upstreamLatencyMs, _ := getContextInt64(c, service.OpsUpstreamLatencyMsKey)
+		responseLatencyMs := forwardDurationMs
+		if upstreamLatencyMs > 0 && forwardDurationMs > upstreamLatencyMs {
+			responseLatencyMs = forwardDurationMs - upstreamLatencyMs
+		}
+		service.SetOpsLatencyMs(c, service.OpsResponseLatencyMsKey, responseLatencyMs)
+		if err == nil && result != nil && result.FirstTokenMs != nil {
+			service.SetOpsLatencyMs(c, service.OpsTimeToFirstTokenMsKey, int64(*result.FirstTokenMs))
+		}
+		if err != nil {
+			var failoverErr *service.UpstreamFailoverError
+			if errors.As(err, &failoverErr) {
+				h.gatewayService.ReportOpenAIAccountScheduleResult(account.ID, false, nil)
+				// Pool mode: retry on the same account
+				if failoverErr.RetryableOnSameAccount {
+					retryLimit := account.GetPoolModeRetryCount()
+					if sameAccountRetryCount[account.ID] < retryLimit {
+						sameAccountRetryCount[account.ID]++
+						reqLog.Warn("openai_chat_completions.pool_mode_same_account_retry",
+							zap.Int64("account_id", account.ID),
+							zap.Int("upstream_status", failoverErr.StatusCode),
+							zap.Int("retry_limit", retryLimit),
+							zap.Int("retry_count", sameAccountRetryCount[account.ID]),
+						)
+						select {
+						case <-c.Request.Context().Done():
+							return
+						case <-time.After(sameAccountRetryDelay):
+						}
+						continue
+					}
+				}
+				h.gatewayService.RecordOpenAIAccountSwitch()
+				failedAccountIDs[account.ID] = struct{}{}
+				lastFailoverErr = failoverErr
+				if switchCount >= maxAccountSwitches {
+					h.handleFailoverExhausted(c, failoverErr, streamStarted)
+					return
+				}
+				switchCount++
+				reqLog.Warn("openai_chat_completions.upstream_failover_switching",
+					zap.Int64("account_id", account.ID),
+					zap.Int("upstream_status", failoverErr.StatusCode),
+					zap.Int("switch_count", switchCount),
+					zap.Int("max_switches", maxAccountSwitches),
+				)
+				continue
+			}
+			h.gatewayService.ReportOpenAIAccountScheduleResult(account.ID, false, nil)
+			wroteFallback := h.ensureForwardErrorResponse(c, streamStarted)
+			reqLog.Warn("openai_chat_completions.forward_failed",
+				zap.Int64("account_id", account.ID),
+				zap.Bool("fallback_error_response_written", wroteFallback),
+				zap.Error(err),
+			)
+			return
+		}
+		if result != nil {
+			h.gatewayService.ReportOpenAIAccountScheduleResult(account.ID, true, result.FirstTokenMs)
+		} else {
+			h.gatewayService.ReportOpenAIAccountScheduleResult(account.ID, true, nil)
+		}
+
+		userAgent := c.GetHeader("User-Agent")
+		clientIP := ip.GetClientIP(c)
+
+		h.submitUsageRecordTask(func(ctx context.Context) {
+			if err := h.gatewayService.RecordUsage(ctx, &service.OpenAIRecordUsageInput{
+				Result:        result,
+				APIKey:        apiKey,
+				User:          apiKey.User,
+				Account:       account,
+				Subscription:  subscription,
+				UserAgent:     userAgent,
+				IPAddress:     clientIP,
+				APIKeyService: h.apiKeyService,
+			}); err != nil {
+				logger.L().With(
+					zap.String("component", "handler.openai_gateway.chat_completions"),
+					zap.Int64("user_id", subject.UserID),
+					zap.Int64("api_key_id", apiKey.ID),
+					zap.Any("group_id", apiKey.GroupID),
+					zap.String("model", reqModel),
+					zap.Int64("account_id", account.ID),
+				).Error("openai_chat_completions.record_usage_failed", zap.Error(err))
+			}
+		})
+		reqLog.Debug("openai_chat_completions.request_completed",
+			zap.Int64("account_id", account.ID),
+			zap.Int("switch_count", switchCount),
+		)
+		return
+	}
+}

backend/internal/handler/openai_gateway_handler.go

@@ -20,6 +20,7 @@ import (
 
 	coderws "github.com/coder/websocket"
 	"github.com/gin-gonic/gin"
+	"github.com/google/uuid"
 	"github.com/tidwall/gjson"
 	"go.uber.org/zap"
 )
@@ -118,6 +119,20 @@ func (h *OpenAIGatewayHandler) Responses(c *gin.Context) {
 	}
 
 	setOpsRequestContext(c, "", false, body)
+	sessionHashBody := body
+	if service.IsOpenAIResponsesCompactPathForTest(c) {
+		if compactSeed := strings.TrimSpace(gjson.GetBytes(body, "prompt_cache_key").String()); compactSeed != "" {
+			c.Set(service.OpenAICompactSessionSeedKeyForTest(), compactSeed)
+		}
+		normalizedCompactBody, normalizedCompact, compactErr := service.NormalizeOpenAICompactRequestBodyForTest(body)
+		if compactErr != nil {
+			h.errorResponse(c, http.StatusBadRequest, "invalid_request_error", "Failed to normalize compact request body")
+			return
+		}
+		if normalizedCompact {
+			body = normalizedCompactBody
+		}
+	}
 
 	// 校验请求体 JSON 合法性
 	if !gjson.ValidBytes(body) {
@@ -193,11 +208,12 @@ func (h *OpenAIGatewayHandler) Responses(c *gin.Context) {
 	}
 
 	// Generate session hash (header first; fallback to prompt_cache_key)
-	sessionHash := h.gatewayService.GenerateSessionHash(c, body)
+	sessionHash := h.gatewayService.GenerateSessionHash(c, sessionHashBody)
 
 	maxAccountSwitches := h.maxAccountSwitches
 	switchCount := 0
 	failedAccountIDs := make(map[int64]struct{})
+	sameAccountRetryCount := make(map[int64]int)
 	var lastFailoverErr *service.UpstreamFailoverError
 
 	for {
@@ -245,6 +261,7 @@ func (h *OpenAIGatewayHandler) Responses(c *gin.Context) {
 			zap.Float64("load_skew", scheduleDecision.LoadSkew),
 		)
 		account := selection.Account
+		sessionHash = ensureOpenAIPoolModeSessionHash(sessionHash, account)
 		reqLog.Debug("openai.account_selected", zap.Int64("account_id", account.ID), zap.String("account_name", account.Name))
 		setOpsSelectedAccount(c, account.ID, account.Platform)
 
@@ -274,6 +291,25 @@ func (h *OpenAIGatewayHandler) Responses(c *gin.Context) {
 			var failoverErr *service.UpstreamFailoverError
 			if errors.As(err, &failoverErr) {
 				h.gatewayService.ReportOpenAIAccountScheduleResult(account.ID, false, nil)
+				// 池模式：同账号重试
+				if failoverErr.RetryableOnSameAccount {
+					retryLimit := account.GetPoolModeRetryCount()
+					if sameAccountRetryCount[account.ID] < retryLimit {
+						sameAccountRetryCount[account.ID]++
+						reqLog.Warn("openai.pool_mode_same_account_retry",
+							zap.Int64("account_id", account.ID),
+							zap.Int("upstream_status", failoverErr.StatusCode),
+							zap.Int("retry_limit", retryLimit),
+							zap.Int("retry_count", sameAccountRetryCount[account.ID]),
+						)
+						select {
+						case <-c.Request.Context().Done():
+							return
+						case <-time.After(sameAccountRetryDelay):
+						}
+						continue
+					}
+				}
 				h.gatewayService.RecordOpenAIAccountSwitch()
 				failedAccountIDs[account.ID] = struct{}{}
 				lastFailoverErr = failoverErr
@@ -305,6 +341,9 @@ func (h *OpenAIGatewayHandler) Responses(c *gin.Context) {
 			return
 		}
 		if result != nil {
+			if account.Type == service.AccountTypeOAuth {
+				h.gatewayService.UpdateCodexUsageSnapshotFromHeaders(c.Request.Context(), account.ID, result.ResponseHeaders)
+			}
 			h.gatewayService.ReportOpenAIAccountScheduleResult(account.ID, true, result.FirstTokenMs)
 		} else {
 			h.gatewayService.ReportOpenAIAccountScheduleResult(account.ID, true, nil)
@@ -313,18 +352,20 @@ func (h *OpenAIGatewayHandler) Responses(c *gin.Context) {
 		// 捕获请求信息（用于异步记录，避免在 goroutine 中访问 gin.Context）
 		userAgent := c.GetHeader("User-Agent")
 		clientIP := ip.GetClientIP(c)
+		requestPayloadHash := service.HashUsageRequestPayload(body)
 
 		// 使用量记录通过有界 worker 池提交，避免请求热路径创建无界 goroutine。
 		h.submitUsageRecordTask(func(ctx context.Context) {
 			if err := h.gatewayService.RecordUsage(ctx, &service.OpenAIRecordUsageInput{
-				Result:        result,
-				APIKey:        apiKey,
-				User:          apiKey.User,
-				Account:       account,
-				Subscription:  subscription,
-				UserAgent:     userAgent,
-				IPAddress:     clientIP,
-				APIKeyService: h.apiKeyService,
+				Result:             result,
+				APIKey:             apiKey,
+				User:               apiKey.User,
+				Account:            account,
+				Subscription:       subscription,
+				UserAgent:          userAgent,
+				IPAddress:          clientIP,
+				RequestPayloadHash: requestPayloadHash,
+				APIKeyService:      h.apiKeyService,
 			}); err != nil {
 				logger.L().With(
 					zap.String("component", "handler.openai_gateway.responses"),
@@ -424,6 +465,349 @@ func (h *OpenAIGatewayHandler) logOpenAIRemoteCompactOutcome(c *gin.Context, sta
 	log.Warn("codex.remote_compact.failed")
 }
 
+// Messages handles Anthropic Messages API requests routed to OpenAI platform.
+// POST /v1/messages (when group platform is OpenAI)
+func (h *OpenAIGatewayHandler) Messages(c *gin.Context) {
+	streamStarted := false
+	defer h.recoverAnthropicMessagesPanic(c, &streamStarted)
+
+	requestStart := time.Now()
+
+	apiKey, ok := middleware2.GetAPIKeyFromContext(c)
+	if !ok {
+		h.anthropicErrorResponse(c, http.StatusUnauthorized, "authentication_error", "Invalid API key")
+		return
+	}
+
+	subject, ok := middleware2.GetAuthSubjectFromContext(c)
+	if !ok {
+		h.anthropicErrorResponse(c, http.StatusInternalServerError, "api_error", "User context not found")
+		return
+	}
+	reqLog := requestLogger(
+		c,
+		"handler.openai_gateway.messages",
+		zap.Int64("user_id", subject.UserID),
+		zap.Int64("api_key_id", apiKey.ID),
+		zap.Any("group_id", apiKey.GroupID),
+	)
+
+	// 检查分组是否允许 /v1/messages 调度
+	if apiKey.Group != nil && !apiKey.Group.AllowMessagesDispatch {
+		h.anthropicErrorResponse(c, http.StatusForbidden, "permission_error",
+			"This group does not allow /v1/messages dispatch")
+		return
+	}
+
+	if !h.ensureResponsesDependencies(c, reqLog) {
+		return
+	}
+
+	body, err := pkghttputil.ReadRequestBodyWithPrealloc(c.Request)
+	if err != nil {
+		if maxErr, ok := extractMaxBytesError(err); ok {
+			h.anthropicErrorResponse(c, http.StatusRequestEntityTooLarge, "invalid_request_error", buildBodyTooLargeMessage(maxErr.Limit))
+			return
+		}
+		h.anthropicErrorResponse(c, http.StatusBadRequest, "invalid_request_error", "Failed to read request body")
+		return
+	}
+	if len(body) == 0 {
+		h.anthropicErrorResponse(c, http.StatusBadRequest, "invalid_request_error", "Request body is empty")
+		return
+	}
+
+	if !gjson.ValidBytes(body) {
+		h.anthropicErrorResponse(c, http.StatusBadRequest, "invalid_request_error", "Failed to parse request body")
+		return
+	}
+
+	modelResult := gjson.GetBytes(body, "model")
+	if !modelResult.Exists() || modelResult.Type != gjson.String || modelResult.String() == "" {
+		h.anthropicErrorResponse(c, http.StatusBadRequest, "invalid_request_error", "model is required")
+		return
+	}
+	reqModel := modelResult.String()
+	reqStream := gjson.GetBytes(body, "stream").Bool()
+
+	reqLog = reqLog.With(zap.String("model", reqModel), zap.Bool("stream", reqStream))
+
+	setOpsRequestContext(c, reqModel, reqStream, body)
+
+	// 绑定错误透传服务，允许 service 层在非 failover 错误场景复用规则。
+	if h.errorPassthroughService != nil {
+		service.BindErrorPassthroughService(c, h.errorPassthroughService)
+	}
+
+	subscription, _ := middleware2.GetSubscriptionFromContext(c)
+
+	service.SetOpsLatencyMs(c, service.OpsAuthLatencyMsKey, time.Since(requestStart).Milliseconds())
+	routingStart := time.Now()
+
+	userReleaseFunc, acquired := h.acquireResponsesUserSlot(c, subject.UserID, subject.Concurrency, reqStream, &streamStarted, reqLog)
+	if !acquired {
+		return
+	}
+	if userReleaseFunc != nil {
+		defer userReleaseFunc()
+	}
+
+	if err := h.billingCacheService.CheckBillingEligibility(c.Request.Context(), apiKey.User, apiKey, apiKey.Group, subscription); err != nil {
+		reqLog.Info("openai_messages.billing_eligibility_check_failed", zap.Error(err))
+		status, code, message := billingErrorDetails(err)
+		h.anthropicStreamingAwareError(c, status, code, message, streamStarted)
+		return
+	}
+
+	sessionHash := h.gatewayService.GenerateSessionHash(c, body)
+	promptCacheKey := h.gatewayService.ExtractSessionID(c, body)
+
+	// Anthropic 格式的请求在 metadata.user_id 中携带 session 标识，
+	// 而非 OpenAI 的 session_id/conversation_id headers。
+	// 从中派生 sessionHash（sticky session）和 promptCacheKey（upstream cache）。
+	if sessionHash == "" || promptCacheKey == "" {
+		if userID := strings.TrimSpace(gjson.GetBytes(body, "metadata.user_id").String()); userID != "" {
+			seed := reqModel + "-" + userID
+			if promptCacheKey == "" {
+				promptCacheKey = service.GenerateSessionUUID(seed)
+			}
+			if sessionHash == "" {
+				sessionHash = service.DeriveSessionHashFromSeed(seed)
+			}
+		}
+	}
+
+	maxAccountSwitches := h.maxAccountSwitches
+	switchCount := 0
+	failedAccountIDs := make(map[int64]struct{})
+	sameAccountRetryCount := make(map[int64]int)
+	var lastFailoverErr *service.UpstreamFailoverError
+
+	for {
+		// 清除上一次迭代的降级模型标记，避免残留影响本次迭代
+		c.Set("openai_messages_fallback_model", "")
+		reqLog.Debug("openai_messages.account_selecting", zap.Int("excluded_account_count", len(failedAccountIDs)))
+		selection, scheduleDecision, err := h.gatewayService.SelectAccountWithScheduler(
+			c.Request.Context(),
+			apiKey.GroupID,
+			"", // no previous_response_id
+			sessionHash,
+			reqModel,
+			failedAccountIDs,
+			service.OpenAIUpstreamTransportAny,
+		)
+		if err != nil {
+			reqLog.Warn("openai_messages.account_select_failed",
+				zap.Error(err),
+				zap.Int("excluded_account_count", len(failedAccountIDs)),
+			)
+			// 首次调度失败 + 有默认映射模型 → 用默认模型重试
+			if len(failedAccountIDs) == 0 {
+				defaultModel := ""
+				if apiKey.Group != nil {
+					defaultModel = apiKey.Group.DefaultMappedModel
+				}
+				if defaultModel != "" && defaultModel != reqModel {
+					reqLog.Info("openai_messages.fallback_to_default_model",
+						zap.String("default_mapped_model", defaultModel),
+					)
+					selection, scheduleDecision, err = h.gatewayService.SelectAccountWithScheduler(
+						c.Request.Context(),
+						apiKey.GroupID,
+						"",
+						sessionHash,
+						defaultModel,
+						failedAccountIDs,
+						service.OpenAIUpstreamTransportAny,
+					)
+					if err == nil && selection != nil {
+						c.Set("openai_messages_fallback_model", defaultModel)
+					}
+				}
+				if err != nil {
+					h.anthropicStreamingAwareError(c, http.StatusServiceUnavailable, "api_error", "Service temporarily unavailable", streamStarted)
+					return
+				}
+			} else {
+				if lastFailoverErr != nil {
+					h.handleAnthropicFailoverExhausted(c, lastFailoverErr, streamStarted)
+				} else {
+					h.anthropicStreamingAwareError(c, http.StatusBadGateway, "api_error", "Upstream request failed", streamStarted)
+				}
+				return
+			}
+		}
+		if selection == nil || selection.Account == nil {
+			h.anthropicStreamingAwareError(c, http.StatusServiceUnavailable, "api_error", "No available accounts", streamStarted)
+			return
+		}
+		account := selection.Account
+		sessionHash = ensureOpenAIPoolModeSessionHash(sessionHash, account)
+		reqLog.Debug("openai_messages.account_selected", zap.Int64("account_id", account.ID), zap.String("account_name", account.Name))
+		_ = scheduleDecision
+		setOpsSelectedAccount(c, account.ID, account.Platform)
+
+		accountReleaseFunc, acquired := h.acquireResponsesAccountSlot(c, apiKey.GroupID, sessionHash, selection, reqStream, &streamStarted, reqLog)
+		if !acquired {
+			return
+		}
+
+		service.SetOpsLatencyMs(c, service.OpsRoutingLatencyMsKey, time.Since(routingStart).Milliseconds())
+		forwardStart := time.Now()
+
+		// 仅在调度时实际触发了降级（原模型无可用账号、改用默认模型重试成功）时，
+		// 才将降级模型传给 Forward 层做模型替换；否则保持用户请求的原始模型。
+		defaultMappedModel := c.GetString("openai_messages_fallback_model")
+		result, err := h.gatewayService.ForwardAsAnthropic(c.Request.Context(), c, account, body, promptCacheKey, defaultMappedModel)
+
+		forwardDurationMs := time.Since(forwardStart).Milliseconds()
+		if accountReleaseFunc != nil {
+			accountReleaseFunc()
+		}
+		upstreamLatencyMs, _ := getContextInt64(c, service.OpsUpstreamLatencyMsKey)
+		responseLatencyMs := forwardDurationMs
+		if upstreamLatencyMs > 0 && forwardDurationMs > upstreamLatencyMs {
+			responseLatencyMs = forwardDurationMs - upstreamLatencyMs
+		}
+		service.SetOpsLatencyMs(c, service.OpsResponseLatencyMsKey, responseLatencyMs)
+		if err == nil && result != nil && result.FirstTokenMs != nil {
+			service.SetOpsLatencyMs(c, service.OpsTimeToFirstTokenMsKey, int64(*result.FirstTokenMs))
+		}
+		if err != nil {
+			var failoverErr *service.UpstreamFailoverError
+			if errors.As(err, &failoverErr) {
+				h.gatewayService.ReportOpenAIAccountScheduleResult(account.ID, false, nil)
+				// 池模式：同账号重试
+				if failoverErr.RetryableOnSameAccount {
+					retryLimit := account.GetPoolModeRetryCount()
+					if sameAccountRetryCount[account.ID] < retryLimit {
+						sameAccountRetryCount[account.ID]++
+						reqLog.Warn("openai_messages.pool_mode_same_account_retry",
+							zap.Int64("account_id", account.ID),
+							zap.Int("upstream_status", failoverErr.StatusCode),
+							zap.Int("retry_limit", retryLimit),
+							zap.Int("retry_count", sameAccountRetryCount[account.ID]),
+						)
+						select {
+						case <-c.Request.Context().Done():
+							return
+						case <-time.After(sameAccountRetryDelay):
+						}
+						continue
+					}
+				}
+				h.gatewayService.RecordOpenAIAccountSwitch()
+				failedAccountIDs[account.ID] = struct{}{}
+				lastFailoverErr = failoverErr
+				if switchCount >= maxAccountSwitches {
+					h.handleAnthropicFailoverExhausted(c, failoverErr, streamStarted)
+					return
+				}
+				switchCount++
+				reqLog.Warn("openai_messages.upstream_failover_switching",
+					zap.Int64("account_id", account.ID),
+					zap.Int("upstream_status", failoverErr.StatusCode),
+					zap.Int("switch_count", switchCount),
+					zap.Int("max_switches", maxAccountSwitches),
+				)
+				continue
+			}
+			h.gatewayService.ReportOpenAIAccountScheduleResult(account.ID, false, nil)
+			wroteFallback := h.ensureAnthropicErrorResponse(c, streamStarted)
+			reqLog.Warn("openai_messages.forward_failed",
+				zap.Int64("account_id", account.ID),
+				zap.Bool("fallback_error_response_written", wroteFallback),
+				zap.Error(err),
+			)
+			return
+		}
+		if result != nil {
+			h.gatewayService.ReportOpenAIAccountScheduleResult(account.ID, true, result.FirstTokenMs)
+		} else {
+			h.gatewayService.ReportOpenAIAccountScheduleResult(account.ID, true, nil)
+		}
+
+		userAgent := c.GetHeader("User-Agent")
+		clientIP := ip.GetClientIP(c)
+		requestPayloadHash := service.HashUsageRequestPayload(body)
+
+		h.submitUsageRecordTask(func(ctx context.Context) {
+			if err := h.gatewayService.RecordUsage(ctx, &service.OpenAIRecordUsageInput{
+				Result:             result,
+				APIKey:             apiKey,
+				User:               apiKey.User,
+				Account:            account,
+				Subscription:       subscription,
+				UserAgent:          userAgent,
+				IPAddress:          clientIP,
+				RequestPayloadHash: requestPayloadHash,
+				APIKeyService:      h.apiKeyService,
+			}); err != nil {
+				logger.L().With(
+					zap.String("component", "handler.openai_gateway.messages"),
+					zap.Int64("user_id", subject.UserID),
+					zap.Int64("api_key_id", apiKey.ID),
+					zap.Any("group_id", apiKey.GroupID),
+					zap.String("model", reqModel),
+					zap.Int64("account_id", account.ID),
+				).Error("openai_messages.record_usage_failed", zap.Error(err))
+			}
+		})
+		reqLog.Debug("openai_messages.request_completed",
+			zap.Int64("account_id", account.ID),
+			zap.Int("switch_count", switchCount),
+		)
+		return
+	}
+}
+
+// anthropicErrorResponse writes an error in Anthropic Messages API format.
+func (h *OpenAIGatewayHandler) anthropicErrorResponse(c *gin.Context, status int, errType, message string) {
+	c.JSON(status, gin.H{
+		"type": "error",
+		"error": gin.H{
+			"type":    errType,
+			"message": message,
+		},
+	})
+}
+
+// anthropicStreamingAwareError handles errors that may occur during streaming,
+// using Anthropic SSE error format.
+func (h *OpenAIGatewayHandler) anthropicStreamingAwareError(c *gin.Context, status int, errType, message string, streamStarted bool) {
+	if streamStarted {
+		flusher, ok := c.Writer.(http.Flusher)
+		if ok {
+			errPayload, _ := json.Marshal(gin.H{
+				"type": "error",
+				"error": gin.H{
+					"type":    errType,
+					"message": message,
+				},
+			})
+			fmt.Fprintf(c.Writer, "event: error\ndata: %s\n\n", errPayload) //nolint:errcheck
+			flusher.Flush()
+		}
+		return
+	}
+	h.anthropicErrorResponse(c, status, errType, message)
+}
+
+// handleAnthropicFailoverExhausted maps upstream failover errors to Anthropic format.
+func (h *OpenAIGatewayHandler) handleAnthropicFailoverExhausted(c *gin.Context, failoverErr *service.UpstreamFailoverError, streamStarted bool) {
+	status, errType, errMsg := h.mapUpstreamError(failoverErr.StatusCode)
+	h.anthropicStreamingAwareError(c, status, errType, errMsg, streamStarted)
+}
+
+// ensureAnthropicErrorResponse writes a fallback Anthropic error if no response was written.
+func (h *OpenAIGatewayHandler) ensureAnthropicErrorResponse(c *gin.Context, streamStarted bool) bool {
+	if c == nil || c.Writer == nil || c.Writer.Written() {
+		return false
+	}
+	h.anthropicStreamingAwareError(c, http.StatusBadGateway, "api_error", "Upstream request failed", streamStarted)
+	return true
+}
+
 func (h *OpenAIGatewayHandler) validateFunctionCallOutputRequest(c *gin.Context, body []byte, reqLog *zap.Logger) bool {
 	if !gjson.GetBytes(body, `input.#(type=="function_call_output")`).Exists() {
 		return true
@@ -840,17 +1224,21 @@ func (h *OpenAIGatewayHandler) ResponsesWebSocket(c *gin.Context) {
 			if turnErr != nil || result == nil {
 				return
 			}
+			if account.Type == service.AccountTypeOAuth {
+				h.gatewayService.UpdateCodexUsageSnapshotFromHeaders(ctx, account.ID, result.ResponseHeaders)
+			}
 			h.gatewayService.ReportOpenAIAccountScheduleResult(account.ID, true, result.FirstTokenMs)
 			h.submitUsageRecordTask(func(taskCtx context.Context) {
 				if err := h.gatewayService.RecordUsage(taskCtx, &service.OpenAIRecordUsageInput{
-					Result:        result,
-					APIKey:        apiKey,
-					User:          apiKey.User,
-					Account:       account,
-					Subscription:  subscription,
-					UserAgent:     userAgent,
-					IPAddress:     clientIP,
-					APIKeyService: h.apiKeyService,
+					Result:             result,
+					APIKey:             apiKey,
+					User:               apiKey.User,
+					Account:            account,
+					Subscription:       subscription,
+					UserAgent:          userAgent,
+					IPAddress:          clientIP,
+					RequestPayloadHash: service.HashUsageRequestPayload(firstMessage),
+					APIKeyService:      h.apiKeyService,
 				}); err != nil {
 					reqLog.Error("openai.websocket_record_usage_failed",
 						zap.Int64("account_id", account.ID),
@@ -901,6 +1289,26 @@ func (h *OpenAIGatewayHandler) recoverResponsesPanic(c *gin.Context, streamStart
 	)
 }
 
+// recoverAnthropicMessagesPanic recovers from panics in the Anthropic Messages
+// handler and returns an Anthropic-formatted error response.
+func (h *OpenAIGatewayHandler) recoverAnthropicMessagesPanic(c *gin.Context, streamStarted *bool) {
+	recovered := recover()
+	if recovered == nil {
+		return
+	}
+
+	started := streamStarted != nil && *streamStarted
+	requestLogger(c, "handler.openai_gateway.messages").Error(
+		"openai.messages_panic_recovered",
+		zap.Bool("stream_started", started),
+		zap.Any("panic", recovered),
+		zap.ByteString("stack", debug.Stack()),
+	)
+	if !started {
+		h.anthropicErrorResponse(c, http.StatusInternalServerError, "api_error", "Internal server error")
+	}
+}
+
 func (h *OpenAIGatewayHandler) ensureResponsesDependencies(c *gin.Context, reqLog *zap.Logger) bool {
 	missing := h.missingResponsesDependencies()
 	if len(missing) == 0 {
@@ -1106,6 +1514,14 @@ func setOpenAIClientTransportWS(c *gin.Context) {
 	service.SetOpenAIClientTransport(c, service.OpenAIClientTransportWS)
 }
 
+func ensureOpenAIPoolModeSessionHash(sessionHash string, account *service.Account) string {
+	if sessionHash != "" || account == nil || !account.IsPoolMode() {
+		return sessionHash
+	}
+	// 为当前请求生成一次性粘性会话键，确保同账号重试不会重新负载均衡到其他账号。
+	return "openai-pool-retry-" + uuid.NewString()
+}
+
 func openAIWSIngressFallbackSessionSeed(userID, apiKeyID int64, groupID *int64) string {
 	gid := int64(0)
 	if groupID != nil {

backend/internal/handler/ops_error_logger.go

@@ -31,6 +31,7 @@ const (
 const (
 	opsErrorLogTimeout      = 5 * time.Second
 	opsErrorLogDrainTimeout = 10 * time.Second
+	opsErrorLogBatchWindow  = 200 * time.Millisecond
 
 	opsErrorLogMinWorkerCount = 4
 	opsErrorLogMaxWorkerCount = 32
@@ -38,6 +39,7 @@ const (
 	opsErrorLogQueueSizePerWorker = 128
 	opsErrorLogMinQueueSize       = 256
 	opsErrorLogMaxQueueSize       = 8192
+	opsErrorLogBatchSize          = 32
 )
 
 type opsErrorLogJob struct {
@@ -82,27 +84,82 @@ func startOpsErrorLogWorkers() {
 	for i := 0; i < workerCount; i++ {
 		go func() {
 			defer opsErrorLogWorkersWg.Done()
-			for job := range opsErrorLogQueue {
-				opsErrorLogQueueLen.Add(-1)
-				if job.ops == nil || job.entry == nil {
-					continue
+			for {
+				job, ok := <-opsErrorLogQueue
+				if !ok {
+					return
 				}
-				func() {
-					defer func() {
-						if r := recover(); r != nil {
-							log.Printf("[OpsErrorLogger] worker panic: %v\n%s", r, debug.Stack())
+				opsErrorLogQueueLen.Add(-1)
+				batch := make([]opsErrorLogJob, 0, opsErrorLogBatchSize)
+				batch = append(batch, job)
+
+				timer := time.NewTimer(opsErrorLogBatchWindow)
+			batchLoop:
+				for len(batch) < opsErrorLogBatchSize {
+					select {
+					case nextJob, ok := <-opsErrorLogQueue:
+						if !ok {
+							if !timer.Stop() {
+								select {
+								case <-timer.C:
+								default:
+								}
+							}
+							flushOpsErrorLogBatch(batch)
+							return
 						}
-					}()
-					ctx, cancel := context.WithTimeout(context.Background(), opsErrorLogTimeout)
-					_ = job.ops.RecordError(ctx, job.entry, nil)
-					cancel()
-					opsErrorLogProcessed.Add(1)
-				}()
+						opsErrorLogQueueLen.Add(-1)
+						batch = append(batch, nextJob)
+					case <-timer.C:
+						break batchLoop
+					}
+				}
+				if !timer.Stop() {
+					select {
+					case <-timer.C:
+					default:
+					}
+				}
+				flushOpsErrorLogBatch(batch)
 			}
 		}()
 	}
 }
 
+func flushOpsErrorLogBatch(batch []opsErrorLogJob) {
+	if len(batch) == 0 {
+		return
+	}
+	defer func() {
+		if r := recover(); r != nil {
+			log.Printf("[OpsErrorLogger] worker panic: %v\n%s", r, debug.Stack())
+		}
+	}()
+
+	grouped := make(map[*service.OpsService][]*service.OpsInsertErrorLogInput, len(batch))
+	var processed int64
+	for _, job := range batch {
+		if job.ops == nil || job.entry == nil {
+			continue
+		}
+		grouped[job.ops] = append(grouped[job.ops], job.entry)
+		processed++
+	}
+	if processed == 0 {
+		return
+	}
+
+	for opsSvc, entries := range grouped {
+		if opsSvc == nil || len(entries) == 0 {
+			continue
+		}
+		ctx, cancel := context.WithTimeout(context.Background(), opsErrorLogTimeout)
+		_ = opsSvc.RecordErrorBatch(ctx, entries)
+		cancel()
+	}
+	opsErrorLogProcessed.Add(processed)
+}
+
 func enqueueOpsErrorLog(ops *service.OpsService, entry *service.OpsInsertErrorLogInput) {
 	if ops == nil || entry == nil {
 		return

backend/internal/handler/setting_handler.go

@@ -54,6 +54,7 @@ func (h *SettingHandler) GetPublicSettings(c *gin.Context) {
 		CustomMenuItems:                  dto.ParseUserVisibleMenuItems(settings.CustomMenuItems),
 		LinuxDoOAuthEnabled:              settings.LinuxDoOAuthEnabled,
 		SoraClientEnabled:                settings.SoraClientEnabled,
+		BackendModeEnabled:               settings.BackendModeEnabled,
 		Version:                          h.version,
 	})
 }

backend/internal/handler/sora_client_handler_test.go

@@ -2132,6 +2132,14 @@ func (r *stubAccountRepoForHandler) BulkUpdate(context.Context, []int64, service
 	return 0, nil
 }
 
+func (r *stubAccountRepoForHandler) IncrementQuotaUsed(context.Context, int64, float64) error {
+	return nil
+}
+
+func (r *stubAccountRepoForHandler) ResetQuotaUsed(context.Context, int64) error {
+	return nil
+}
+
 // ==================== Stub: SoraClient (用于 SoraGatewayService) ====================
 
 var _ service.SoraClient = (*stubSoraClientForHandler)(nil)
@@ -2198,8 +2206,8 @@ func (s *stubSoraClientForHandler) GetVideoTask(_ context.Context, _ *service.Ac
 // newMinimalGatewayService 创建仅包含 accountRepo 的最小 GatewayService（用于测试 SelectAccountForModel）。
 func newMinimalGatewayService(accountRepo service.AccountRepository) *service.GatewayService {
 	return service.NewGatewayService(
-		accountRepo, nil, nil, nil, nil, nil, nil, nil,
-		nil, nil, nil, nil, nil, nil, nil, nil, nil, nil, nil, nil,
+		accountRepo, nil, nil, nil, nil, nil, nil, nil, nil,
+		nil, nil, nil, nil, nil, nil, nil, nil, nil, nil, nil, nil, nil,
 	)
 }
 

backend/internal/handler/sora_gateway_handler.go

@@ -399,17 +399,19 @@ func (h *SoraGatewayHandler) ChatCompletions(c *gin.Context) {
 
 		userAgent := c.GetHeader("User-Agent")
 		clientIP := ip.GetClientIP(c)
+		requestPayloadHash := service.HashUsageRequestPayload(body)
 
 		// 使用量记录通过有界 worker 池提交，避免请求热路径创建无界 goroutine。
 		h.submitUsageRecordTask(func(ctx context.Context) {
 			if err := h.gatewayService.RecordUsage(ctx, &service.RecordUsageInput{
-				Result:       result,
-				APIKey:       apiKey,
-				User:         apiKey.User,
-				Account:      account,
-				Subscription: subscription,
-				UserAgent:    userAgent,
-				IPAddress:    clientIP,
+				Result:             result,
+				APIKey:             apiKey,
+				User:               apiKey.User,
+				Account:            account,
+				Subscription:       subscription,
+				UserAgent:          userAgent,
+				IPAddress:          clientIP,
+				RequestPayloadHash: requestPayloadHash,
 			}); err != nil {
 				logger.L().With(
 					zap.String("component", "handler.sora_gateway.chat_completions"),

backend/internal/handler/sora_gateway_handler_test.go

@@ -216,6 +216,14 @@ func (r *stubAccountRepo) BulkUpdate(ctx context.Context, ids []int64, updates s
 	return 0, nil
 }
 
+func (r *stubAccountRepo) IncrementQuotaUsed(ctx context.Context, id int64, amount float64) error {
+	return nil
+}
+
+func (r *stubAccountRepo) ResetQuotaUsed(ctx context.Context, id int64) error {
+	return nil
+}
+
 func (r *stubAccountRepo) listSchedulable() []service.Account {
 	var result []service.Account
 	for _, acc := range r.accounts {
@@ -335,6 +343,9 @@ func (s *stubUsageLogRepo) GetAPIKeyUsageTrend(ctx context.Context, startTime, e
 func (s *stubUsageLogRepo) GetUserUsageTrend(ctx context.Context, startTime, endTime time.Time, granularity string, limit int) ([]usagestats.UserUsageTrendPoint, error) {
 	return nil, nil
 }
+func (s *stubUsageLogRepo) GetUserSpendingRanking(ctx context.Context, startTime, endTime time.Time, limit int) (*usagestats.UserSpendingRankingResponse, error) {
+	return nil, nil
+}
 func (s *stubUsageLogRepo) GetBatchUserUsageStats(ctx context.Context, userIDs []int64, startTime, endTime time.Time) (map[int64]*usagestats.BatchUserUsageStats, error) {
 	return nil, nil
 }
@@ -423,6 +434,7 @@ func TestSoraGatewayHandler_ChatCompletions(t *testing.T) {
 		nil,
 		nil,
 		nil,
+		nil,
 		testutil.StubGatewayCache{},
 		cfg,
 		nil,
@@ -437,6 +449,7 @@ func TestSoraGatewayHandler_ChatCompletions(t *testing.T) {
 		testutil.StubSessionLimitCache{},
 		nil, // rpmCache
 		nil, // digestStore
+		nil, // settingService
 	)
 
 	soraClient := &stubSoraClient{imageURLs: []string{"https://example.com/a.png"}}

backend/internal/handler/wire.go

@@ -15,6 +15,7 @@ func ProvideAdminHandlers(
 	accountHandler *admin.AccountHandler,
 	announcementHandler *admin.AnnouncementHandler,
 	dataManagementHandler *admin.DataManagementHandler,
+	backupHandler *admin.BackupHandler,
 	oauthHandler *admin.OAuthHandler,
 	openaiOAuthHandler *admin.OpenAIOAuthHandler,
 	geminiOAuthHandler *admin.GeminiOAuthHandler,
@@ -39,6 +40,7 @@ func ProvideAdminHandlers(
 		Account:          accountHandler,
 		Announcement:     announcementHandler,
 		DataManagement:   dataManagementHandler,
+		Backup:           backupHandler,
 		OAuth:            oauthHandler,
 		OpenAIOAuth:      openaiOAuthHandler,
 		GeminiOAuth:      geminiOAuthHandler,
@@ -128,6 +130,7 @@ var ProviderSet = wire.NewSet(
 	admin.NewAccountHandler,
 	admin.NewAnnouncementHandler,
 	admin.NewDataManagementHandler,
+	admin.NewBackupHandler,
 	admin.NewOAuthHandler,
 	admin.NewOpenAIOAuthHandler,
 	admin.NewGeminiOAuthHandler,

backend/internal/pkg/antigravity/claude_types.go

@@ -159,6 +159,8 @@ var claudeModels = []modelDef{
 // Antigravity 支持的 Gemini 模型
 var geminiModels = []modelDef{
 	{ID: "gemini-2.5-flash", DisplayName: "Gemini 2.5 Flash", CreatedAt: "2025-01-01T00:00:00Z"},
+	{ID: "gemini-2.5-flash-image", DisplayName: "Gemini 2.5 Flash Image", CreatedAt: "2025-01-01T00:00:00Z"},
+	{ID: "gemini-2.5-flash-image-preview", DisplayName: "Gemini 2.5 Flash Image Preview", CreatedAt: "2025-01-01T00:00:00Z"},
 	{ID: "gemini-2.5-flash-lite", DisplayName: "Gemini 2.5 Flash Lite", CreatedAt: "2025-01-01T00:00:00Z"},
 	{ID: "gemini-2.5-flash-thinking", DisplayName: "Gemini 2.5 Flash Thinking", CreatedAt: "2025-01-01T00:00:00Z"},
 	{ID: "gemini-3-flash", DisplayName: "Gemini 3 Flash", CreatedAt: "2025-06-01T00:00:00Z"},

backend/internal/pkg/antigravity/claude_types_test.go

@@ -13,6 +13,8 @@ func TestDefaultModels_ContainsNewAndLegacyImageModels(t *testing.T) {
 
 	requiredIDs := []string{
 		"claude-opus-4-6-thinking",
+		"gemini-2.5-flash-image",
+		"gemini-2.5-flash-image-preview",
 		"gemini-3.1-flash-image",
 		"gemini-3.1-flash-image-preview",
 		"gemini-3-pro-image", // legacy compatibility

backend/internal/pkg/antigravity/client.go

@@ -19,6 +19,16 @@ import (
 	"github.com/Wei-Shaw/sub2api/internal/pkg/proxyutil"
 )
 
+// ForbiddenError 表示上游返回 403 Forbidden
+type ForbiddenError struct {
+	StatusCode int
+	Body       string
+}
+
+func (e *ForbiddenError) Error() string {
+	return fmt.Sprintf("fetchAvailableModels 失败 (HTTP %d): %s", e.StatusCode, e.Body)
+}
+
 // NewAPIRequestWithURL 使用指定的 base URL 创建 Antigravity API 请求（v1internal 端点）
 func NewAPIRequestWithURL(ctx context.Context, baseURL, action, accessToken string, body []byte) (*http.Request, error) {
 	// 构建 URL，流式请求添加 ?alt=sse 参数
@@ -514,7 +524,20 @@ type ModelQuotaInfo struct {
 
 // ModelInfo 模型信息
 type ModelInfo struct {
-	QuotaInfo *ModelQuotaInfo `json:"quotaInfo,omitempty"`
+	QuotaInfo          *ModelQuotaInfo `json:"quotaInfo,omitempty"`
+	DisplayName        string          `json:"displayName,omitempty"`
+	SupportsImages     *bool           `json:"supportsImages,omitempty"`
+	SupportsThinking   *bool           `json:"supportsThinking,omitempty"`
+	ThinkingBudget     *int            `json:"thinkingBudget,omitempty"`
+	Recommended        *bool           `json:"recommended,omitempty"`
+	MaxTokens          *int            `json:"maxTokens,omitempty"`
+	MaxOutputTokens    *int            `json:"maxOutputTokens,omitempty"`
+	SupportedMimeTypes map[string]bool `json:"supportedMimeTypes,omitempty"`
+}
+
+// DeprecatedModelInfo 废弃模型转发信息
+type DeprecatedModelInfo struct {
+	NewModelID string `json:"newModelId"`
 }
 
 // FetchAvailableModelsRequest fetchAvailableModels 请求
@@ -524,7 +547,8 @@ type FetchAvailableModelsRequest struct {
 
 // FetchAvailableModelsResponse fetchAvailableModels 响应
 type FetchAvailableModelsResponse struct {
-	Models map[string]ModelInfo `json:"models"`
+	Models             map[string]ModelInfo           `json:"models"`
+	DeprecatedModelIDs map[string]DeprecatedModelInfo `json:"deprecatedModelIds,omitempty"`
 }
 
 // FetchAvailableModels 获取可用模型和配额信息，返回解析后的结构体和原始 JSON
@@ -573,6 +597,13 @@ func (c *Client) FetchAvailableModels(ctx context.Context, accessToken, projectI
 			continue
 		}
 
+		if resp.StatusCode == http.StatusForbidden {
+			return nil, nil, &ForbiddenError{
+				StatusCode: resp.StatusCode,
+				Body:       string(respBodyBytes),
+			}
+		}
+
 		if resp.StatusCode != http.StatusOK {
 			return nil, nil, fmt.Errorf("fetchAvailableModels 失败 (HTTP %d): %s", resp.StatusCode, string(respBodyBytes))
 		}

backend/internal/pkg/antigravity/gemini_types.go

@@ -189,6 +189,5 @@ var DefaultStopSequences = []string{
 	"<|user|>",
 	"<|endoftext|>",
 	"<|end_of_turn|>",
-	"[DONE]",
 	"\n\nHuman:",
 }

backend/internal/pkg/antigravity/oauth.go

@@ -49,8 +49,8 @@ const (
 	antigravityDailyBaseURL = "https://daily-cloudcode-pa.sandbox.googleapis.com"
 )
 
-// defaultUserAgentVersion 可通过环境变量 ANTIGRAVITY_USER_AGENT_VERSION 配置，默认 1.19.6
-var defaultUserAgentVersion = "1.19.6"
+// defaultUserAgentVersion 可通过环境变量 ANTIGRAVITY_USER_AGENT_VERSION 配置，默认 1.20.4
+var defaultUserAgentVersion = "1.20.4"
 
 // defaultClientSecret 可通过环境变量 ANTIGRAVITY_OAUTH_CLIENT_SECRET 配置
 var defaultClientSecret = "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"

backend/internal/pkg/antigravity/oauth_test.go

@@ -690,7 +690,7 @@ func TestConstants_值正确(t *testing.T) {
 	if RedirectURI != "http://localhost:8085/callback" {
 		t.Errorf("RedirectURI 不匹配: got %s", RedirectURI)
 	}
-	if GetUserAgent() != "antigravity/1.19.6 windows/amd64" {
+	if GetUserAgent() != "antigravity/1.20.4 windows/amd64" {
 		t.Errorf("UserAgent 不匹配: got %s", GetUserAgent())
 	}
 	if SessionTTL != 30*time.Minute {

backend/internal/pkg/antigravity/stream_transformer.go

@@ -119,23 +119,33 @@ func (p *StreamingProcessor) ProcessLine(line string) []byte {
 	return result.Bytes()
 }
 
-// Finish 结束处理，返回最终事件和用量
+// Finish 结束处理，返回最终事件和用量。
+// 若整个流未收到任何可解析的上游数据（messageStartSent == false），
+// 则不补发任何结束事件，防止客户端收到没有 message_start 的残缺流。
 func (p *StreamingProcessor) Finish() ([]byte, *ClaudeUsage) {
-	var result bytes.Buffer
-
-	if !p.messageStopSent {
-		_, _ = result.Write(p.emitFinish(""))
-	}
-
 	usage := &ClaudeUsage{
 		InputTokens:          p.inputTokens,
 		OutputTokens:         p.outputTokens,
 		CacheReadInputTokens: p.cacheReadTokens,
 	}
 
+	if !p.messageStartSent {
+		return nil, usage
+	}
+
+	var result bytes.Buffer
+	if !p.messageStopSent {
+		_, _ = result.Write(p.emitFinish(""))
+	}
+
 	return result.Bytes(), usage
 }
 
+// MessageStartSent 报告流中是否已发出过 message_start 事件（即是否收到过有效的上游数据）
+func (p *StreamingProcessor) MessageStartSent() bool {
+	return p.messageStartSent
+}
+
 // emitMessageStart 发送 message_start 事件
 func (p *StreamingProcessor) emitMessageStart(v1Resp *V1InternalResponse) []byte {
 	if p.messageStartSent {

backend/internal/pkg/apicompat/anthropic_to_responses.go

@@ -0,0 +1,416 @@
+package apicompat
+
+import (
+	"encoding/json"
+	"fmt"
+	"strings"
+)
+
+// AnthropicToResponses converts an Anthropic Messages request directly into
+// a Responses API request. This preserves fields that would be lost in a
+// Chat Completions intermediary round-trip (e.g. thinking, cache_control,
+// structured system prompts).
+func AnthropicToResponses(req *AnthropicRequest) (*ResponsesRequest, error) {
+	input, err := convertAnthropicToResponsesInput(req.System, req.Messages)
+	if err != nil {
+		return nil, err
+	}
+
+	inputJSON, err := json.Marshal(input)
+	if err != nil {
+		return nil, err
+	}
+
+	out := &ResponsesRequest{
+		Model:       req.Model,
+		Input:       inputJSON,
+		Temperature: req.Temperature,
+		TopP:        req.TopP,
+		Stream:      req.Stream,
+		Include:     []string{"reasoning.encrypted_content"},
+	}
+
+	storeFalse := false
+	out.Store = &storeFalse
+
+	if req.MaxTokens > 0 {
+		v := req.MaxTokens
+		if v < minMaxOutputTokens {
+			v = minMaxOutputTokens
+		}
+		out.MaxOutputTokens = &v
+	}
+
+	if len(req.Tools) > 0 {
+		out.Tools = convertAnthropicToolsToResponses(req.Tools)
+	}
+
+	// Determine reasoning effort: only output_config.effort controls the
+	// level; thinking.type is ignored. Default is xhigh when unset.
+	// Anthropic levels map to OpenAI: low→low, medium→high, high→xhigh.
+	effort := "high" // default → maps to xhigh
+	if req.OutputConfig != nil && req.OutputConfig.Effort != "" {
+		effort = req.OutputConfig.Effort
+	}
+	out.Reasoning = &ResponsesReasoning{
+		Effort:  mapAnthropicEffortToResponses(effort),
+		Summary: "auto",
+	}
+
+	// Convert tool_choice
+	if len(req.ToolChoice) > 0 {
+		tc, err := convertAnthropicToolChoiceToResponses(req.ToolChoice)
+		if err != nil {
+			return nil, fmt.Errorf("convert tool_choice: %w", err)
+		}
+		out.ToolChoice = tc
+	}
+
+	return out, nil
+}
+
+// convertAnthropicToolChoiceToResponses maps Anthropic tool_choice to Responses format.
+//
+//	{"type":"auto"}            → "auto"
+//	{"type":"any"}             → "required"
+//	{"type":"none"}            → "none"
+//	{"type":"tool","name":"X"} → {"type":"function","function":{"name":"X"}}
+func convertAnthropicToolChoiceToResponses(raw json.RawMessage) (json.RawMessage, error) {
+	var tc struct {
+		Type string `json:"type"`
+		Name string `json:"name"`
+	}
+	if err := json.Unmarshal(raw, &tc); err != nil {
+		return nil, err
+	}
+
+	switch tc.Type {
+	case "auto":
+		return json.Marshal("auto")
+	case "any":
+		return json.Marshal("required")
+	case "none":
+		return json.Marshal("none")
+	case "tool":
+		return json.Marshal(map[string]any{
+			"type":     "function",
+			"function": map[string]string{"name": tc.Name},
+		})
+	default:
+		// Pass through unknown types as-is
+		return raw, nil
+	}
+}
+
+// convertAnthropicToResponsesInput builds the Responses API input items array
+// from the Anthropic system field and message list.
+func convertAnthropicToResponsesInput(system json.RawMessage, msgs []AnthropicMessage) ([]ResponsesInputItem, error) {
+	var out []ResponsesInputItem
+
+	// System prompt → system role input item.
+	if len(system) > 0 {
+		sysText, err := parseAnthropicSystemPrompt(system)
+		if err != nil {
+			return nil, err
+		}
+		if sysText != "" {
+			content, _ := json.Marshal(sysText)
+			out = append(out, ResponsesInputItem{
+				Role:    "system",
+				Content: content,
+			})
+		}
+	}
+
+	for _, m := range msgs {
+		items, err := anthropicMsgToResponsesItems(m)
+		if err != nil {
+			return nil, err
+		}
+		out = append(out, items...)
+	}
+	return out, nil
+}
+
+// parseAnthropicSystemPrompt handles the Anthropic system field which can be
+// a plain string or an array of text blocks.
+func parseAnthropicSystemPrompt(raw json.RawMessage) (string, error) {
+	var s string
+	if err := json.Unmarshal(raw, &s); err == nil {
+		return s, nil
+	}
+	var blocks []AnthropicContentBlock
+	if err := json.Unmarshal(raw, &blocks); err != nil {
+		return "", err
+	}
+	var parts []string
+	for _, b := range blocks {
+		if b.Type == "text" && b.Text != "" {
+			parts = append(parts, b.Text)
+		}
+	}
+	return strings.Join(parts, "\n\n"), nil
+}
+
+// anthropicMsgToResponsesItems converts a single Anthropic message into one
+// or more Responses API input items.
+func anthropicMsgToResponsesItems(m AnthropicMessage) ([]ResponsesInputItem, error) {
+	switch m.Role {
+	case "user":
+		return anthropicUserToResponses(m.Content)
+	case "assistant":
+		return anthropicAssistantToResponses(m.Content)
+	default:
+		return anthropicUserToResponses(m.Content)
+	}
+}
+
+// anthropicUserToResponses handles an Anthropic user message. Content can be a
+// plain string or an array of blocks. tool_result blocks are extracted into
+// function_call_output items. Image blocks are converted to input_image parts.
+func anthropicUserToResponses(raw json.RawMessage) ([]ResponsesInputItem, error) {
+	// Try plain string.
+	var s string
+	if err := json.Unmarshal(raw, &s); err == nil {
+		content, _ := json.Marshal(s)
+		return []ResponsesInputItem{{Role: "user", Content: content}}, nil
+	}
+
+	var blocks []AnthropicContentBlock
+	if err := json.Unmarshal(raw, &blocks); err != nil {
+		return nil, err
+	}
+
+	var out []ResponsesInputItem
+	var toolResultImageParts []ResponsesContentPart
+
+	// Extract tool_result blocks → function_call_output items.
+	// Images inside tool_results are extracted separately because the
+	// Responses API function_call_output.output only accepts strings.
+	for _, b := range blocks {
+		if b.Type != "tool_result" {
+			continue
+		}
+		outputText, imageParts := convertToolResultOutput(b)
+		out = append(out, ResponsesInputItem{
+			Type:   "function_call_output",
+			CallID: toResponsesCallID(b.ToolUseID),
+			Output: outputText,
+		})
+		toolResultImageParts = append(toolResultImageParts, imageParts...)
+	}
+
+	// Remaining text + image blocks → user message with content parts.
+	// Also include images extracted from tool_results so the model can see them.
+	var parts []ResponsesContentPart
+	for _, b := range blocks {
+		switch b.Type {
+		case "text":
+			if b.Text != "" {
+				parts = append(parts, ResponsesContentPart{Type: "input_text", Text: b.Text})
+			}
+		case "image":
+			if uri := anthropicImageToDataURI(b.Source); uri != "" {
+				parts = append(parts, ResponsesContentPart{Type: "input_image", ImageURL: uri})
+			}
+		}
+	}
+	parts = append(parts, toolResultImageParts...)
+
+	if len(parts) > 0 {
+		content, err := json.Marshal(parts)
+		if err != nil {
+			return nil, err
+		}
+		out = append(out, ResponsesInputItem{Role: "user", Content: content})
+	}
+
+	return out, nil
+}
+
+// anthropicAssistantToResponses handles an Anthropic assistant message.
+// Text content → assistant message with output_text parts.
+// tool_use blocks → function_call items.
+// thinking blocks → ignored (OpenAI doesn't accept them as input).
+func anthropicAssistantToResponses(raw json.RawMessage) ([]ResponsesInputItem, error) {
+	// Try plain string.
+	var s string
+	if err := json.Unmarshal(raw, &s); err == nil {
+		parts := []ResponsesContentPart{{Type: "output_text", Text: s}}
+		partsJSON, err := json.Marshal(parts)
+		if err != nil {
+			return nil, err
+		}
+		return []ResponsesInputItem{{Role: "assistant", Content: partsJSON}}, nil
+	}
+
+	var blocks []AnthropicContentBlock
+	if err := json.Unmarshal(raw, &blocks); err != nil {
+		return nil, err
+	}
+
+	var items []ResponsesInputItem
+
+	// Text content → assistant message with output_text content parts.
+	text := extractAnthropicTextFromBlocks(blocks)
+	if text != "" {
+		parts := []ResponsesContentPart{{Type: "output_text", Text: text}}
+		partsJSON, err := json.Marshal(parts)
+		if err != nil {
+			return nil, err
+		}
+		items = append(items, ResponsesInputItem{Role: "assistant", Content: partsJSON})
+	}
+
+	// tool_use → function_call items.
+	for _, b := range blocks {
+		if b.Type != "tool_use" {
+			continue
+		}
+		args := "{}"
+		if len(b.Input) > 0 {
+			args = string(b.Input)
+		}
+		fcID := toResponsesCallID(b.ID)
+		items = append(items, ResponsesInputItem{
+			Type:      "function_call",
+			CallID:    fcID,
+			Name:      b.Name,
+			Arguments: args,
+		})
+	}
+
+	return items, nil
+}
+
+// toResponsesCallID converts an Anthropic tool ID (toolu_xxx / call_xxx) to a
+// Responses API function_call ID that starts with "fc_".
+func toResponsesCallID(id string) string {
+	if strings.HasPrefix(id, "fc_") {
+		return id
+	}
+	return "fc_" + id
+}
+
+// fromResponsesCallID reverses toResponsesCallID, stripping the "fc_" prefix
+// that was added during request conversion.
+func fromResponsesCallID(id string) string {
+	if after, ok := strings.CutPrefix(id, "fc_"); ok {
+		// Only strip if the remainder doesn't look like it was already "fc_" prefixed.
+		// E.g. "fc_toolu_xxx" → "toolu_xxx", "fc_call_xxx" → "call_xxx"
+		if strings.HasPrefix(after, "toolu_") || strings.HasPrefix(after, "call_") {
+			return after
+		}
+	}
+	return id
+}
+
+// anthropicImageToDataURI converts an AnthropicImageSource to a data URI string.
+// Returns "" if the source is nil or has no data.
+func anthropicImageToDataURI(src *AnthropicImageSource) string {
+	if src == nil || src.Data == "" {
+		return ""
+	}
+	mediaType := src.MediaType
+	if mediaType == "" {
+		mediaType = "image/png"
+	}
+	return "data:" + mediaType + ";base64," + src.Data
+}
+
+// convertToolResultOutput extracts text and image content from a tool_result
+// block. Returns the text as a string for the function_call_output Output
+// field, plus any image parts that must be sent in a separate user message
+// (the Responses API output field only accepts strings).
+func convertToolResultOutput(b AnthropicContentBlock) (string, []ResponsesContentPart) {
+	if len(b.Content) == 0 {
+		return "(empty)", nil
+	}
+
+	// Try plain string content.
+	var s string
+	if err := json.Unmarshal(b.Content, &s); err == nil {
+		if s == "" {
+			s = "(empty)"
+		}
+		return s, nil
+	}
+
+	// Array of content blocks — may contain text and/or images.
+	var inner []AnthropicContentBlock
+	if err := json.Unmarshal(b.Content, &inner); err != nil {
+		return "(empty)", nil
+	}
+
+	// Separate text (for function_call_output) from images (for user message).
+	var textParts []string
+	var imageParts []ResponsesContentPart
+	for _, ib := range inner {
+		switch ib.Type {
+		case "text":
+			if ib.Text != "" {
+				textParts = append(textParts, ib.Text)
+			}
+		case "image":
+			if uri := anthropicImageToDataURI(ib.Source); uri != "" {
+				imageParts = append(imageParts, ResponsesContentPart{Type: "input_image", ImageURL: uri})
+			}
+		}
+	}
+
+	text := strings.Join(textParts, "\n\n")
+	if text == "" {
+		text = "(empty)"
+	}
+	return text, imageParts
+}
+
+// extractAnthropicTextFromBlocks joins all text blocks, ignoring thinking/
+// tool_use/tool_result blocks.
+func extractAnthropicTextFromBlocks(blocks []AnthropicContentBlock) string {
+	var parts []string
+	for _, b := range blocks {
+		if b.Type == "text" && b.Text != "" {
+			parts = append(parts, b.Text)
+		}
+	}
+	return strings.Join(parts, "\n\n")
+}
+
+// mapAnthropicEffortToResponses converts Anthropic reasoning effort levels to
+// OpenAI Responses API effort levels.
+//
+//	low    → low
+//	medium → high
+//	high   → xhigh
+func mapAnthropicEffortToResponses(effort string) string {
+	switch effort {
+	case "medium":
+		return "high"
+	case "high":
+		return "xhigh"
+	default:
+		return effort // "low" and any unknown values pass through unchanged
+	}
+}
+
+// convertAnthropicToolsToResponses maps Anthropic tool definitions to
+// Responses API tools. Server-side tools like web_search are mapped to their
+// OpenAI equivalents; regular tools become function tools.
+func convertAnthropicToolsToResponses(tools []AnthropicTool) []ResponsesTool {
+	var out []ResponsesTool
+	for _, t := range tools {
+		// Anthropic server tools like "web_search_20250305" → OpenAI {"type":"web_search"}
+		if strings.HasPrefix(t.Type, "web_search") {
+			out = append(out, ResponsesTool{Type: "web_search"})
+			continue
+		}
+		out = append(out, ResponsesTool{
+			Type:        "function",
+			Name:        t.Name,
+			Description: t.Description,
+			Parameters:  t.InputSchema,
+		})
+	}
+	return out
+}

backend/internal/pkg/apicompat/chatcompletions_responses_test.go

@@ -0,0 +1,810 @@
+package apicompat
+
+import (
+	"encoding/json"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// ---------------------------------------------------------------------------
+// ChatCompletionsToResponses tests
+// ---------------------------------------------------------------------------
+
+func TestChatCompletionsToResponses_BasicText(t *testing.T) {
+	req := &ChatCompletionsRequest{
+		Model: "gpt-4o",
+		Messages: []ChatMessage{
+			{Role: "user", Content: json.RawMessage(`"Hello"`)},
+		},
+	}
+
+	resp, err := ChatCompletionsToResponses(req)
+	require.NoError(t, err)
+	assert.Equal(t, "gpt-4o", resp.Model)
+	assert.True(t, resp.Stream) // always forced true
+	assert.False(t, *resp.Store)
+
+	var items []ResponsesInputItem
+	require.NoError(t, json.Unmarshal(resp.Input, &items))
+	require.Len(t, items, 1)
+	assert.Equal(t, "user", items[0].Role)
+}
+
+func TestChatCompletionsToResponses_SystemMessage(t *testing.T) {
+	req := &ChatCompletionsRequest{
+		Model: "gpt-4o",
+		Messages: []ChatMessage{
+			{Role: "system", Content: json.RawMessage(`"You are helpful."`)},
+			{Role: "user", Content: json.RawMessage(`"Hi"`)},
+		},
+	}
+
+	resp, err := ChatCompletionsToResponses(req)
+	require.NoError(t, err)
+
+	var items []ResponsesInputItem
+	require.NoError(t, json.Unmarshal(resp.Input, &items))
+	require.Len(t, items, 2)
+	assert.Equal(t, "system", items[0].Role)
+	assert.Equal(t, "user", items[1].Role)
+}
+
+func TestChatCompletionsToResponses_ToolCalls(t *testing.T) {
+	req := &ChatCompletionsRequest{
+		Model: "gpt-4o",
+		Messages: []ChatMessage{
+			{Role: "user", Content: json.RawMessage(`"Call the function"`)},
+			{
+				Role: "assistant",
+				ToolCalls: []ChatToolCall{
+					{
+						ID:   "call_1",
+						Type: "function",
+						Function: ChatFunctionCall{
+							Name:      "ping",
+							Arguments: `{"host":"example.com"}`,
+						},
+					},
+				},
+			},
+			{
+				Role:       "tool",
+				ToolCallID: "call_1",
+				Content:    json.RawMessage(`"pong"`),
+			},
+		},
+		Tools: []ChatTool{
+			{
+				Type: "function",
+				Function: &ChatFunction{
+					Name:        "ping",
+					Description: "Ping a host",
+					Parameters:  json.RawMessage(`{"type":"object"}`),
+				},
+			},
+		},
+	}
+
+	resp, err := ChatCompletionsToResponses(req)
+	require.NoError(t, err)
+
+	var items []ResponsesInputItem
+	require.NoError(t, json.Unmarshal(resp.Input, &items))
+	// user + function_call + function_call_output = 3
+	// (assistant message with empty content + tool_calls → only function_call items emitted)
+	require.Len(t, items, 3)
+
+	// Check function_call item
+	assert.Equal(t, "function_call", items[1].Type)
+	assert.Equal(t, "call_1", items[1].CallID)
+	assert.Empty(t, items[1].ID)
+	assert.Equal(t, "ping", items[1].Name)
+
+	// Check function_call_output item
+	assert.Equal(t, "function_call_output", items[2].Type)
+	assert.Equal(t, "call_1", items[2].CallID)
+	assert.Equal(t, "pong", items[2].Output)
+
+	// Check tools
+	require.Len(t, resp.Tools, 1)
+	assert.Equal(t, "function", resp.Tools[0].Type)
+	assert.Equal(t, "ping", resp.Tools[0].Name)
+}
+
+func TestChatCompletionsToResponses_MaxTokens(t *testing.T) {
+	t.Run("max_tokens", func(t *testing.T) {
+		maxTokens := 100
+		req := &ChatCompletionsRequest{
+			Model:     "gpt-4o",
+			MaxTokens: &maxTokens,
+			Messages:  []ChatMessage{{Role: "user", Content: json.RawMessage(`"Hi"`)}},
+		}
+		resp, err := ChatCompletionsToResponses(req)
+		require.NoError(t, err)
+		require.NotNil(t, resp.MaxOutputTokens)
+		// Below minMaxOutputTokens (128), should be clamped
+		assert.Equal(t, minMaxOutputTokens, *resp.MaxOutputTokens)
+	})
+
+	t.Run("max_completion_tokens_preferred", func(t *testing.T) {
+		maxTokens := 100
+		maxCompletion := 500
+		req := &ChatCompletionsRequest{
+			Model:               "gpt-4o",
+			MaxTokens:           &maxTokens,
+			MaxCompletionTokens: &maxCompletion,
+			Messages:            []ChatMessage{{Role: "user", Content: json.RawMessage(`"Hi"`)}},
+		}
+		resp, err := ChatCompletionsToResponses(req)
+		require.NoError(t, err)
+		require.NotNil(t, resp.MaxOutputTokens)
+		assert.Equal(t, 500, *resp.MaxOutputTokens)
+	})
+}
+
+func TestChatCompletionsToResponses_ReasoningEffort(t *testing.T) {
+	req := &ChatCompletionsRequest{
+		Model:           "gpt-4o",
+		ReasoningEffort: "high",
+		Messages:        []ChatMessage{{Role: "user", Content: json.RawMessage(`"Hi"`)}},
+	}
+	resp, err := ChatCompletionsToResponses(req)
+	require.NoError(t, err)
+	require.NotNil(t, resp.Reasoning)
+	assert.Equal(t, "high", resp.Reasoning.Effort)
+	assert.Equal(t, "auto", resp.Reasoning.Summary)
+}
+
+func TestChatCompletionsToResponses_ImageURL(t *testing.T) {
+	content := `[{"type":"text","text":"Describe this"},{"type":"image_url","image_url":{"url":"data:image/png;base64,abc123"}}]`
+	req := &ChatCompletionsRequest{
+		Model: "gpt-4o",
+		Messages: []ChatMessage{
+			{Role: "user", Content: json.RawMessage(content)},
+		},
+	}
+	resp, err := ChatCompletionsToResponses(req)
+	require.NoError(t, err)
+
+	var items []ResponsesInputItem
+	require.NoError(t, json.Unmarshal(resp.Input, &items))
+	require.Len(t, items, 1)
+
+	var parts []ResponsesContentPart
+	require.NoError(t, json.Unmarshal(items[0].Content, &parts))
+	require.Len(t, parts, 2)
+	assert.Equal(t, "input_text", parts[0].Type)
+	assert.Equal(t, "Describe this", parts[0].Text)
+	assert.Equal(t, "input_image", parts[1].Type)
+	assert.Equal(t, "data:image/png;base64,abc123", parts[1].ImageURL)
+}
+
+func TestChatCompletionsToResponses_LegacyFunctions(t *testing.T) {
+	req := &ChatCompletionsRequest{
+		Model: "gpt-4o",
+		Messages: []ChatMessage{
+			{Role: "user", Content: json.RawMessage(`"Hi"`)},
+		},
+		Functions: []ChatFunction{
+			{
+				Name:        "get_weather",
+				Description: "Get weather",
+				Parameters:  json.RawMessage(`{"type":"object"}`),
+			},
+		},
+		FunctionCall: json.RawMessage(`{"name":"get_weather"}`),
+	}
+
+	resp, err := ChatCompletionsToResponses(req)
+	require.NoError(t, err)
+	require.Len(t, resp.Tools, 1)
+	assert.Equal(t, "function", resp.Tools[0].Type)
+	assert.Equal(t, "get_weather", resp.Tools[0].Name)
+
+	// tool_choice should be converted
+	require.NotNil(t, resp.ToolChoice)
+	var tc map[string]any
+	require.NoError(t, json.Unmarshal(resp.ToolChoice, &tc))
+	assert.Equal(t, "function", tc["type"])
+}
+
+func TestChatCompletionsToResponses_ServiceTier(t *testing.T) {
+	req := &ChatCompletionsRequest{
+		Model:       "gpt-4o",
+		ServiceTier: "flex",
+		Messages:    []ChatMessage{{Role: "user", Content: json.RawMessage(`"Hi"`)}},
+	}
+	resp, err := ChatCompletionsToResponses(req)
+	require.NoError(t, err)
+	assert.Equal(t, "flex", resp.ServiceTier)
+}
+
+func TestChatCompletionsToResponses_AssistantWithTextAndToolCalls(t *testing.T) {
+	req := &ChatCompletionsRequest{
+		Model: "gpt-4o",
+		Messages: []ChatMessage{
+			{Role: "user", Content: json.RawMessage(`"Do something"`)},
+			{
+				Role:    "assistant",
+				Content: json.RawMessage(`"Let me call a function."`),
+				ToolCalls: []ChatToolCall{
+					{
+						ID:   "call_abc",
+						Type: "function",
+						Function: ChatFunctionCall{
+							Name:      "do_thing",
+							Arguments: `{}`,
+						},
+					},
+				},
+			},
+		},
+	}
+
+	resp, err := ChatCompletionsToResponses(req)
+	require.NoError(t, err)
+
+	var items []ResponsesInputItem
+	require.NoError(t, json.Unmarshal(resp.Input, &items))
+	// user + assistant message (with text) + function_call
+	require.Len(t, items, 3)
+	assert.Equal(t, "user", items[0].Role)
+	assert.Equal(t, "assistant", items[1].Role)
+	assert.Equal(t, "function_call", items[2].Type)
+	assert.Empty(t, items[2].ID)
+}
+
+func TestChatCompletionsToResponses_AssistantArrayContentPreserved(t *testing.T) {
+	req := &ChatCompletionsRequest{
+		Model: "gpt-4o",
+		Messages: []ChatMessage{
+			{Role: "user", Content: json.RawMessage(`"Hi"`)},
+			{Role: "assistant", Content: json.RawMessage(`[{"type":"text","text":"A"},{"type":"text","text":"B"}]`)},
+		},
+	}
+
+	resp, err := ChatCompletionsToResponses(req)
+	require.NoError(t, err)
+
+	var items []ResponsesInputItem
+	require.NoError(t, json.Unmarshal(resp.Input, &items))
+	require.Len(t, items, 2)
+	assert.Equal(t, "assistant", items[1].Role)
+
+	var parts []ResponsesContentPart
+	require.NoError(t, json.Unmarshal(items[1].Content, &parts))
+	require.Len(t, parts, 1)
+	assert.Equal(t, "output_text", parts[0].Type)
+	assert.Equal(t, "AB", parts[0].Text)
+}
+
+func TestChatCompletionsToResponses_AssistantThinkingTagPreserved(t *testing.T) {
+	req := &ChatCompletionsRequest{
+		Model: "gpt-4o",
+		Messages: []ChatMessage{
+			{Role: "user", Content: json.RawMessage(`"Hi"`)},
+			{Role: "assistant", Content: json.RawMessage(`[{"type":"thinking","thinking":"internal plan"},{"type":"text","text":"final answer"}]`)},
+		},
+	}
+
+	resp, err := ChatCompletionsToResponses(req)
+	require.NoError(t, err)
+
+	var items []ResponsesInputItem
+	require.NoError(t, json.Unmarshal(resp.Input, &items))
+	require.Len(t, items, 2)
+
+	var parts []ResponsesContentPart
+	require.NoError(t, json.Unmarshal(items[1].Content, &parts))
+	require.Len(t, parts, 1)
+	assert.Equal(t, "output_text", parts[0].Type)
+	assert.Contains(t, parts[0].Text, "<thinking>internal plan</thinking>")
+	assert.Contains(t, parts[0].Text, "final answer")
+}
+
+// ---------------------------------------------------------------------------
+// ResponsesToChatCompletions tests
+// ---------------------------------------------------------------------------
+
+func TestResponsesToChatCompletions_BasicText(t *testing.T) {
+	resp := &ResponsesResponse{
+		ID:     "resp_123",
+		Status: "completed",
+		Output: []ResponsesOutput{
+			{
+				Type: "message",
+				Content: []ResponsesContentPart{
+					{Type: "output_text", Text: "Hello, world!"},
+				},
+			},
+		},
+		Usage: &ResponsesUsage{
+			InputTokens:  10,
+			OutputTokens: 5,
+			TotalTokens:  15,
+		},
+	}
+
+	chat := ResponsesToChatCompletions(resp, "gpt-4o")
+	assert.Equal(t, "chat.completion", chat.Object)
+	assert.Equal(t, "gpt-4o", chat.Model)
+	require.Len(t, chat.Choices, 1)
+	assert.Equal(t, "stop", chat.Choices[0].FinishReason)
+
+	var content string
+	require.NoError(t, json.Unmarshal(chat.Choices[0].Message.Content, &content))
+	assert.Equal(t, "Hello, world!", content)
+
+	require.NotNil(t, chat.Usage)
+	assert.Equal(t, 10, chat.Usage.PromptTokens)
+	assert.Equal(t, 5, chat.Usage.CompletionTokens)
+	assert.Equal(t, 15, chat.Usage.TotalTokens)
+}
+
+func TestResponsesToChatCompletions_ToolCalls(t *testing.T) {
+	resp := &ResponsesResponse{
+		ID:     "resp_456",
+		Status: "completed",
+		Output: []ResponsesOutput{
+			{
+				Type:      "function_call",
+				CallID:    "call_xyz",
+				Name:      "get_weather",
+				Arguments: `{"city":"NYC"}`,
+			},
+		},
+	}
+
+	chat := ResponsesToChatCompletions(resp, "gpt-4o")
+	require.Len(t, chat.Choices, 1)
+	assert.Equal(t, "tool_calls", chat.Choices[0].FinishReason)
+
+	msg := chat.Choices[0].Message
+	require.Len(t, msg.ToolCalls, 1)
+	assert.Equal(t, "call_xyz", msg.ToolCalls[0].ID)
+	assert.Equal(t, "function", msg.ToolCalls[0].Type)
+	assert.Equal(t, "get_weather", msg.ToolCalls[0].Function.Name)
+	assert.Equal(t, `{"city":"NYC"}`, msg.ToolCalls[0].Function.Arguments)
+}
+
+func TestResponsesToChatCompletions_Reasoning(t *testing.T) {
+	resp := &ResponsesResponse{
+		ID:     "resp_789",
+		Status: "completed",
+		Output: []ResponsesOutput{
+			{
+				Type: "reasoning",
+				Summary: []ResponsesSummary{
+					{Type: "summary_text", Text: "I thought about it."},
+				},
+			},
+			{
+				Type: "message",
+				Content: []ResponsesContentPart{
+					{Type: "output_text", Text: "The answer is 42."},
+				},
+			},
+		},
+	}
+
+	chat := ResponsesToChatCompletions(resp, "gpt-4o")
+	require.Len(t, chat.Choices, 1)
+
+	var content string
+	require.NoError(t, json.Unmarshal(chat.Choices[0].Message.Content, &content))
+	assert.Equal(t, "The answer is 42.", content)
+	assert.Equal(t, "I thought about it.", chat.Choices[0].Message.ReasoningContent)
+}
+
+func TestResponsesToChatCompletions_Incomplete(t *testing.T) {
+	resp := &ResponsesResponse{
+		ID:                "resp_inc",
+		Status:            "incomplete",
+		IncompleteDetails: &ResponsesIncompleteDetails{Reason: "max_output_tokens"},
+		Output: []ResponsesOutput{
+			{
+				Type: "message",
+				Content: []ResponsesContentPart{
+					{Type: "output_text", Text: "partial..."},
+				},
+			},
+		},
+	}
+
+	chat := ResponsesToChatCompletions(resp, "gpt-4o")
+	require.Len(t, chat.Choices, 1)
+	assert.Equal(t, "length", chat.Choices[0].FinishReason)
+}
+
+func TestResponsesToChatCompletions_CachedTokens(t *testing.T) {
+	resp := &ResponsesResponse{
+		ID:     "resp_cache",
+		Status: "completed",
+		Output: []ResponsesOutput{
+			{
+				Type:    "message",
+				Content: []ResponsesContentPart{{Type: "output_text", Text: "cached"}},
+			},
+		},
+		Usage: &ResponsesUsage{
+			InputTokens:  100,
+			OutputTokens: 10,
+			TotalTokens:  110,
+			InputTokensDetails: &ResponsesInputTokensDetails{
+				CachedTokens: 80,
+			},
+		},
+	}
+
+	chat := ResponsesToChatCompletions(resp, "gpt-4o")
+	require.NotNil(t, chat.Usage)
+	require.NotNil(t, chat.Usage.PromptTokensDetails)
+	assert.Equal(t, 80, chat.Usage.PromptTokensDetails.CachedTokens)
+}
+
+func TestResponsesToChatCompletions_WebSearch(t *testing.T) {
+	resp := &ResponsesResponse{
+		ID:     "resp_ws",
+		Status: "completed",
+		Output: []ResponsesOutput{
+			{
+				Type:   "web_search_call",
+				Action: &WebSearchAction{Type: "search", Query: "test"},
+			},
+			{
+				Type:    "message",
+				Content: []ResponsesContentPart{{Type: "output_text", Text: "search results"}},
+			},
+		},
+	}
+
+	chat := ResponsesToChatCompletions(resp, "gpt-4o")
+	require.Len(t, chat.Choices, 1)
+	assert.Equal(t, "stop", chat.Choices[0].FinishReason)
+
+	var content string
+	require.NoError(t, json.Unmarshal(chat.Choices[0].Message.Content, &content))
+	assert.Equal(t, "search results", content)
+}
+
+// ---------------------------------------------------------------------------
+// Streaming: ResponsesEventToChatChunks tests
+// ---------------------------------------------------------------------------
+
+func TestResponsesEventToChatChunks_TextDelta(t *testing.T) {
+	state := NewResponsesEventToChatState()
+	state.Model = "gpt-4o"
+
+	// response.created → role chunk
+	chunks := ResponsesEventToChatChunks(&ResponsesStreamEvent{
+		Type: "response.created",
+		Response: &ResponsesResponse{
+			ID: "resp_stream",
+		},
+	}, state)
+	require.Len(t, chunks, 1)
+	assert.Equal(t, "assistant", chunks[0].Choices[0].Delta.Role)
+	assert.True(t, state.SentRole)
+
+	// response.output_text.delta → content chunk
+	chunks = ResponsesEventToChatChunks(&ResponsesStreamEvent{
+		Type:  "response.output_text.delta",
+		Delta: "Hello",
+	}, state)
+	require.Len(t, chunks, 1)
+	require.NotNil(t, chunks[0].Choices[0].Delta.Content)
+	assert.Equal(t, "Hello", *chunks[0].Choices[0].Delta.Content)
+}
+
+func TestResponsesEventToChatChunks_ToolCallDelta(t *testing.T) {
+	state := NewResponsesEventToChatState()
+	state.Model = "gpt-4o"
+	state.SentRole = true
+
+	// response.output_item.added (function_call) — output_index=1 (e.g. after a message item at 0)
+	chunks := ResponsesEventToChatChunks(&ResponsesStreamEvent{
+		Type:        "response.output_item.added",
+		OutputIndex: 1,
+		Item: &ResponsesOutput{
+			Type:   "function_call",
+			CallID: "call_1",
+			Name:   "get_weather",
+		},
+	}, state)
+	require.Len(t, chunks, 1)
+	require.Len(t, chunks[0].Choices[0].Delta.ToolCalls, 1)
+	tc := chunks[0].Choices[0].Delta.ToolCalls[0]
+	assert.Equal(t, "call_1", tc.ID)
+	assert.Equal(t, "get_weather", tc.Function.Name)
+	require.NotNil(t, tc.Index)
+	assert.Equal(t, 0, *tc.Index)
+
+	// response.function_call_arguments.delta — uses output_index (NOT call_id) to find tool
+	chunks = ResponsesEventToChatChunks(&ResponsesStreamEvent{
+		Type:        "response.function_call_arguments.delta",
+		OutputIndex: 1, // matches the output_index from output_item.added above
+		Delta:       `{"city":`,
+	}, state)
+	require.Len(t, chunks, 1)
+	tc = chunks[0].Choices[0].Delta.ToolCalls[0]
+	require.NotNil(t, tc.Index)
+	assert.Equal(t, 0, *tc.Index, "argument delta must use same index as the tool call")
+	assert.Equal(t, `{"city":`, tc.Function.Arguments)
+
+	// Add a second function call at output_index=2
+	chunks = ResponsesEventToChatChunks(&ResponsesStreamEvent{
+		Type:        "response.output_item.added",
+		OutputIndex: 2,
+		Item: &ResponsesOutput{
+			Type:   "function_call",
+			CallID: "call_2",
+			Name:   "get_time",
+		},
+	}, state)
+	require.Len(t, chunks, 1)
+	tc = chunks[0].Choices[0].Delta.ToolCalls[0]
+	require.NotNil(t, tc.Index)
+	assert.Equal(t, 1, *tc.Index, "second tool call should get index 1")
+
+	// Argument delta for second tool call
+	chunks = ResponsesEventToChatChunks(&ResponsesStreamEvent{
+		Type:        "response.function_call_arguments.delta",
+		OutputIndex: 2,
+		Delta:       `{"tz":"UTC"}`,
+	}, state)
+	require.Len(t, chunks, 1)
+	tc = chunks[0].Choices[0].Delta.ToolCalls[0]
+	require.NotNil(t, tc.Index)
+	assert.Equal(t, 1, *tc.Index, "second tool arg delta must use index 1")
+
+	// Argument delta for first tool call (interleaved)
+	chunks = ResponsesEventToChatChunks(&ResponsesStreamEvent{
+		Type:        "response.function_call_arguments.delta",
+		OutputIndex: 1,
+		Delta:       `"Tokyo"}`,
+	}, state)
+	require.Len(t, chunks, 1)
+	tc = chunks[0].Choices[0].Delta.ToolCalls[0]
+	require.NotNil(t, tc.Index)
+	assert.Equal(t, 0, *tc.Index, "first tool arg delta must still use index 0")
+}
+
+func TestResponsesEventToChatChunks_Completed(t *testing.T) {
+	state := NewResponsesEventToChatState()
+	state.Model = "gpt-4o"
+	state.IncludeUsage = true
+
+	chunks := ResponsesEventToChatChunks(&ResponsesStreamEvent{
+		Type: "response.completed",
+		Response: &ResponsesResponse{
+			Status: "completed",
+			Usage: &ResponsesUsage{
+				InputTokens:  50,
+				OutputTokens: 20,
+				TotalTokens:  70,
+				InputTokensDetails: &ResponsesInputTokensDetails{
+					CachedTokens: 30,
+				},
+			},
+		},
+	}, state)
+	// finish chunk + usage chunk
+	require.Len(t, chunks, 2)
+
+	// First chunk: finish_reason
+	require.NotNil(t, chunks[0].Choices[0].FinishReason)
+	assert.Equal(t, "stop", *chunks[0].Choices[0].FinishReason)
+
+	// Second chunk: usage
+	require.NotNil(t, chunks[1].Usage)
+	assert.Equal(t, 50, chunks[1].Usage.PromptTokens)
+	assert.Equal(t, 20, chunks[1].Usage.CompletionTokens)
+	assert.Equal(t, 70, chunks[1].Usage.TotalTokens)
+	require.NotNil(t, chunks[1].Usage.PromptTokensDetails)
+	assert.Equal(t, 30, chunks[1].Usage.PromptTokensDetails.CachedTokens)
+}
+
+func TestResponsesEventToChatChunks_CompletedWithToolCalls(t *testing.T) {
+	state := NewResponsesEventToChatState()
+	state.Model = "gpt-4o"
+	state.SawToolCall = true
+
+	chunks := ResponsesEventToChatChunks(&ResponsesStreamEvent{
+		Type: "response.completed",
+		Response: &ResponsesResponse{
+			Status: "completed",
+		},
+	}, state)
+	require.Len(t, chunks, 1)
+	require.NotNil(t, chunks[0].Choices[0].FinishReason)
+	assert.Equal(t, "tool_calls", *chunks[0].Choices[0].FinishReason)
+}
+
+func TestResponsesEventToChatChunks_ReasoningDelta(t *testing.T) {
+	state := NewResponsesEventToChatState()
+	state.Model = "gpt-4o"
+	state.SentRole = true
+
+	chunks := ResponsesEventToChatChunks(&ResponsesStreamEvent{
+		Type:  "response.reasoning_summary_text.delta",
+		Delta: "Thinking...",
+	}, state)
+	require.Len(t, chunks, 1)
+	require.NotNil(t, chunks[0].Choices[0].Delta.ReasoningContent)
+	assert.Equal(t, "Thinking...", *chunks[0].Choices[0].Delta.ReasoningContent)
+
+	chunks = ResponsesEventToChatChunks(&ResponsesStreamEvent{
+		Type: "response.reasoning_summary_text.done",
+	}, state)
+	require.Len(t, chunks, 0)
+}
+
+func TestResponsesEventToChatChunks_ReasoningThenTextAutoCloseTag(t *testing.T) {
+	state := NewResponsesEventToChatState()
+	state.Model = "gpt-4o"
+	state.SentRole = true
+
+	chunks := ResponsesEventToChatChunks(&ResponsesStreamEvent{
+		Type:  "response.reasoning_summary_text.delta",
+		Delta: "plan",
+	}, state)
+	require.Len(t, chunks, 1)
+	require.NotNil(t, chunks[0].Choices[0].Delta.ReasoningContent)
+	assert.Equal(t, "plan", *chunks[0].Choices[0].Delta.ReasoningContent)
+
+	chunks = ResponsesEventToChatChunks(&ResponsesStreamEvent{
+		Type:  "response.output_text.delta",
+		Delta: "answer",
+	}, state)
+	require.Len(t, chunks, 1)
+	require.NotNil(t, chunks[0].Choices[0].Delta.Content)
+	assert.Equal(t, "answer", *chunks[0].Choices[0].Delta.Content)
+}
+
+func TestFinalizeResponsesChatStream(t *testing.T) {
+	state := NewResponsesEventToChatState()
+	state.Model = "gpt-4o"
+	state.IncludeUsage = true
+	state.Usage = &ChatUsage{
+		PromptTokens:     100,
+		CompletionTokens: 50,
+		TotalTokens:      150,
+	}
+
+	chunks := FinalizeResponsesChatStream(state)
+	require.Len(t, chunks, 2)
+
+	// Finish chunk
+	require.NotNil(t, chunks[0].Choices[0].FinishReason)
+	assert.Equal(t, "stop", *chunks[0].Choices[0].FinishReason)
+
+	// Usage chunk
+	require.NotNil(t, chunks[1].Usage)
+	assert.Equal(t, 100, chunks[1].Usage.PromptTokens)
+
+	// Idempotent: second call returns nil
+	assert.Nil(t, FinalizeResponsesChatStream(state))
+}
+
+func TestFinalizeResponsesChatStream_AfterCompleted(t *testing.T) {
+	// If response.completed already emitted the finish chunk, FinalizeResponsesChatStream
+	// must be a no-op (prevents double finish_reason being sent to the client).
+	state := NewResponsesEventToChatState()
+	state.Model = "gpt-4o"
+	state.IncludeUsage = true
+
+	// Simulate response.completed
+	chunks := ResponsesEventToChatChunks(&ResponsesStreamEvent{
+		Type: "response.completed",
+		Response: &ResponsesResponse{
+			Status: "completed",
+			Usage: &ResponsesUsage{
+				InputTokens:  10,
+				OutputTokens: 5,
+				TotalTokens:  15,
+			},
+		},
+	}, state)
+	require.NotEmpty(t, chunks) // finish + usage chunks
+
+	// Now FinalizeResponsesChatStream should return nil — already finalized.
+	assert.Nil(t, FinalizeResponsesChatStream(state))
+}
+
+func TestChatChunkToSSE(t *testing.T) {
+	chunk := ChatCompletionsChunk{
+		ID:      "chatcmpl-test",
+		Object:  "chat.completion.chunk",
+		Created: 1700000000,
+		Model:   "gpt-4o",
+		Choices: []ChatChunkChoice{
+			{
+				Index:        0,
+				Delta:        ChatDelta{Role: "assistant"},
+				FinishReason: nil,
+			},
+		},
+	}
+
+	sse, err := ChatChunkToSSE(chunk)
+	require.NoError(t, err)
+	assert.Contains(t, sse, "data: ")
+	assert.Contains(t, sse, "chatcmpl-test")
+	assert.Contains(t, sse, "assistant")
+	assert.True(t, len(sse) > 10)
+}
+
+// ---------------------------------------------------------------------------
+// Stream round-trip test
+// ---------------------------------------------------------------------------
+
+func TestChatCompletionsStreamRoundTrip(t *testing.T) {
+	// Simulate: client sends chat completions request, upstream returns Responses SSE events.
+	// Verify that the streaming state machine produces correct chat completions chunks.
+
+	state := NewResponsesEventToChatState()
+	state.Model = "gpt-4o"
+	state.IncludeUsage = true
+
+	var allChunks []ChatCompletionsChunk
+
+	// 1. response.created
+	chunks := ResponsesEventToChatChunks(&ResponsesStreamEvent{
+		Type:     "response.created",
+		Response: &ResponsesResponse{ID: "resp_rt"},
+	}, state)
+	allChunks = append(allChunks, chunks...)
+
+	// 2. text deltas
+	for _, text := range []string{"Hello", ", ", "world", "!"} {
+		chunks = ResponsesEventToChatChunks(&ResponsesStreamEvent{
+			Type:  "response.output_text.delta",
+			Delta: text,
+		}, state)
+		allChunks = append(allChunks, chunks...)
+	}
+
+	// 3. response.completed
+	chunks = ResponsesEventToChatChunks(&ResponsesStreamEvent{
+		Type: "response.completed",
+		Response: &ResponsesResponse{
+			Status: "completed",
+			Usage: &ResponsesUsage{
+				InputTokens:  10,
+				OutputTokens: 4,
+				TotalTokens:  14,
+			},
+		},
+	}, state)
+	allChunks = append(allChunks, chunks...)
+
+	// Verify: role chunk + 4 text chunks + finish chunk + usage chunk = 7
+	require.Len(t, allChunks, 7)
+
+	// First chunk has role
+	assert.Equal(t, "assistant", allChunks[0].Choices[0].Delta.Role)
+
+	// Text chunks
+	var fullText string
+	for i := 1; i <= 4; i++ {
+		require.NotNil(t, allChunks[i].Choices[0].Delta.Content)
+		fullText += *allChunks[i].Choices[0].Delta.Content
+	}
+	assert.Equal(t, "Hello, world!", fullText)
+
+	// Finish chunk
+	require.NotNil(t, allChunks[5].Choices[0].FinishReason)
+	assert.Equal(t, "stop", *allChunks[5].Choices[0].FinishReason)
+
+	// Usage chunk
+	require.NotNil(t, allChunks[6].Usage)
+	assert.Equal(t, 10, allChunks[6].Usage.PromptTokens)
+	assert.Equal(t, 4, allChunks[6].Usage.CompletionTokens)
+
+	// All chunks share the same ID
+	for _, c := range allChunks {
+		assert.Equal(t, "resp_rt", c.ID)
+	}
+}

backend/internal/pkg/apicompat/chatcompletions_to_responses.go

@@ -0,0 +1,385 @@
+package apicompat
+
+import (
+	"encoding/json"
+	"fmt"
+	"strings"
+)
+
+// ChatCompletionsToResponses converts a Chat Completions request into a
+// Responses API request. The upstream always streams, so Stream is forced to
+// true. store is always false and reasoning.encrypted_content is always
+// included so that the response translator has full context.
+func ChatCompletionsToResponses(req *ChatCompletionsRequest) (*ResponsesRequest, error) {
+	input, err := convertChatMessagesToResponsesInput(req.Messages)
+	if err != nil {
+		return nil, err
+	}
+
+	inputJSON, err := json.Marshal(input)
+	if err != nil {
+		return nil, err
+	}
+
+	out := &ResponsesRequest{
+		Model:       req.Model,
+		Input:       inputJSON,
+		Temperature: req.Temperature,
+		TopP:        req.TopP,
+		Stream:      true, // upstream always streams
+		Include:     []string{"reasoning.encrypted_content"},
+		ServiceTier: req.ServiceTier,
+	}
+
+	storeFalse := false
+	out.Store = &storeFalse
+
+	// max_tokens / max_completion_tokens → max_output_tokens, prefer max_completion_tokens
+	maxTokens := 0
+	if req.MaxTokens != nil {
+		maxTokens = *req.MaxTokens
+	}
+	if req.MaxCompletionTokens != nil {
+		maxTokens = *req.MaxCompletionTokens
+	}
+	if maxTokens > 0 {
+		v := maxTokens
+		if v < minMaxOutputTokens {
+			v = minMaxOutputTokens
+		}
+		out.MaxOutputTokens = &v
+	}
+
+	// reasoning_effort → reasoning.effort + reasoning.summary="auto"
+	if req.ReasoningEffort != "" {
+		out.Reasoning = &ResponsesReasoning{
+			Effort:  req.ReasoningEffort,
+			Summary: "auto",
+		}
+	}
+
+	// tools[] and legacy functions[] → ResponsesTool[]
+	if len(req.Tools) > 0 || len(req.Functions) > 0 {
+		out.Tools = convertChatToolsToResponses(req.Tools, req.Functions)
+	}
+
+	// tool_choice: already compatible format — pass through directly.
+	// Legacy function_call needs mapping.
+	if len(req.ToolChoice) > 0 {
+		out.ToolChoice = req.ToolChoice
+	} else if len(req.FunctionCall) > 0 {
+		tc, err := convertChatFunctionCallToToolChoice(req.FunctionCall)
+		if err != nil {
+			return nil, fmt.Errorf("convert function_call: %w", err)
+		}
+		out.ToolChoice = tc
+	}
+
+	return out, nil
+}
+
+// convertChatMessagesToResponsesInput converts the Chat Completions messages
+// array into a Responses API input items array.
+func convertChatMessagesToResponsesInput(msgs []ChatMessage) ([]ResponsesInputItem, error) {
+	var out []ResponsesInputItem
+	for _, m := range msgs {
+		items, err := chatMessageToResponsesItems(m)
+		if err != nil {
+			return nil, err
+		}
+		out = append(out, items...)
+	}
+	return out, nil
+}
+
+// chatMessageToResponsesItems converts a single ChatMessage into one or more
+// ResponsesInputItem values.
+func chatMessageToResponsesItems(m ChatMessage) ([]ResponsesInputItem, error) {
+	switch m.Role {
+	case "system":
+		return chatSystemToResponses(m)
+	case "user":
+		return chatUserToResponses(m)
+	case "assistant":
+		return chatAssistantToResponses(m)
+	case "tool":
+		return chatToolToResponses(m)
+	case "function":
+		return chatFunctionToResponses(m)
+	default:
+		return chatUserToResponses(m)
+	}
+}
+
+// chatSystemToResponses converts a system message.
+func chatSystemToResponses(m ChatMessage) ([]ResponsesInputItem, error) {
+	text, err := parseChatContent(m.Content)
+	if err != nil {
+		return nil, err
+	}
+	content, err := json.Marshal(text)
+	if err != nil {
+		return nil, err
+	}
+	return []ResponsesInputItem{{Role: "system", Content: content}}, nil
+}
+
+// chatUserToResponses converts a user message, handling both plain strings and
+// multi-modal content arrays.
+func chatUserToResponses(m ChatMessage) ([]ResponsesInputItem, error) {
+	// Try plain string first.
+	var s string
+	if err := json.Unmarshal(m.Content, &s); err == nil {
+		content, _ := json.Marshal(s)
+		return []ResponsesInputItem{{Role: "user", Content: content}}, nil
+	}
+
+	var parts []ChatContentPart
+	if err := json.Unmarshal(m.Content, &parts); err != nil {
+		return nil, fmt.Errorf("parse user content: %w", err)
+	}
+
+	var responseParts []ResponsesContentPart
+	for _, p := range parts {
+		switch p.Type {
+		case "text":
+			if p.Text != "" {
+				responseParts = append(responseParts, ResponsesContentPart{
+					Type: "input_text",
+					Text: p.Text,
+				})
+			}
+		case "image_url":
+			if p.ImageURL != nil && p.ImageURL.URL != "" {
+				responseParts = append(responseParts, ResponsesContentPart{
+					Type:     "input_image",
+					ImageURL: p.ImageURL.URL,
+				})
+			}
+		}
+	}
+
+	content, err := json.Marshal(responseParts)
+	if err != nil {
+		return nil, err
+	}
+	return []ResponsesInputItem{{Role: "user", Content: content}}, nil
+}
+
+// chatAssistantToResponses converts an assistant message. If there is both
+// text content and tool_calls, the text is emitted as an assistant message
+// first, then each tool_call becomes a function_call item. If the content is
+// empty/nil and there are tool_calls, only function_call items are emitted.
+func chatAssistantToResponses(m ChatMessage) ([]ResponsesInputItem, error) {
+	var items []ResponsesInputItem
+
+	// Emit assistant message with output_text if content is non-empty.
+	if len(m.Content) > 0 {
+		s, err := parseAssistantContent(m.Content)
+		if err != nil {
+			return nil, err
+		}
+		if s != "" {
+			parts := []ResponsesContentPart{{Type: "output_text", Text: s}}
+			partsJSON, err := json.Marshal(parts)
+			if err != nil {
+				return nil, err
+			}
+			items = append(items, ResponsesInputItem{Role: "assistant", Content: partsJSON})
+		}
+	}
+
+	// Emit one function_call item per tool_call.
+	for _, tc := range m.ToolCalls {
+		args := tc.Function.Arguments
+		if args == "" {
+			args = "{}"
+		}
+		items = append(items, ResponsesInputItem{
+			Type:      "function_call",
+			CallID:    tc.ID,
+			Name:      tc.Function.Name,
+			Arguments: args,
+		})
+	}
+
+	return items, nil
+}
+
+// parseAssistantContent returns assistant content as plain text.
+//
+// Supported formats:
+// - JSON string
+// - JSON array of typed parts (e.g. [{"type":"text","text":"..."}])
+//
+// For structured thinking/reasoning parts, it preserves semantics by wrapping
+// the text in explicit tags so downstream can still distinguish it from normal text.
+func parseAssistantContent(raw json.RawMessage) (string, error) {
+	if len(raw) == 0 {
+		return "", nil
+	}
+
+	var s string
+	if err := json.Unmarshal(raw, &s); err == nil {
+		return s, nil
+	}
+
+	var parts []map[string]any
+	if err := json.Unmarshal(raw, &parts); err != nil {
+		// Keep compatibility with prior behavior: unsupported assistant content
+		// formats are ignored instead of failing the whole request conversion.
+		return "", nil
+	}
+
+	var b strings.Builder
+	write := func(v string) error {
+		_, err := b.WriteString(v)
+		return err
+	}
+	for _, p := range parts {
+		typ, _ := p["type"].(string)
+		text, _ := p["text"].(string)
+		thinking, _ := p["thinking"].(string)
+
+		switch typ {
+		case "thinking", "reasoning":
+			if thinking != "" {
+				if err := write("<thinking>"); err != nil {
+					return "", err
+				}
+				if err := write(thinking); err != nil {
+					return "", err
+				}
+				if err := write("</thinking>"); err != nil {
+					return "", err
+				}
+			} else if text != "" {
+				if err := write("<thinking>"); err != nil {
+					return "", err
+				}
+				if err := write(text); err != nil {
+					return "", err
+				}
+				if err := write("</thinking>"); err != nil {
+					return "", err
+				}
+			}
+		default:
+			if text != "" {
+				if err := write(text); err != nil {
+					return "", err
+				}
+			}
+		}
+	}
+
+	return b.String(), nil
+}
+
+// chatToolToResponses converts a tool result message (role=tool) into a
+// function_call_output item.
+func chatToolToResponses(m ChatMessage) ([]ResponsesInputItem, error) {
+	output, err := parseChatContent(m.Content)
+	if err != nil {
+		return nil, err
+	}
+	if output == "" {
+		output = "(empty)"
+	}
+	return []ResponsesInputItem{{
+		Type:   "function_call_output",
+		CallID: m.ToolCallID,
+		Output: output,
+	}}, nil
+}
+
+// chatFunctionToResponses converts a legacy function result message
+// (role=function) into a function_call_output item. The Name field is used as
+// call_id since legacy function calls do not carry a separate call_id.
+func chatFunctionToResponses(m ChatMessage) ([]ResponsesInputItem, error) {
+	output, err := parseChatContent(m.Content)
+	if err != nil {
+		return nil, err
+	}
+	if output == "" {
+		output = "(empty)"
+	}
+	return []ResponsesInputItem{{
+		Type:   "function_call_output",
+		CallID: m.Name,
+		Output: output,
+	}}, nil
+}
+
+// parseChatContent returns the string value of a ChatMessage Content field.
+// Content must be a JSON string. Returns "" if content is null or empty.
+func parseChatContent(raw json.RawMessage) (string, error) {
+	if len(raw) == 0 {
+		return "", nil
+	}
+	var s string
+	if err := json.Unmarshal(raw, &s); err != nil {
+		return "", fmt.Errorf("parse content as string: %w", err)
+	}
+	return s, nil
+}
+
+// convertChatToolsToResponses maps Chat Completions tool definitions and legacy
+// function definitions to Responses API tool definitions.
+func convertChatToolsToResponses(tools []ChatTool, functions []ChatFunction) []ResponsesTool {
+	var out []ResponsesTool
+
+	for _, t := range tools {
+		if t.Type != "function" || t.Function == nil {
+			continue
+		}
+		rt := ResponsesTool{
+			Type:        "function",
+			Name:        t.Function.Name,
+			Description: t.Function.Description,
+			Parameters:  t.Function.Parameters,
+			Strict:      t.Function.Strict,
+		}
+		out = append(out, rt)
+	}
+
+	// Legacy functions[] are treated as function-type tools.
+	for _, f := range functions {
+		rt := ResponsesTool{
+			Type:        "function",
+			Name:        f.Name,
+			Description: f.Description,
+			Parameters:  f.Parameters,
+			Strict:      f.Strict,
+		}
+		out = append(out, rt)
+	}
+
+	return out
+}
+
+// convertChatFunctionCallToToolChoice maps the legacy function_call field to a
+// Responses API tool_choice value.
+//
+//	"auto" → "auto"
+//	"none" → "none"
+//	{"name":"X"} → {"type":"function","function":{"name":"X"}}
+func convertChatFunctionCallToToolChoice(raw json.RawMessage) (json.RawMessage, error) {
+	// Try string first ("auto", "none", etc.) — pass through as-is.
+	var s string
+	if err := json.Unmarshal(raw, &s); err == nil {
+		return json.Marshal(s)
+	}
+
+	// Object form: {"name":"X"}
+	var obj struct {
+		Name string `json:"name"`
+	}
+	if err := json.Unmarshal(raw, &obj); err != nil {
+		return nil, err
+	}
+	return json.Marshal(map[string]any{
+		"type":     "function",
+		"function": map[string]string{"name": obj.Name},
+	})
+}

backend/internal/pkg/apicompat/responses_to_anthropic.go

@@ -0,0 +1,516 @@
+package apicompat
+
+import (
+	"encoding/json"
+	"fmt"
+	"time"
+)
+
+// ---------------------------------------------------------------------------
+// Non-streaming: ResponsesResponse → AnthropicResponse
+// ---------------------------------------------------------------------------
+
+// ResponsesToAnthropic converts a Responses API response directly into an
+// Anthropic Messages response. Reasoning output items are mapped to thinking
+// blocks; function_call items become tool_use blocks.
+func ResponsesToAnthropic(resp *ResponsesResponse, model string) *AnthropicResponse {
+	out := &AnthropicResponse{
+		ID:    resp.ID,
+		Type:  "message",
+		Role:  "assistant",
+		Model: model,
+	}
+
+	var blocks []AnthropicContentBlock
+
+	for _, item := range resp.Output {
+		switch item.Type {
+		case "reasoning":
+			summaryText := ""
+			for _, s := range item.Summary {
+				if s.Type == "summary_text" && s.Text != "" {
+					summaryText += s.Text
+				}
+			}
+			if summaryText != "" {
+				blocks = append(blocks, AnthropicContentBlock{
+					Type:     "thinking",
+					Thinking: summaryText,
+				})
+			}
+		case "message":
+			for _, part := range item.Content {
+				if part.Type == "output_text" && part.Text != "" {
+					blocks = append(blocks, AnthropicContentBlock{
+						Type: "text",
+						Text: part.Text,
+					})
+				}
+			}
+		case "function_call":
+			blocks = append(blocks, AnthropicContentBlock{
+				Type:  "tool_use",
+				ID:    fromResponsesCallID(item.CallID),
+				Name:  item.Name,
+				Input: json.RawMessage(item.Arguments),
+			})
+		case "web_search_call":
+			toolUseID := "srvtoolu_" + item.ID
+			query := ""
+			if item.Action != nil {
+				query = item.Action.Query
+			}
+			inputJSON, _ := json.Marshal(map[string]string{"query": query})
+			blocks = append(blocks, AnthropicContentBlock{
+				Type:  "server_tool_use",
+				ID:    toolUseID,
+				Name:  "web_search",
+				Input: inputJSON,
+			})
+			emptyResults, _ := json.Marshal([]struct{}{})
+			blocks = append(blocks, AnthropicContentBlock{
+				Type:      "web_search_tool_result",
+				ToolUseID: toolUseID,
+				Content:   emptyResults,
+			})
+		}
+	}
+
+	if len(blocks) == 0 {
+		blocks = append(blocks, AnthropicContentBlock{Type: "text", Text: ""})
+	}
+	out.Content = blocks
+
+	out.StopReason = responsesStatusToAnthropicStopReason(resp.Status, resp.IncompleteDetails, blocks)
+
+	if resp.Usage != nil {
+		out.Usage = AnthropicUsage{
+			InputTokens:  resp.Usage.InputTokens,
+			OutputTokens: resp.Usage.OutputTokens,
+		}
+		if resp.Usage.InputTokensDetails != nil {
+			out.Usage.CacheReadInputTokens = resp.Usage.InputTokensDetails.CachedTokens
+		}
+	}
+
+	return out
+}
+
+func responsesStatusToAnthropicStopReason(status string, details *ResponsesIncompleteDetails, blocks []AnthropicContentBlock) string {
+	switch status {
+	case "incomplete":
+		if details != nil && details.Reason == "max_output_tokens" {
+			return "max_tokens"
+		}
+		return "end_turn"
+	case "completed":
+		if len(blocks) > 0 && blocks[len(blocks)-1].Type == "tool_use" {
+			return "tool_use"
+		}
+		return "end_turn"
+	default:
+		return "end_turn"
+	}
+}
+
+// ---------------------------------------------------------------------------
+// Streaming: ResponsesStreamEvent → []AnthropicStreamEvent (stateful converter)
+// ---------------------------------------------------------------------------
+
+// ResponsesEventToAnthropicState tracks state for converting a sequence of
+// Responses SSE events directly into Anthropic SSE events.
+type ResponsesEventToAnthropicState struct {
+	MessageStartSent bool
+	MessageStopSent  bool
+
+	ContentBlockIndex int
+	ContentBlockOpen  bool
+	CurrentBlockType  string // "text" | "thinking" | "tool_use"
+
+	// OutputIndexToBlockIdx maps Responses output_index → Anthropic content block index.
+	OutputIndexToBlockIdx map[int]int
+
+	InputTokens          int
+	OutputTokens         int
+	CacheReadInputTokens int
+
+	ResponseID string
+	Model      string
+	Created    int64
+}
+
+// NewResponsesEventToAnthropicState returns an initialised stream state.
+func NewResponsesEventToAnthropicState() *ResponsesEventToAnthropicState {
+	return &ResponsesEventToAnthropicState{
+		OutputIndexToBlockIdx: make(map[int]int),
+		Created:               time.Now().Unix(),
+	}
+}
+
+// ResponsesEventToAnthropicEvents converts a single Responses SSE event into
+// zero or more Anthropic SSE events, updating state as it goes.
+func ResponsesEventToAnthropicEvents(
+	evt *ResponsesStreamEvent,
+	state *ResponsesEventToAnthropicState,
+) []AnthropicStreamEvent {
+	switch evt.Type {
+	case "response.created":
+		return resToAnthHandleCreated(evt, state)
+	case "response.output_item.added":
+		return resToAnthHandleOutputItemAdded(evt, state)
+	case "response.output_text.delta":
+		return resToAnthHandleTextDelta(evt, state)
+	case "response.output_text.done":
+		return resToAnthHandleBlockDone(state)
+	case "response.function_call_arguments.delta":
+		return resToAnthHandleFuncArgsDelta(evt, state)
+	case "response.function_call_arguments.done":
+		return resToAnthHandleBlockDone(state)
+	case "response.output_item.done":
+		return resToAnthHandleOutputItemDone(evt, state)
+	case "response.reasoning_summary_text.delta":
+		return resToAnthHandleReasoningDelta(evt, state)
+	case "response.reasoning_summary_text.done":
+		return resToAnthHandleBlockDone(state)
+	case "response.completed", "response.incomplete", "response.failed":
+		return resToAnthHandleCompleted(evt, state)
+	default:
+		return nil
+	}
+}
+
+// FinalizeResponsesAnthropicStream emits synthetic termination events if the
+// stream ended without a proper completion event.
+func FinalizeResponsesAnthropicStream(state *ResponsesEventToAnthropicState) []AnthropicStreamEvent {
+	if !state.MessageStartSent || state.MessageStopSent {
+		return nil
+	}
+
+	var events []AnthropicStreamEvent
+	events = append(events, closeCurrentBlock(state)...)
+
+	events = append(events,
+		AnthropicStreamEvent{
+			Type: "message_delta",
+			Delta: &AnthropicDelta{
+				StopReason: "end_turn",
+			},
+			Usage: &AnthropicUsage{
+				InputTokens:          state.InputTokens,
+				OutputTokens:         state.OutputTokens,
+				CacheReadInputTokens: state.CacheReadInputTokens,
+			},
+		},
+		AnthropicStreamEvent{Type: "message_stop"},
+	)
+	state.MessageStopSent = true
+	return events
+}
+
+// ResponsesAnthropicEventToSSE formats an AnthropicStreamEvent as an SSE line pair.
+func ResponsesAnthropicEventToSSE(evt AnthropicStreamEvent) (string, error) {
+	data, err := json.Marshal(evt)
+	if err != nil {
+		return "", err
+	}
+	return fmt.Sprintf("event: %s\ndata: %s\n\n", evt.Type, data), nil
+}
+
+// --- internal handlers ---
+
+func resToAnthHandleCreated(evt *ResponsesStreamEvent, state *ResponsesEventToAnthropicState) []AnthropicStreamEvent {
+	if evt.Response != nil {
+		state.ResponseID = evt.Response.ID
+		// Only use upstream model if no override was set (e.g. originalModel)
+		if state.Model == "" {
+			state.Model = evt.Response.Model
+		}
+	}
+
+	if state.MessageStartSent {
+		return nil
+	}
+	state.MessageStartSent = true
+
+	return []AnthropicStreamEvent{{
+		Type: "message_start",
+		Message: &AnthropicResponse{
+			ID:      state.ResponseID,
+			Type:    "message",
+			Role:    "assistant",
+			Content: []AnthropicContentBlock{},
+			Model:   state.Model,
+			Usage: AnthropicUsage{
+				InputTokens:  0,
+				OutputTokens: 0,
+			},
+		},
+	}}
+}
+
+func resToAnthHandleOutputItemAdded(evt *ResponsesStreamEvent, state *ResponsesEventToAnthropicState) []AnthropicStreamEvent {
+	if evt.Item == nil {
+		return nil
+	}
+
+	switch evt.Item.Type {
+	case "function_call":
+		var events []AnthropicStreamEvent
+		events = append(events, closeCurrentBlock(state)...)
+
+		idx := state.ContentBlockIndex
+		state.OutputIndexToBlockIdx[evt.OutputIndex] = idx
+		state.ContentBlockOpen = true
+		state.CurrentBlockType = "tool_use"
+
+		events = append(events, AnthropicStreamEvent{
+			Type:  "content_block_start",
+			Index: &idx,
+			ContentBlock: &AnthropicContentBlock{
+				Type:  "tool_use",
+				ID:    fromResponsesCallID(evt.Item.CallID),
+				Name:  evt.Item.Name,
+				Input: json.RawMessage("{}"),
+			},
+		})
+		return events
+
+	case "reasoning":
+		var events []AnthropicStreamEvent
+		events = append(events, closeCurrentBlock(state)...)
+
+		idx := state.ContentBlockIndex
+		state.OutputIndexToBlockIdx[evt.OutputIndex] = idx
+		state.ContentBlockOpen = true
+		state.CurrentBlockType = "thinking"
+
+		events = append(events, AnthropicStreamEvent{
+			Type:  "content_block_start",
+			Index: &idx,
+			ContentBlock: &AnthropicContentBlock{
+				Type:     "thinking",
+				Thinking: "",
+			},
+		})
+		return events
+
+	case "message":
+		return nil
+	}
+
+	return nil
+}
+
+func resToAnthHandleTextDelta(evt *ResponsesStreamEvent, state *ResponsesEventToAnthropicState) []AnthropicStreamEvent {
+	if evt.Delta == "" {
+		return nil
+	}
+
+	var events []AnthropicStreamEvent
+
+	if !state.ContentBlockOpen || state.CurrentBlockType != "text" {
+		events = append(events, closeCurrentBlock(state)...)
+
+		idx := state.ContentBlockIndex
+		state.ContentBlockOpen = true
+		state.CurrentBlockType = "text"
+
+		events = append(events, AnthropicStreamEvent{
+			Type:  "content_block_start",
+			Index: &idx,
+			ContentBlock: &AnthropicContentBlock{
+				Type: "text",
+				Text: "",
+			},
+		})
+	}
+
+	idx := state.ContentBlockIndex
+	events = append(events, AnthropicStreamEvent{
+		Type:  "content_block_delta",
+		Index: &idx,
+		Delta: &AnthropicDelta{
+			Type: "text_delta",
+			Text: evt.Delta,
+		},
+	})
+	return events
+}
+
+func resToAnthHandleFuncArgsDelta(evt *ResponsesStreamEvent, state *ResponsesEventToAnthropicState) []AnthropicStreamEvent {
+	if evt.Delta == "" {
+		return nil
+	}
+
+	blockIdx, ok := state.OutputIndexToBlockIdx[evt.OutputIndex]
+	if !ok {
+		return nil
+	}
+
+	return []AnthropicStreamEvent{{
+		Type:  "content_block_delta",
+		Index: &blockIdx,
+		Delta: &AnthropicDelta{
+			Type:        "input_json_delta",
+			PartialJSON: evt.Delta,
+		},
+	}}
+}
+
+func resToAnthHandleReasoningDelta(evt *ResponsesStreamEvent, state *ResponsesEventToAnthropicState) []AnthropicStreamEvent {
+	if evt.Delta == "" {
+		return nil
+	}
+
+	blockIdx, ok := state.OutputIndexToBlockIdx[evt.OutputIndex]
+	if !ok {
+		return nil
+	}
+
+	return []AnthropicStreamEvent{{
+		Type:  "content_block_delta",
+		Index: &blockIdx,
+		Delta: &AnthropicDelta{
+			Type:     "thinking_delta",
+			Thinking: evt.Delta,
+		},
+	}}
+}
+
+func resToAnthHandleBlockDone(state *ResponsesEventToAnthropicState) []AnthropicStreamEvent {
+	if !state.ContentBlockOpen {
+		return nil
+	}
+	return closeCurrentBlock(state)
+}
+
+func resToAnthHandleOutputItemDone(evt *ResponsesStreamEvent, state *ResponsesEventToAnthropicState) []AnthropicStreamEvent {
+	if evt.Item == nil {
+		return nil
+	}
+
+	// Handle web_search_call → synthesize server_tool_use + web_search_tool_result blocks.
+	if evt.Item.Type == "web_search_call" && evt.Item.Status == "completed" {
+		return resToAnthHandleWebSearchDone(evt, state)
+	}
+
+	if state.ContentBlockOpen {
+		return closeCurrentBlock(state)
+	}
+	return nil
+}
+
+// resToAnthHandleWebSearchDone converts an OpenAI web_search_call output item
+// into Anthropic server_tool_use + web_search_tool_result content block pairs.
+// This allows Claude Code to count the searches performed.
+func resToAnthHandleWebSearchDone(evt *ResponsesStreamEvent, state *ResponsesEventToAnthropicState) []AnthropicStreamEvent {
+	var events []AnthropicStreamEvent
+	events = append(events, closeCurrentBlock(state)...)
+
+	toolUseID := "srvtoolu_" + evt.Item.ID
+	query := ""
+	if evt.Item.Action != nil {
+		query = evt.Item.Action.Query
+	}
+	inputJSON, _ := json.Marshal(map[string]string{"query": query})
+
+	// Emit server_tool_use block (start + stop).
+	idx1 := state.ContentBlockIndex
+	events = append(events, AnthropicStreamEvent{
+		Type:  "content_block_start",
+		Index: &idx1,
+		ContentBlock: &AnthropicContentBlock{
+			Type:  "server_tool_use",
+			ID:    toolUseID,
+			Name:  "web_search",
+			Input: inputJSON,
+		},
+	})
+	events = append(events, AnthropicStreamEvent{
+		Type:  "content_block_stop",
+		Index: &idx1,
+	})
+	state.ContentBlockIndex++
+
+	// Emit web_search_tool_result block (start + stop).
+	// Content is empty because OpenAI does not expose individual search results;
+	// the model consumes them internally and produces text output.
+	emptyResults, _ := json.Marshal([]struct{}{})
+	idx2 := state.ContentBlockIndex
+	events = append(events, AnthropicStreamEvent{
+		Type:  "content_block_start",
+		Index: &idx2,
+		ContentBlock: &AnthropicContentBlock{
+			Type:      "web_search_tool_result",
+			ToolUseID: toolUseID,
+			Content:   emptyResults,
+		},
+	})
+	events = append(events, AnthropicStreamEvent{
+		Type:  "content_block_stop",
+		Index: &idx2,
+	})
+	state.ContentBlockIndex++
+
+	return events
+}
+
+func resToAnthHandleCompleted(evt *ResponsesStreamEvent, state *ResponsesEventToAnthropicState) []AnthropicStreamEvent {
+	if state.MessageStopSent {
+		return nil
+	}
+
+	var events []AnthropicStreamEvent
+	events = append(events, closeCurrentBlock(state)...)
+
+	stopReason := "end_turn"
+	if evt.Response != nil {
+		if evt.Response.Usage != nil {
+			state.InputTokens = evt.Response.Usage.InputTokens
+			state.OutputTokens = evt.Response.Usage.OutputTokens
+			if evt.Response.Usage.InputTokensDetails != nil {
+				state.CacheReadInputTokens = evt.Response.Usage.InputTokensDetails.CachedTokens
+			}
+		}
+		switch evt.Response.Status {
+		case "incomplete":
+			if evt.Response.IncompleteDetails != nil && evt.Response.IncompleteDetails.Reason == "max_output_tokens" {
+				stopReason = "max_tokens"
+			}
+		case "completed":
+			if state.ContentBlockIndex > 0 && state.CurrentBlockType == "tool_use" {
+				stopReason = "tool_use"
+			}
+		}
+	}
+
+	events = append(events,
+		AnthropicStreamEvent{
+			Type: "message_delta",
+			Delta: &AnthropicDelta{
+				StopReason: stopReason,
+			},
+			Usage: &AnthropicUsage{
+				InputTokens:          state.InputTokens,
+				OutputTokens:         state.OutputTokens,
+				CacheReadInputTokens: state.CacheReadInputTokens,
+			},
+		},
+		AnthropicStreamEvent{Type: "message_stop"},
+	)
+	state.MessageStopSent = true
+	return events
+}
+
+func closeCurrentBlock(state *ResponsesEventToAnthropicState) []AnthropicStreamEvent {
+	if !state.ContentBlockOpen {
+		return nil
+	}
+	idx := state.ContentBlockIndex
+	state.ContentBlockOpen = false
+	state.ContentBlockIndex++
+	return []AnthropicStreamEvent{{
+		Type:  "content_block_stop",
+		Index: &idx,
+	}}
+}

backend/internal/pkg/apicompat/responses_to_chatcompletions.go

@@ -0,0 +1,374 @@
+package apicompat
+
+import (
+	"crypto/rand"
+	"encoding/hex"
+	"encoding/json"
+	"fmt"
+	"time"
+)
+
+// ---------------------------------------------------------------------------
+// Non-streaming: ResponsesResponse → ChatCompletionsResponse
+// ---------------------------------------------------------------------------
+
+// ResponsesToChatCompletions converts a Responses API response into a Chat
+// Completions response. Text output items are concatenated into
+// choices[0].message.content; function_call items become tool_calls.
+func ResponsesToChatCompletions(resp *ResponsesResponse, model string) *ChatCompletionsResponse {
+	id := resp.ID
+	if id == "" {
+		id = generateChatCmplID()
+	}
+
+	out := &ChatCompletionsResponse{
+		ID:      id,
+		Object:  "chat.completion",
+		Created: time.Now().Unix(),
+		Model:   model,
+	}
+
+	var contentText string
+	var reasoningText string
+	var toolCalls []ChatToolCall
+
+	for _, item := range resp.Output {
+		switch item.Type {
+		case "message":
+			for _, part := range item.Content {
+				if part.Type == "output_text" && part.Text != "" {
+					contentText += part.Text
+				}
+			}
+		case "function_call":
+			toolCalls = append(toolCalls, ChatToolCall{
+				ID:   item.CallID,
+				Type: "function",
+				Function: ChatFunctionCall{
+					Name:      item.Name,
+					Arguments: item.Arguments,
+				},
+			})
+		case "reasoning":
+			for _, s := range item.Summary {
+				if s.Type == "summary_text" && s.Text != "" {
+					reasoningText += s.Text
+				}
+			}
+		case "web_search_call":
+			// silently consumed — results already incorporated into text output
+		}
+	}
+
+	msg := ChatMessage{Role: "assistant"}
+	if len(toolCalls) > 0 {
+		msg.ToolCalls = toolCalls
+	}
+	if contentText != "" {
+		raw, _ := json.Marshal(contentText)
+		msg.Content = raw
+	}
+	if reasoningText != "" {
+		msg.ReasoningContent = reasoningText
+	}
+
+	finishReason := responsesStatusToChatFinishReason(resp.Status, resp.IncompleteDetails, toolCalls)
+
+	out.Choices = []ChatChoice{{
+		Index:        0,
+		Message:      msg,
+		FinishReason: finishReason,
+	}}
+
+	if resp.Usage != nil {
+		usage := &ChatUsage{
+			PromptTokens:     resp.Usage.InputTokens,
+			CompletionTokens: resp.Usage.OutputTokens,
+			TotalTokens:      resp.Usage.InputTokens + resp.Usage.OutputTokens,
+		}
+		if resp.Usage.InputTokensDetails != nil && resp.Usage.InputTokensDetails.CachedTokens > 0 {
+			usage.PromptTokensDetails = &ChatTokenDetails{
+				CachedTokens: resp.Usage.InputTokensDetails.CachedTokens,
+			}
+		}
+		out.Usage = usage
+	}
+
+	return out
+}
+
+func responsesStatusToChatFinishReason(status string, details *ResponsesIncompleteDetails, toolCalls []ChatToolCall) string {
+	switch status {
+	case "incomplete":
+		if details != nil && details.Reason == "max_output_tokens" {
+			return "length"
+		}
+		return "stop"
+	case "completed":
+		if len(toolCalls) > 0 {
+			return "tool_calls"
+		}
+		return "stop"
+	default:
+		return "stop"
+	}
+}
+
+// ---------------------------------------------------------------------------
+// Streaming: ResponsesStreamEvent → []ChatCompletionsChunk (stateful converter)
+// ---------------------------------------------------------------------------
+
+// ResponsesEventToChatState tracks state for converting a sequence of Responses
+// SSE events into Chat Completions SSE chunks.
+type ResponsesEventToChatState struct {
+	ID                     string
+	Model                  string
+	Created                int64
+	SentRole               bool
+	SawToolCall            bool
+	SawText                bool
+	Finalized              bool        // true after finish chunk has been emitted
+	NextToolCallIndex      int         // next sequential tool_call index to assign
+	OutputIndexToToolIndex map[int]int // Responses output_index → Chat tool_calls index
+	IncludeUsage           bool
+	Usage                  *ChatUsage
+}
+
+// NewResponsesEventToChatState returns an initialised stream state.
+func NewResponsesEventToChatState() *ResponsesEventToChatState {
+	return &ResponsesEventToChatState{
+		ID:                     generateChatCmplID(),
+		Created:                time.Now().Unix(),
+		OutputIndexToToolIndex: make(map[int]int),
+	}
+}
+
+// ResponsesEventToChatChunks converts a single Responses SSE event into zero
+// or more Chat Completions chunks, updating state as it goes.
+func ResponsesEventToChatChunks(evt *ResponsesStreamEvent, state *ResponsesEventToChatState) []ChatCompletionsChunk {
+	switch evt.Type {
+	case "response.created":
+		return resToChatHandleCreated(evt, state)
+	case "response.output_text.delta":
+		return resToChatHandleTextDelta(evt, state)
+	case "response.output_item.added":
+		return resToChatHandleOutputItemAdded(evt, state)
+	case "response.function_call_arguments.delta":
+		return resToChatHandleFuncArgsDelta(evt, state)
+	case "response.reasoning_summary_text.delta":
+		return resToChatHandleReasoningDelta(evt, state)
+	case "response.reasoning_summary_text.done":
+		return nil
+	case "response.completed", "response.incomplete", "response.failed":
+		return resToChatHandleCompleted(evt, state)
+	default:
+		return nil
+	}
+}
+
+// FinalizeResponsesChatStream emits a final chunk with finish_reason if the
+// stream ended without a proper completion event (e.g. upstream disconnect).
+// It is idempotent: if a completion event already emitted the finish chunk,
+// this returns nil.
+func FinalizeResponsesChatStream(state *ResponsesEventToChatState) []ChatCompletionsChunk {
+	if state.Finalized {
+		return nil
+	}
+	state.Finalized = true
+
+	finishReason := "stop"
+	if state.SawToolCall {
+		finishReason = "tool_calls"
+	}
+
+	chunks := []ChatCompletionsChunk{makeChatFinishChunk(state, finishReason)}
+
+	if state.IncludeUsage && state.Usage != nil {
+		chunks = append(chunks, ChatCompletionsChunk{
+			ID:      state.ID,
+			Object:  "chat.completion.chunk",
+			Created: state.Created,
+			Model:   state.Model,
+			Choices: []ChatChunkChoice{},
+			Usage:   state.Usage,
+		})
+	}
+
+	return chunks
+}
+
+// ChatChunkToSSE formats a ChatCompletionsChunk as an SSE data line.
+func ChatChunkToSSE(chunk ChatCompletionsChunk) (string, error) {
+	data, err := json.Marshal(chunk)
+	if err != nil {
+		return "", err
+	}
+	return fmt.Sprintf("data: %s\n\n", data), nil
+}
+
+// --- internal handlers ---
+
+func resToChatHandleCreated(evt *ResponsesStreamEvent, state *ResponsesEventToChatState) []ChatCompletionsChunk {
+	if evt.Response != nil {
+		if evt.Response.ID != "" {
+			state.ID = evt.Response.ID
+		}
+		if state.Model == "" && evt.Response.Model != "" {
+			state.Model = evt.Response.Model
+		}
+	}
+	// Emit the role chunk.
+	if state.SentRole {
+		return nil
+	}
+	state.SentRole = true
+
+	role := "assistant"
+	return []ChatCompletionsChunk{makeChatDeltaChunk(state, ChatDelta{Role: role})}
+}
+
+func resToChatHandleTextDelta(evt *ResponsesStreamEvent, state *ResponsesEventToChatState) []ChatCompletionsChunk {
+	if evt.Delta == "" {
+		return nil
+	}
+	state.SawText = true
+	content := evt.Delta
+	return []ChatCompletionsChunk{makeChatDeltaChunk(state, ChatDelta{Content: &content})}
+}
+
+func resToChatHandleOutputItemAdded(evt *ResponsesStreamEvent, state *ResponsesEventToChatState) []ChatCompletionsChunk {
+	if evt.Item == nil || evt.Item.Type != "function_call" {
+		return nil
+	}
+
+	state.SawToolCall = true
+	idx := state.NextToolCallIndex
+	state.OutputIndexToToolIndex[evt.OutputIndex] = idx
+	state.NextToolCallIndex++
+
+	return []ChatCompletionsChunk{makeChatDeltaChunk(state, ChatDelta{
+		ToolCalls: []ChatToolCall{{
+			Index: &idx,
+			ID:    evt.Item.CallID,
+			Type:  "function",
+			Function: ChatFunctionCall{
+				Name: evt.Item.Name,
+			},
+		}},
+	})}
+}
+
+func resToChatHandleFuncArgsDelta(evt *ResponsesStreamEvent, state *ResponsesEventToChatState) []ChatCompletionsChunk {
+	if evt.Delta == "" {
+		return nil
+	}
+
+	idx, ok := state.OutputIndexToToolIndex[evt.OutputIndex]
+	if !ok {
+		return nil
+	}
+
+	return []ChatCompletionsChunk{makeChatDeltaChunk(state, ChatDelta{
+		ToolCalls: []ChatToolCall{{
+			Index: &idx,
+			Function: ChatFunctionCall{
+				Arguments: evt.Delta,
+			},
+		}},
+	})}
+}
+
+func resToChatHandleReasoningDelta(evt *ResponsesStreamEvent, state *ResponsesEventToChatState) []ChatCompletionsChunk {
+	if evt.Delta == "" {
+		return nil
+	}
+	reasoning := evt.Delta
+	return []ChatCompletionsChunk{makeChatDeltaChunk(state, ChatDelta{ReasoningContent: &reasoning})}
+}
+
+func resToChatHandleCompleted(evt *ResponsesStreamEvent, state *ResponsesEventToChatState) []ChatCompletionsChunk {
+	state.Finalized = true
+	finishReason := "stop"
+
+	if evt.Response != nil {
+		if evt.Response.Usage != nil {
+			u := evt.Response.Usage
+			usage := &ChatUsage{
+				PromptTokens:     u.InputTokens,
+				CompletionTokens: u.OutputTokens,
+				TotalTokens:      u.InputTokens + u.OutputTokens,
+			}
+			if u.InputTokensDetails != nil && u.InputTokensDetails.CachedTokens > 0 {
+				usage.PromptTokensDetails = &ChatTokenDetails{
+					CachedTokens: u.InputTokensDetails.CachedTokens,
+				}
+			}
+			state.Usage = usage
+		}
+
+		switch evt.Response.Status {
+		case "incomplete":
+			if evt.Response.IncompleteDetails != nil && evt.Response.IncompleteDetails.Reason == "max_output_tokens" {
+				finishReason = "length"
+			}
+		case "completed":
+			if state.SawToolCall {
+				finishReason = "tool_calls"
+			}
+		}
+	} else if state.SawToolCall {
+		finishReason = "tool_calls"
+	}
+
+	var chunks []ChatCompletionsChunk
+	chunks = append(chunks, makeChatFinishChunk(state, finishReason))
+
+	if state.IncludeUsage && state.Usage != nil {
+		chunks = append(chunks, ChatCompletionsChunk{
+			ID:      state.ID,
+			Object:  "chat.completion.chunk",
+			Created: state.Created,
+			Model:   state.Model,
+			Choices: []ChatChunkChoice{},
+			Usage:   state.Usage,
+		})
+	}
+
+	return chunks
+}
+
+func makeChatDeltaChunk(state *ResponsesEventToChatState, delta ChatDelta) ChatCompletionsChunk {
+	return ChatCompletionsChunk{
+		ID:      state.ID,
+		Object:  "chat.completion.chunk",
+		Created: state.Created,
+		Model:   state.Model,
+		Choices: []ChatChunkChoice{{
+			Index:        0,
+			Delta:        delta,
+			FinishReason: nil,
+		}},
+	}
+}
+
+func makeChatFinishChunk(state *ResponsesEventToChatState, finishReason string) ChatCompletionsChunk {
+	empty := ""
+	return ChatCompletionsChunk{
+		ID:      state.ID,
+		Object:  "chat.completion.chunk",
+		Created: state.Created,
+		Model:   state.Model,
+		Choices: []ChatChunkChoice{{
+			Index:        0,
+			Delta:        ChatDelta{Content: &empty},
+			FinishReason: &finishReason,
+		}},
+	}
+}
+
+// generateChatCmplID returns a "chatcmpl-" prefixed random hex ID.
+func generateChatCmplID() string {
+	b := make([]byte, 12)
+	_, _ = rand.Read(b)
+	return "chatcmpl-" + hex.EncodeToString(b)
+}

backend/internal/pkg/apicompat/types.go

@@ -0,0 +1,482 @@
+// Package apicompat provides type definitions and conversion utilities for
+// translating between Anthropic Messages and OpenAI Responses API formats.
+// It enables multi-protocol support so that clients using different API
+// formats can be served through a unified gateway.
+package apicompat
+
+import "encoding/json"
+
+// ---------------------------------------------------------------------------
+// Anthropic Messages API types
+// ---------------------------------------------------------------------------
+
+// AnthropicRequest is the request body for POST /v1/messages.
+type AnthropicRequest struct {
+	Model        string                 `json:"model"`
+	MaxTokens    int                    `json:"max_tokens"`
+	System       json.RawMessage        `json:"system,omitempty"` // string or []AnthropicContentBlock
+	Messages     []AnthropicMessage     `json:"messages"`
+	Tools        []AnthropicTool        `json:"tools,omitempty"`
+	Stream       bool                   `json:"stream,omitempty"`
+	Temperature  *float64               `json:"temperature,omitempty"`
+	TopP         *float64               `json:"top_p,omitempty"`
+	StopSeqs     []string               `json:"stop_sequences,omitempty"`
+	Thinking     *AnthropicThinking     `json:"thinking,omitempty"`
+	ToolChoice   json.RawMessage        `json:"tool_choice,omitempty"`
+	OutputConfig *AnthropicOutputConfig `json:"output_config,omitempty"`
+}
+
+// AnthropicOutputConfig controls output generation parameters.
+type AnthropicOutputConfig struct {
+	Effort string `json:"effort,omitempty"` // "low" | "medium" | "high"
+}
+
+// AnthropicThinking configures extended thinking in the Anthropic API.
+type AnthropicThinking struct {
+	Type         string `json:"type"`                    // "enabled" | "adaptive" | "disabled"
+	BudgetTokens int    `json:"budget_tokens,omitempty"` // max thinking tokens
+}
+
+// AnthropicMessage is a single message in the Anthropic conversation.
+type AnthropicMessage struct {
+	Role    string          `json:"role"` // "user" | "assistant"
+	Content json.RawMessage `json:"content"`
+}
+
+// AnthropicContentBlock is one block inside a message's content array.
+type AnthropicContentBlock struct {
+	Type string `json:"type"`
+
+	// type=text
+	Text string `json:"text,omitempty"`
+
+	// type=thinking
+	Thinking string `json:"thinking,omitempty"`
+
+	// type=image
+	Source *AnthropicImageSource `json:"source,omitempty"`
+
+	// type=tool_use
+	ID    string          `json:"id,omitempty"`
+	Name  string          `json:"name,omitempty"`
+	Input json.RawMessage `json:"input,omitempty"`
+
+	// type=tool_result
+	ToolUseID string          `json:"tool_use_id,omitempty"`
+	Content   json.RawMessage `json:"content,omitempty"` // string or []AnthropicContentBlock
+	IsError   bool            `json:"is_error,omitempty"`
+}
+
+// AnthropicImageSource describes the source data for an image content block.
+type AnthropicImageSource struct {
+	Type      string `json:"type"` // "base64"
+	MediaType string `json:"media_type"`
+	Data      string `json:"data"`
+}
+
+// AnthropicTool describes a tool available to the model.
+type AnthropicTool struct {
+	Type        string          `json:"type,omitempty"` // e.g. "web_search_20250305" for server tools
+	Name        string          `json:"name"`
+	Description string          `json:"description,omitempty"`
+	InputSchema json.RawMessage `json:"input_schema"` // JSON Schema object
+}
+
+// AnthropicResponse is the non-streaming response from POST /v1/messages.
+type AnthropicResponse struct {
+	ID           string                  `json:"id"`
+	Type         string                  `json:"type"` // "message"
+	Role         string                  `json:"role"` // "assistant"
+	Content      []AnthropicContentBlock `json:"content"`
+	Model        string                  `json:"model"`
+	StopReason   string                  `json:"stop_reason"`
+	StopSequence *string                 `json:"stop_sequence,omitempty"`
+	Usage        AnthropicUsage          `json:"usage"`
+}
+
+// AnthropicUsage holds token counts in Anthropic format.
+type AnthropicUsage struct {
+	InputTokens              int `json:"input_tokens"`
+	OutputTokens             int `json:"output_tokens"`
+	CacheCreationInputTokens int `json:"cache_creation_input_tokens"`
+	CacheReadInputTokens     int `json:"cache_read_input_tokens"`
+}
+
+// ---------------------------------------------------------------------------
+// Anthropic SSE event types
+// ---------------------------------------------------------------------------
+
+// AnthropicStreamEvent is a single SSE event in the Anthropic streaming protocol.
+type AnthropicStreamEvent struct {
+	Type string `json:"type"`
+
+	// message_start
+	Message *AnthropicResponse `json:"message,omitempty"`
+
+	// content_block_start
+	Index        *int                   `json:"index,omitempty"`
+	ContentBlock *AnthropicContentBlock `json:"content_block,omitempty"`
+
+	// content_block_delta
+	Delta *AnthropicDelta `json:"delta,omitempty"`
+
+	// message_delta
+	Usage *AnthropicUsage `json:"usage,omitempty"`
+}
+
+// AnthropicDelta carries incremental content in streaming events.
+type AnthropicDelta struct {
+	Type string `json:"type,omitempty"` // "text_delta" | "input_json_delta" | "thinking_delta" | "signature_delta"
+
+	// text_delta
+	Text string `json:"text,omitempty"`
+
+	// input_json_delta
+	PartialJSON string `json:"partial_json,omitempty"`
+
+	// thinking_delta
+	Thinking string `json:"thinking,omitempty"`
+
+	// signature_delta
+	Signature string `json:"signature,omitempty"`
+
+	// message_delta fields
+	StopReason   string  `json:"stop_reason,omitempty"`
+	StopSequence *string `json:"stop_sequence,omitempty"`
+}
+
+// ---------------------------------------------------------------------------
+// OpenAI Responses API types
+// ---------------------------------------------------------------------------
+
+// ResponsesRequest is the request body for POST /v1/responses.
+type ResponsesRequest struct {
+	Model           string              `json:"model"`
+	Input           json.RawMessage     `json:"input"` // string or []ResponsesInputItem
+	MaxOutputTokens *int                `json:"max_output_tokens,omitempty"`
+	Temperature     *float64            `json:"temperature,omitempty"`
+	TopP            *float64            `json:"top_p,omitempty"`
+	Stream          bool                `json:"stream,omitempty"`
+	Tools           []ResponsesTool     `json:"tools,omitempty"`
+	Include         []string            `json:"include,omitempty"`
+	Store           *bool               `json:"store,omitempty"`
+	Reasoning       *ResponsesReasoning `json:"reasoning,omitempty"`
+	ToolChoice      json.RawMessage     `json:"tool_choice,omitempty"`
+	ServiceTier     string              `json:"service_tier,omitempty"`
+}
+
+// ResponsesReasoning configures reasoning effort in the Responses API.
+type ResponsesReasoning struct {
+	Effort  string `json:"effort"`            // "low" | "medium" | "high"
+	Summary string `json:"summary,omitempty"` // "auto" | "concise" | "detailed"
+}
+
+// ResponsesInputItem is one item in the Responses API input array.
+// The Type field determines which other fields are populated.
+type ResponsesInputItem struct {
+	// Common
+	Type string `json:"type,omitempty"` // "" for role-based messages
+
+	// Role-based messages (system/user/assistant)
+	Role    string          `json:"role,omitempty"`
+	Content json.RawMessage `json:"content,omitempty"` // string or []ResponsesContentPart
+
+	// type=function_call
+	CallID    string `json:"call_id,omitempty"`
+	Name      string `json:"name,omitempty"`
+	Arguments string `json:"arguments,omitempty"`
+	ID        string `json:"id,omitempty"`
+
+	// type=function_call_output
+	Output string `json:"output,omitempty"`
+}
+
+// ResponsesContentPart is a typed content part in a Responses message.
+type ResponsesContentPart struct {
+	Type     string `json:"type"` // "input_text" | "output_text" | "input_image"
+	Text     string `json:"text,omitempty"`
+	ImageURL string `json:"image_url,omitempty"` // data URI for input_image
+}
+
+// ResponsesTool describes a tool in the Responses API.
+type ResponsesTool struct {
+	Type        string          `json:"type"` // "function" | "web_search" | "local_shell" etc.
+	Name        string          `json:"name,omitempty"`
+	Description string          `json:"description,omitempty"`
+	Parameters  json.RawMessage `json:"parameters,omitempty"`
+	Strict      *bool           `json:"strict,omitempty"`
+}
+
+// ResponsesResponse is the non-streaming response from POST /v1/responses.
+type ResponsesResponse struct {
+	ID     string            `json:"id"`
+	Object string            `json:"object"` // "response"
+	Model  string            `json:"model"`
+	Status string            `json:"status"` // "completed" | "incomplete" | "failed"
+	Output []ResponsesOutput `json:"output"`
+	Usage  *ResponsesUsage   `json:"usage,omitempty"`
+
+	// incomplete_details is present when status="incomplete"
+	IncompleteDetails *ResponsesIncompleteDetails `json:"incomplete_details,omitempty"`
+
+	// Error is present when status="failed"
+	Error *ResponsesError `json:"error,omitempty"`
+}
+
+// ResponsesError describes an error in a failed response.
+type ResponsesError struct {
+	Code    string `json:"code"`
+	Message string `json:"message"`
+}
+
+// ResponsesIncompleteDetails explains why a response is incomplete.
+type ResponsesIncompleteDetails struct {
+	Reason string `json:"reason"` // "max_output_tokens" | "content_filter"
+}
+
+// ResponsesOutput is one output item in a Responses API response.
+type ResponsesOutput struct {
+	Type string `json:"type"` // "message" | "reasoning" | "function_call" | "web_search_call"
+
+	// type=message
+	ID      string                 `json:"id,omitempty"`
+	Role    string                 `json:"role,omitempty"`
+	Content []ResponsesContentPart `json:"content,omitempty"`
+	Status  string                 `json:"status,omitempty"`
+
+	// type=reasoning
+	EncryptedContent string             `json:"encrypted_content,omitempty"`
+	Summary          []ResponsesSummary `json:"summary,omitempty"`
+
+	// type=function_call
+	CallID    string `json:"call_id,omitempty"`
+	Name      string `json:"name,omitempty"`
+	Arguments string `json:"arguments,omitempty"`
+
+	// type=web_search_call
+	Action *WebSearchAction `json:"action,omitempty"`
+}
+
+// WebSearchAction describes the search action in a web_search_call output item.
+type WebSearchAction struct {
+	Type  string `json:"type,omitempty"`  // "search"
+	Query string `json:"query,omitempty"` // primary search query
+}
+
+// ResponsesSummary is a summary text block inside a reasoning output.
+type ResponsesSummary struct {
+	Type string `json:"type"` // "summary_text"
+	Text string `json:"text"`
+}
+
+// ResponsesUsage holds token counts in Responses API format.
+type ResponsesUsage struct {
+	InputTokens  int `json:"input_tokens"`
+	OutputTokens int `json:"output_tokens"`
+	TotalTokens  int `json:"total_tokens"`
+
+	// Optional detailed breakdown
+	InputTokensDetails  *ResponsesInputTokensDetails  `json:"input_tokens_details,omitempty"`
+	OutputTokensDetails *ResponsesOutputTokensDetails `json:"output_tokens_details,omitempty"`
+}
+
+// ResponsesInputTokensDetails breaks down input token usage.
+type ResponsesInputTokensDetails struct {
+	CachedTokens int `json:"cached_tokens,omitempty"`
+}
+
+// ResponsesOutputTokensDetails breaks down output token usage.
+type ResponsesOutputTokensDetails struct {
+	ReasoningTokens int `json:"reasoning_tokens,omitempty"`
+}
+
+// ---------------------------------------------------------------------------
+// Responses SSE event types
+// ---------------------------------------------------------------------------
+
+// ResponsesStreamEvent is a single SSE event in the Responses streaming protocol.
+// The Type field corresponds to the "type" in the JSON payload.
+type ResponsesStreamEvent struct {
+	Type string `json:"type"`
+
+	// response.created / response.completed / response.failed / response.incomplete
+	Response *ResponsesResponse `json:"response,omitempty"`
+
+	// response.output_item.added / response.output_item.done
+	Item *ResponsesOutput `json:"item,omitempty"`
+
+	// response.output_text.delta / response.output_text.done
+	OutputIndex  int    `json:"output_index,omitempty"`
+	ContentIndex int    `json:"content_index,omitempty"`
+	Delta        string `json:"delta,omitempty"`
+	Text         string `json:"text,omitempty"`
+	ItemID       string `json:"item_id,omitempty"`
+
+	// response.function_call_arguments.delta / done
+	CallID    string `json:"call_id,omitempty"`
+	Name      string `json:"name,omitempty"`
+	Arguments string `json:"arguments,omitempty"`
+
+	// response.reasoning_summary_text.delta / done
+	// Reuses Text/Delta fields above, SummaryIndex identifies which summary part
+	SummaryIndex int `json:"summary_index,omitempty"`
+
+	// error event fields
+	Code  string `json:"code,omitempty"`
+	Param string `json:"param,omitempty"`
+
+	// Sequence number for ordering events
+	SequenceNumber int `json:"sequence_number,omitempty"`
+}
+
+// ---------------------------------------------------------------------------
+// OpenAI Chat Completions API types
+// ---------------------------------------------------------------------------
+
+// ChatCompletionsRequest is the request body for POST /v1/chat/completions.
+type ChatCompletionsRequest struct {
+	Model               string             `json:"model"`
+	Messages            []ChatMessage      `json:"messages"`
+	MaxTokens           *int               `json:"max_tokens,omitempty"`
+	MaxCompletionTokens *int               `json:"max_completion_tokens,omitempty"`
+	Temperature         *float64           `json:"temperature,omitempty"`
+	TopP                *float64           `json:"top_p,omitempty"`
+	Stream              bool               `json:"stream,omitempty"`
+	StreamOptions       *ChatStreamOptions `json:"stream_options,omitempty"`
+	Tools               []ChatTool         `json:"tools,omitempty"`
+	ToolChoice          json.RawMessage    `json:"tool_choice,omitempty"`
+	ReasoningEffort     string             `json:"reasoning_effort,omitempty"` // "low" | "medium" | "high"
+	ServiceTier         string             `json:"service_tier,omitempty"`
+	Stop                json.RawMessage    `json:"stop,omitempty"` // string or []string
+
+	// Legacy function calling (deprecated but still supported)
+	Functions    []ChatFunction  `json:"functions,omitempty"`
+	FunctionCall json.RawMessage `json:"function_call,omitempty"`
+}
+
+// ChatStreamOptions configures streaming behavior.
+type ChatStreamOptions struct {
+	IncludeUsage bool `json:"include_usage,omitempty"`
+}
+
+// ChatMessage is a single message in the Chat Completions conversation.
+type ChatMessage struct {
+	Role             string          `json:"role"` // "system" | "user" | "assistant" | "tool" | "function"
+	Content          json.RawMessage `json:"content,omitempty"`
+	ReasoningContent string          `json:"reasoning_content,omitempty"`
+	Name             string          `json:"name,omitempty"`
+	ToolCalls        []ChatToolCall  `json:"tool_calls,omitempty"`
+	ToolCallID       string          `json:"tool_call_id,omitempty"`
+
+	// Legacy function calling
+	FunctionCall *ChatFunctionCall `json:"function_call,omitempty"`
+}
+
+// ChatContentPart is a typed content part in a multi-modal message.
+type ChatContentPart struct {
+	Type     string        `json:"type"` // "text" | "image_url"
+	Text     string        `json:"text,omitempty"`
+	ImageURL *ChatImageURL `json:"image_url,omitempty"`
+}
+
+// ChatImageURL contains the URL for an image content part.
+type ChatImageURL struct {
+	URL    string `json:"url"`
+	Detail string `json:"detail,omitempty"` // "auto" | "low" | "high"
+}
+
+// ChatTool describes a tool available to the model.
+type ChatTool struct {
+	Type     string        `json:"type"` // "function"
+	Function *ChatFunction `json:"function,omitempty"`
+}
+
+// ChatFunction describes a function tool definition.
+type ChatFunction struct {
+	Name        string          `json:"name"`
+	Description string          `json:"description,omitempty"`
+	Parameters  json.RawMessage `json:"parameters,omitempty"`
+	Strict      *bool           `json:"strict,omitempty"`
+}
+
+// ChatToolCall represents a tool call made by the assistant.
+// Index is only populated in streaming chunks (omitted in non-streaming responses).
+type ChatToolCall struct {
+	Index    *int             `json:"index,omitempty"`
+	ID       string           `json:"id,omitempty"`
+	Type     string           `json:"type,omitempty"` // "function"
+	Function ChatFunctionCall `json:"function"`
+}
+
+// ChatFunctionCall contains the function name and arguments.
+type ChatFunctionCall struct {
+	Name      string `json:"name"`
+	Arguments string `json:"arguments"`
+}
+
+// ChatCompletionsResponse is the non-streaming response from POST /v1/chat/completions.
+type ChatCompletionsResponse struct {
+	ID                string       `json:"id"`
+	Object            string       `json:"object"` // "chat.completion"
+	Created           int64        `json:"created"`
+	Model             string       `json:"model"`
+	Choices           []ChatChoice `json:"choices"`
+	Usage             *ChatUsage   `json:"usage,omitempty"`
+	SystemFingerprint string       `json:"system_fingerprint,omitempty"`
+	ServiceTier       string       `json:"service_tier,omitempty"`
+}
+
+// ChatChoice is a single completion choice.
+type ChatChoice struct {
+	Index        int         `json:"index"`
+	Message      ChatMessage `json:"message"`
+	FinishReason string      `json:"finish_reason"` // "stop" | "length" | "tool_calls" | "content_filter"
+}
+
+// ChatUsage holds token counts in Chat Completions format.
+type ChatUsage struct {
+	PromptTokens        int               `json:"prompt_tokens"`
+	CompletionTokens    int               `json:"completion_tokens"`
+	TotalTokens         int               `json:"total_tokens"`
+	PromptTokensDetails *ChatTokenDetails `json:"prompt_tokens_details,omitempty"`
+}
+
+// ChatTokenDetails provides a breakdown of token usage.
+type ChatTokenDetails struct {
+	CachedTokens int `json:"cached_tokens,omitempty"`
+}
+
+// ChatCompletionsChunk is a single streaming chunk from POST /v1/chat/completions.
+type ChatCompletionsChunk struct {
+	ID                string            `json:"id"`
+	Object            string            `json:"object"` // "chat.completion.chunk"
+	Created           int64             `json:"created"`
+	Model             string            `json:"model"`
+	Choices           []ChatChunkChoice `json:"choices"`
+	Usage             *ChatUsage        `json:"usage,omitempty"`
+	SystemFingerprint string            `json:"system_fingerprint,omitempty"`
+	ServiceTier       string            `json:"service_tier,omitempty"`
+}
+
+// ChatChunkChoice is a single choice in a streaming chunk.
+type ChatChunkChoice struct {
+	Index        int       `json:"index"`
+	Delta        ChatDelta `json:"delta"`
+	FinishReason *string   `json:"finish_reason"` // pointer: null when not final
+}
+
+// ChatDelta carries incremental content in a streaming chunk.
+type ChatDelta struct {
+	Role             string         `json:"role,omitempty"`
+	Content          *string        `json:"content,omitempty"` // pointer: omit when not present, null vs "" matters
+	ReasoningContent *string        `json:"reasoning_content,omitempty"`
+	ToolCalls        []ChatToolCall `json:"tool_calls,omitempty"`
+}
+
+// ---------------------------------------------------------------------------
+// Shared constants
+// ---------------------------------------------------------------------------
+
+// minMaxOutputTokens is the floor for max_output_tokens in a Responses request.
+// Very small values may cause upstream API errors, so we enforce a minimum.
+const minMaxOutputTokens = 128

backend/internal/pkg/claude/constants.go

@@ -16,7 +16,7 @@ const (
 
 // DroppedBetas 是转发时需要从 anthropic-beta header 中移除的 beta token 列表。
 // 这些 token 是客户端特有的，不应透传给上游 API。
-var DroppedBetas = []string{BetaContext1M, BetaFastMode}
+var DroppedBetas = []string{}
 
 // DefaultBetaHeader Claude Code 客户端默认的 anthropic-beta header
 const DefaultBetaHeader = BetaClaudeCode + "," + BetaOAuth + "," + BetaInterleavedThinking + "," + BetaFineGrainedToolStreaming

backend/internal/pkg/gemini/models.go

@@ -18,10 +18,12 @@ func DefaultModels() []Model {
 	return []Model{
 		{Name: "models/gemini-2.0-flash", SupportedGenerationMethods: methods},
 		{Name: "models/gemini-2.5-flash", SupportedGenerationMethods: methods},
+		{Name: "models/gemini-2.5-flash-image", SupportedGenerationMethods: methods},
 		{Name: "models/gemini-2.5-pro", SupportedGenerationMethods: methods},
 		{Name: "models/gemini-3-flash-preview", SupportedGenerationMethods: methods},
 		{Name: "models/gemini-3-pro-preview", SupportedGenerationMethods: methods},
 		{Name: "models/gemini-3.1-pro-preview", SupportedGenerationMethods: methods},
+		{Name: "models/gemini-3.1-flash-image", SupportedGenerationMethods: methods},
 	}
 }
 

backend/internal/pkg/gemini/models_test.go

@@ -0,0 +1,28 @@
+package gemini
+
+import "testing"
+
+func TestDefaultModels_ContainsImageModels(t *testing.T) {
+	t.Parallel()
+
+	models := DefaultModels()
+	byName := make(map[string]Model, len(models))
+	for _, model := range models {
+		byName[model.Name] = model
+	}
+
+	required := []string{
+		"models/gemini-2.5-flash-image",
+		"models/gemini-3.1-flash-image",
+	}
+
+	for _, name := range required {
+		model, ok := byName[name]
+		if !ok {
+			t.Fatalf("expected fallback model %q to exist", name)
+		}
+		if len(model.SupportedGenerationMethods) == 0 {
+			t.Fatalf("expected fallback model %q to advertise generation methods", name)
+		}
+	}
+}