zfc/sub2api
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity
3,752 Commits 3 Branches 0 Tags
Commit Graph

442 Commits

Author SHA1 Message Date
win
a420179abb chore: merge upstream Wei-Shaw/sub2api v0.1.133 2026-05-29 17:48:27 +08:00
Wesley Liddick
1d46be02ae
Merge pull request #2830 from stabey/fix/anthropic-to-responses-cache-tokens 
fix(apicompat): Anthropic 转 Responses 时按 OpenAI 语义汇总 input_tokens
 2026-05-29 11:00:19 +08:00
Wesley Liddick
fc9d79d96b
Merge pull request #2835 from JIA-ss/fix/responses-passthrough-token-details 
fix(apicompat): Responses→Chat 转换补齐 completion_tokens_details 透传
 2026-05-29 10:31:52 +08:00
Wesley Liddick
433f8dcd13
Merge pull request #2834 from DaydreamCoding/pr/openai-codex-cli-allow-claude-code 
feat(openai): codex_cli_only 新增放行 Claude Code Codex 插件的机制
 2026-05-29 10:30:33 +08:00
shaw
514ac5c6a1 feat: 适配 claude-opus-4-8 2026-05-29 09:56:48 +08:00
win
f519a02ec9 chore: merge upstream Wei-Shaw/sub2api v0.1.132 
Conflicts resolved (preserving fork customizations):
- config.go: keep NodeTLSProxy + add upstream OpenAIHTTP2
- gateway_service.go: NewGatewayService now takes both rpmTokenBucketSvc
  (local) and userPlatformQuotaRepo (upstream)
- wire_gen.go: wire both new args into the call site
- http_upstream.go: drop redundant settings re-assignment; keep proxy
  URL log redaction
- http_upstream_test.go: adopt upstream's explicit-0-disables semantics;
  keep 600s default constant in nil-cfg fallback test
- user_handler_test.go / gateway_record_usage_test.go: pick up new
  userPlatformQuotaRepo nil parameter

Also updated test stubs (windsurf_google_login_test.go,
windsurf_tier_access_service_test.go, gateway_models_test.go) for new
SetModelRateLimit variadic signature and the extra NewGatewayService arg.

Upstream highlights: OpenAI embeddings gateway, user x platform USD
quota, content-moderation risk thresholds, OAuth 401 credentials
no-overwrite fix, HTTP/2 OpenAI upstream config, pool retry status code
configurability, long-context cache pricing multipliers.
 2026-05-29 07:21:32 +08:00
JIA-ss
20f5340784 fix(apicompat): Responses→Chat 转换补齐 completion_tokens_details 透传 
OpenAI Responses API 在 gpt-5.x 等 reasoning 模型上会返回
output_tokens_details.reasoning_tokens, 但 ResponsesToChatCompletions
只映射了 input_tokens_details.cached_tokens, 导致客户端拿到的
chat.completion.usage 中 completion_tokens 出现无法解释的波动
(短 prompt 也可能 30+ token), 且缺失 reasoning_tokens 细分字段,
难以与 OpenAI 原生 Chat Completions 响应对账。

按 OpenAI 官方 CompletionUsage schema (openai/openai-go SDK
completion.go) 补齐所有 token-details 字段, 全部 omitempty:

  prompt_tokens_details:
    - cached_tokens   (原已支持)
    - audio_tokens    (新增)
  completion_tokens_details:
    - reasoning_tokens             (新增)
    - audio_tokens                 (新增)
    - accepted_prediction_tokens   (新增)
    - rejected_prediction_tokens   (新增)

实现细节:
- 抽出 promptDetailsFromResponses / completionDetailsFromResponses
  两个 helper, 全零字段返回 nil
- 非流路径 ResponsesToChatCompletions 复用已存在的
  chatUsageFromResponsesUsage helper, 消除两条路径间的重复
- 非 reasoning / 非 audio 上游 (Anthropic, Gemini, gpt-4o) 不填这些
  字段, helper 返回 nil → CompletionTokensDetails 不输出, 对现有响应
  字节级兼容

新增单测:
- TestResponsesToChatCompletions_ReasoningTokens
- TestResponsesToChatCompletions_AllTokenDetailsPassThrough
- TestResponsesToChatCompletions_NoReasoningTokensWhenZero
- TestResponsesEventToChatChunks_CompletedWithReasoningTokens
 2026-05-28 00:38:25 +08:00
DaydreamCoding
56908d3c4c feat(openai): codex_cli_only 新增放行 Claude Code Codex 插件的机制 
适用场景：在 Claude Code 中使用 https://github.com/openai/codex-plugin-cc
插件时，插件经官方 codex app-server 以 clientInfo.name="Claude Code" 完成
initialize 握手，请求头被设为 originator=Claude Code、User-Agent 含
"Claude Code/"，不在官方客户端白名单内，原本会被 codex_cli_only 拦截 403。

在官方客户端白名单未命中时评估两层独立放行（OR 语义）：

- 按账号：account.Extra.codex_cli_only_allowed_clients 引用命名预设
  （目前仅 claude_code），detector reason=allowed_client_matched
- 全局开关：/admin/settings 网关服务 OpenAI 区块新增
  openai_allow_claude_code_codex_plugin（默认 false），开启后对所有
  codex_cli_only 账号统一放行，detector reason=global_allowed_client_matched

签名仍要求 originator=Claude Code 精确等值 + UA 含 "Claude Code/"。
上游转发保持透传不变。

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-27 23:55:34 +08:00
stabey
89dffdd2e1 fix(apicompat): emit OpenAI-semantic input_tokens when converting Anthropic to Responses 
Anthropic Messages reports input_tokens excluding cache_read/cache_creation, but
OpenAI Responses input_tokens is the total including cached tokens. The reverse
converter passed Anthropic's input_tokens straight through, so client-facing
prompt_tokens/input_tokens were short by the cached count and cache_creation
was dropped entirely.

Fix the non-stream path and the streaming state machine to add cache_read +
cache_creation back into input_tokens, and track CacheCreationInputTokens on
the streaming state. Six downstream paths benefit (Anthropic->Responses,
Anthropic->ChatCompletions, Gemini->ChatCompletions, each sync + stream).

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
 2026-05-27 22:36:52 +08:00
shaw
f7ac5e5931 fix(openai): preserve chat responses usage billing 2026-05-26 21:33:28 +08:00
DaydreamCoding
6b39b344d8 feat(quota): 用户 × 平台 USD 配额 
为用户在 anthropic/openai/gemini/antigravity 四个平台上提供日/周/月
三个窗口的 USD 配额管控。配额语义：未设置=不限制，0=禁用，>0=美元上限。

两层模型：
- 配置层：系统默认配额，以及 email/linuxdo/oidc/wechat/github/google/
  dingtalk 七个鉴权来源的默认配额，存于 settings，以嵌套 JSON 整体读写
  （系统 1 个 key + 每个来源 1 个 key），整体替换语义。
- 运行时层：user_platform_quota 表按用户记录实际配额，与配置层解耦。

后端：新增 ent schema 与 140_user_platform_quotas.sql 迁移、repository
与 service 端口、计费链路集成、管理端与用户端读写接口。
前端：管理端设置页配额编辑、用户配额管理 Modal、用户 Dashboard 展示、
中英文案。

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-26 10:49:20 +08:00
win
e938be5f3f chore: merge upstream Wei-Shaw/sub2api latest (v0.1.130+) 
Upstream features: bedrock CC compat, email whitelist wildcard,
content moderation per-model toggle, redeem code batch update,
OIDC verified-email fast path, subscription expiry email,
cache hit rate fix, audit dedup, js-cookie security fix,
x/net vulnerability fix, OpenAI account cooldown optimization,
reverse proxy client IP fix, API key ACL trusted forwarded IP.

Local additions preserved: rpmTokenBucketService, quotaFactor
scoring, P2C scheduler selection.
 2026-05-24 15:54:54 +08:00
wucm667
c4d7edba08 fix(apicompat): map developer role to system 2026-05-21 16:37:05 +08:00
win
92433656f5 chore: merge upstream Wei-Shaw/sub2api v0.1.128 — keep fork customizations 
Upstream 新功能 (34 commits, ~main..origin/main):
- feat(email): 通知邮件模板服务、模板编辑器、订阅/余额提醒邮件
- feat(notification): NotificationEmailService 注入到 Balance/Payment/Setting
- feat(payment): 支付成功通知邮件
- feat(usage): 用户 API Key 用量页支持按日明细
- feat(openai-gateway): Codex OAuth 浏览器 UA 自动改写规避 Cloudflare 质询
- feat(admin): 邮件模板管理接口
- fix(auth): 停用/删除分组后阻断 API Key
- fix(group): 修正分组账号可用计数口径
- fix(openai): /v1/responses respect force chat completions, images n 参数透传
- test(repository): AES Encryptor 单元测试
- chore: VERSION 0.1.128

冲突解决 (backend/cmd/server/wire_gen.go):
- 引入 upstream 新 wire providers: notificationEmailService,
  ProvidePaymentService(10 args), ProvideAdminSettingHandler(8 args)
- 保留 fork 独有依赖: rpmTokenBucketService (RPM 平滑),
  NewOpsHandler 3 参数版本 (requestEventBus, opsLogBroadcaster)
- ProvideBalanceNotifyService 接受 4 参数 (含 notificationEmailService)

修复 session-id helper 设计 (claude_code_session_id.go):
- 发现: TestGatewayService_AnthropicOAuth_InjectsClaudeCodeSessionHeaderFromMetadata
  在 OAuth + mimicClaudeCode=false 场景失败
- 重新审视设计原则: OAuth 凭证本身就是 Claude Code 客户端,可信任 metadata
  派生 session_id;不应受 mimicClaudeCode 标志阻止
- 修复: metadata 派生只看 tokenType=="oauth";UUID 兜底仍需 oauth && mimic
- 更新测试: OAuthNonMimicDerivesFromMetadata 取代原 IgnoresMetadata

所有 fork 独有功能保留:
- Claude Code 2.1.145 mimicry bundle (上个 commit 引入)
- RPM token bucket smoothing (commit 95814974)
- Windsurf/Antigravity/Omniroute 定制
- claudemask/ 校验包 (upstream 已删除,我们仍在 gateway_service 中使用)

不在范围:
- 不修复 baseline 既存的 2 个测试失败 (TestProxyImportData...,
  TestWindsurfTierAccessService_Snapshot_HappyPath) - 与 merge 无关
 2026-05-20 17:50:44 +08:00
win
0fefedf9cd feat(claude-mimic): upgrade Claude Code mimicry to 2.1.145 via bundle abstraction 
反编译本地 Claude Code 2.1.145 二进制 (Bun 1.3.14 打包,
@anthropic-ai/sdk@0.94.0 嵌入) 提取真实指纹,系统性升级 mimicry。

核心改动:
- 新增 ClaudeCodeBundle struct 作为单一事实源,DefaultBundle 描述当前
  伪装目标的完整快照 (CLIVersion/SDKVersion/RuntimeVersion/OS/Arch)
- DefaultCLIVersion/DefaultStainlessPackageVersion/CLICurrentVersion/
  DefaultHeaders 全部派生自 DefaultBundle,消除三处 (2.1.92, 2.1.104,
  0.70.0, 0.81.0) 版本分裂
- CLI 版本 2.1.92/2.1.104 -> 2.1.145
- SDK 版本 0.70.0/0.81.0 -> 0.94.0
- 新增 12 个 2.1.145 反编译确认的 anthropic-beta token:
  advanced-tool-use, tool-search-tool, mcp-servers, mcp-client,
  mid-conversation-system, afk-mode, cache-diagnosis, context-hint,
  environments, managed-agents, skills, compact
- FullClaudeCodeMimicryBetas() 从 7 个 token 升级到 21 个 ordered list
- 修正 BetaTokenEfficientTools 错日期 (2026-03-28 -> 2025-02-19)
- 从默认 beta header 移除已 GA 的 BetaFineGrainedToolStreaming /
  BetaTokenEfficientTools (常量保留供客户端显式 merge)
- claudemask.RequiredNodeHeaders 加 X-Claude-Code-Session-Id 强制

新增 ensureClaudeCodeSessionID helper (claude_code_session_id.go):
- 真实 CLI 在 SDK 内强制 X-Claude-Code-Session-Id:y_(),缺失被判第三方
- OAuth mimic 路径: metadata.user_id 派生 -> canonical UUID 写入
  -> 兜底 uuid.NewString()
- API key passthrough 路径: 不从 body 派生,保护客户端原始语义
- 所有路径均对客户端传入的非法 UUID 执行删除 (避免恶意值上游透传)
- 所有写入 header 的 session-id 都通过 uuid.Parse 校验

测试:
- 新增 14 个 ensureClaudeCodeSessionID 单元测试,含恶意 UUID 注入拒绝
  + API key 路径隔离 + canonical 形式校验
- 新增 3 个 bundle 派生一致性测试
- mask_test 加 session-id 缺失校验 case
- 老 UA 断言 2.1.104 -> 2.1.145

不在范围:
- TLS 指纹 (utls 已处理)
- Bun.hash vs xxHash64 算法验证 (需 golden vectors,独立项目)

References:
- VERSION:2.1.145 BUILD_TIME:2026-05-19T01:36:35Z
  GIT_SHA:daa4c3755d45ab0cf97bb41db8c03bd2dfd2ff5f
 2026-05-20 17:18:47 +08:00
wucm667
90b2b2a757 feat(usage): 用户 API Key 用量页支持按日明细 2026-05-20 15:48:38 +08:00
Wesley Liddick
7ec61eb2f5
Merge pull request #2606 from wucm667/fix/openai-responses-respect-force-chat-completions 
fix(openai): /v1/responses 入口尊重 force_chat_completions 设置
 2026-05-20 15:13:43 +08:00
shaw
878ad3b569 feat(openai-gateway): Codex OAuth 账号浏览器 UA 自动改写规避 Cloudflare 
质询
 2026-05-20 14:33:51 +08:00
wucm667
cae93ae137 fix(openai): /v1/responses respect force chat completions 2026-05-20 14:17:26 +08:00
win
158785bfc9 chore: merge upstream v0.1.127 — keep omniroute customizations 
Upstream highlights:
- v0.1.127 release (150 commits): channel-monitor 协议管理、OpenAI
  Responses 路由配置、模型定价 LiteLLM 默认、payment 强制扫码、
  钉钉 OAuth、用户用量按平台拆分、Ops 错误分类 SLA 调整、
  Anthropic passthrough keepalive、Gemini chat completions 路由 ...
- 91da8159 feat(risk-control): 内容审计新增关键词拦截
- 3d22dd34 feat: gemini-3.5-flash 模型支持

Conflicts resolved:
- Dockerfile: keep pnpm pin to 9.15.9 (upstream pinned generic v9 floating).
- wire_gen.go: combine upstream NewSettingHandler(+userAttributeService)
  with local NewOpsHandler(opsService, requestEventBus, opsLogBroadcaster).
  Verified by re-running wire generate.
- scheduler_cache.go: keep both upstream openai_responses_{mode,supported}
  keys and local model_rate_limits key in filterSchedulerExtra().
- gateway_service.go: keep local context-compression block; drop now-dead
  setOpsUpstreamRequestBody call (upstream removed ops retry replay).
- docker-compose.yml: keep local windsurf-ls service profile and named
  volumes; keep local healthcheck start_period values.

Test mock signatures bumped to match current constructors:
- gateway_models_test.go: add nil for RPMTokenBucketService.
- account_handler_available_models_test.go: add nil for windsurfChatService.
 2026-05-20 12:39:40 +08:00
shaw
3d22dd34d3 feat: add gemini-3.5-flash model support across backend and frontend 2026-05-20 09:28:46 +08:00
wucm667
276b5c7755 fix(apicompat): strip temperature/top_p for reasoning models in Responses conversion 
gpt-5.x models served via the OpenAI Responses API reject requests that
include temperature or top_p with:
  {"detail":"Unsupported parameter: temperature"}

This caused ClaudeCode agent/subagent tool requests to fail with a 400
error when an OpenAI group had the Messages-format support enabled.

Root cause: AnthropicToResponses and ChatCompletionsToResponses were
unconditionally forwarding temperature and top_p from the incoming
request to the ResponsesRequest, even though all gpt-5.x reasoning
models reject these sampling parameters.

Fix:
- Add isReasoningModel(model string) bool helper that returns true for
  any model whose name starts with "gpt-5".
- Skip temperature and top_p when converting to ResponsesRequest for
  reasoning models. Non-reasoning models (e.g. gpt-4o) are unaffected.
- ResponsesRequest.Temperature and TopP are already *float64 with
  omitempty, so nil values are safely omitted from the JSON body.

Tests:
- TestAnthropicToResponses_TemperatureStrippedForReasoningModel
- TestAnthropicToResponses_TemperatureStrippedForAllGpt5Variants
- TestChatCompletionsToResponses_TemperatureStrippedForReasoningModel
- TestChatCompletionsToResponses_TemperaturePreservedForNonReasoningModel

Fixes #2487
 2026-05-19 20:03:16 +08:00
Wesley Liddick
e65fb8b086
Merge pull request #2543 from L494264Tt/fix/deepseek-reasoning-content 
fix: preserve DeepSeek reasoning_content in chat compatibility paths
 2026-05-19 17:34:58 +08:00
L494264Tt
fe3283a1d5 fix: satisfy errcheck for reasoning content conversion 2026-05-19 17:17:39 +08:00
Wesley Liddick
6c8b6843fd
Merge pull request #2546 from nanobanana123/fix/anthropic-empty-thinking-sse 
fix(apicompat): preserve empty streaming thinking blocks
 2026-05-19 17:03:54 +08:00
L494264Tt
6082d02d22 Merge origin/main into fix/deepseek-reasoning-content 2026-05-19 17:00:57 +08:00
DaydreamCoding
664e9fdcd4 feat(usage): 用户用量按平台拆分 + UsersView 列设置可配置 + 用量列排序 
后端
- BatchUserUsageStats / UserDashboardStats 新增 ByPlatform 字段
  复用 ops 路径 COALESCE(g.platform, a.platform) 语义，不冗余 DB 字段
- 抽出 usageLogEffectivePlatformExpr 常量供管理员与用户两路径共用
- GetBatchUsersUsage cacheKey 加 v=2 + 当日日期，修复跨午夜旧缓存兼容新字段

前端
- 新建 PlatformUsageBreakdown：管理员用量列 hover tooltip 展示各平台 today/total
- 新建 PlatformCostCell：单平台 today/total 紧凑单元格
- UsersView 列设置新增 Claude/OpenAI/Gemini/Antigravity 四个平台子列，默认隐藏可手动启用
- 普通用户 Dashboard 新增 Row 3 平台拆分卡片，受 isSimple 控制
- 平台之和 < 总值时显式展示"其他"行，避免数字对不齐
- last_active_at 从 FORCED_VISIBLE_COLUMNS 移除，允许用户隐藏并持久化
- 列设置加 schema 版本号 + 迁移机制，老用户升级时新增默认隐藏列自动应用
- UsersView 用量列（汇总 + 4 平台子列）加入前端单页排序：列头单按钮 + 弹出菜单
  切换"今日 / 近30天"，三态循环 desc → asc → off；菜单底部备注"仅对本页数据排序"
- sortedUsers computed 在 server-side-sort 结果之上叠加本地排序，缺失值按 0 处理；
  usageSort 状态独立 localStorage 持久化，互不干扰后端 sort_by
- i18n 新增 admin.users.sortBy / sortCurrentPageOnly

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-19 15:25:34 +08:00
Wesley Liddick
e365aae450
Merge pull request #2450 from wucm667/codex/issue-2431-responses-api-support 
feat: 支持后台配置 OpenAI Responses API 路由
 2026-05-19 14:47:10 +08:00
Wesley Liddick
23e95b77b7
Merge pull request #2528 from wucm667/fix/openai-responses-null-content 
fix(openai): 修复 chat-completions 转 responses 时 content 为 null 导致上游 400
 2026-05-19 14:43:55 +08:00
Wesley Liddick
03473d3ee8
Merge pull request #2554 from Arron196/feature/sync-upstream-models-pr 
feat: 支持从上游同步账号可用模型列表
 2026-05-19 14:42:47 +08:00
benjamin
b9ecf25207 fix: harden Antigravity model list requests 
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)

Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
 2026-05-18 19:01:23 +08:00
nanobabanan
e9a25e7b92 fix(apicompat): preserve empty streaming thinking blocks 
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)

Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>
 2026-05-18 15:10:35 +08:00
L494264Tt
1d47fd6300 This preserves DeepSeek reasoning_content across chat compatibility paths. 
DeepSeek thinking-mode tool-call conversations may require the assistant
  reasoning_content from previous turns to be sent back in later requests. Without
  preserving it, those conversations can fail or lose reasoning context.

  Changes:
  - Preserve assistant reasoning_content when converting Chat Completions messages
    to Responses input by wrapping it as a thinking block.
  - Add regression coverage for non-streaming DeepSeek responses.
  - Add regression coverage for streaming DeepSeek deltas.
  - Add regression coverage that request-side messages[].reasoning_content is
    passed through with tool calls.

  Tests:
  go test -tags=unit ./internal/pkg/apicompat ./internal/service -run
  'TestChatCompletionsToResponses_AssistantReasoningContentPreserved|
  TestChatCompletionsToResponses_AssistantThinkingTagPreserved|
  TestForwardAsRawChatCompletions_PreservesDeepSeekReasoningContent|
  TestForwardAsRawChatCompletions_ForcesStreamUsageUpstreamAndPassesUsageDownstream
 2026-05-18 11:22:23 +08:00
wucm667
df82a3bc69 fix(openai): avoid null content when converting chat-completions to responses 
When a chat-completions message has no usable content parts (empty array,
empty text part, or filtered-out image part), marshalChatInputContent
marshalled a nil slice to JSON null. The upstream Responses API rejects a
null content field with HTTP 400. Fall back to an empty string instead.

Fixes #2515
 2026-05-17 11:20:05 +08:00
name
0393bd7c82 Fix OpenAI compat usage parsing 2026-05-16 03:03:43 +08:00
wucm667
862819042c feat(openai): 支持后台配置 Responses API 路由 2026-05-14 11:46:24 +08:00
win
35c6c2b097 chore: merge upstream v0.1.126 — Airwallex, OpenAI fixes, Antigravity UA config 
吸收上游 26 个新 commit:
- feat: Airwallex 支付 + 多币种支持 (b23055af)
- feat: Antigravity user agent 版本可配置 (a07a0dac)
- fix(mimic): 同步 messages 里 tool_use 名称 (f97b8534)
- fix: cache_control 改写默认关闭 (9377c967)
- fix(openai): 多 tool_use 上下文延续 (87d73236)
- fix(openai): 未定价模型零成本记录 (6d69ae87)
- fix(openai): WS replay tool 输出延续 (16a31557)
- fix(openai): 429 plan type 同步 (c3a14717)
- fix(gemini): Vertex token 走 account proxy (2a17c0b2)
- fix(ccswitch): codex 模型 import deeplink (65493df9)
- fix: 订单详情/支付页 NaN 修复 (ba1c6fa5, 6884b03e)
- 系统设置标签导航优化 (18cc4691)

本地解决:
- config.go CSP: 合并 Firebase Auth (Windsurf) + Airwallex 域名
- KeysView.vue: 删除死代码（已被 buildCcSwitchImportDeeplink 取代）
- ccswitchImport.ts: 补充 windsurf 平台 case
- 修复 NewOpsHandler/RegisterGatewayRoutes/SelectAccountWithScheduler 测试签名

保留:
- Antigravity newapi 兼容 (ForwardUpstream /v1/messages 透传)
- Antigravity 核心（gateway_service, oauth, client, credits_overages 等）
- Windsurf 全套
- Claude 网关 + TLS 指纹路由
- 其他本地 feat: P2C 调度 / viewer / context 压缩 / RPM / fallback / health
 2026-05-12 13:48:40 +08:00
win
8d0ef3d2ef chore: antigravity 瘦身，删除边缘文件保留核心 2026-05-12 12:19:22 +08:00
shaw
a07a0dac63 feat: add configurable Antigravity user agent version 2026-05-11 22:25:20 +08:00
shaw
297b54d066 fix: 完善工具名改写测试和格式 2026-05-11 17:27:04 +08:00
win
7347dfffc1 chore: merge upstream v0.1.124-125, keep Windsurf/Antigravity customizations 
Upstream changes:
- feat: 邮箱 + GitHub + Google OAuth 快捷登录
- feat: Codex image bridge 开关
- feat: 内容审核 (content moderation) — 新增 contentModerationService/Handler
- feat: redeem code 返利、批量并发 API、markdown 页面渲染
- feat: 登录注册条款确认
- fix(security): pages API 加 JWT + 可见性校验
- fix: 修复 markdown 页面图片路径
- fix(gateway): 不再默认注入 redact thinking beta
- fix: 稳定 anthropic passthrough 超时错误
- chore: VERSION 升到 0.1.125 + golang:1.26.3-alpine

Conflict resolutions:
- Dockerfile/backend/Dockerfile: 取 upstream golang:1.26.3-alpine
- backend/go.mod: 取 upstream term v0.42.0，保留定制 protobuf v1.36.10
- frontend/src/api/admin/index.ts: 并集 (windsurf + riskControl)
- backend/cmd/server/wire_gen.go: 接 upstream contentModeration*，保留 windsurfHandler/windsurfGatewayService/billingCacheService/requestEventBus；并通过 wire 重生成
- frontend/src/views/admin/AccountsView.vue: 采用 upstream 双层布局 + OpenAI Meta，保留 is_enterprise prop 和 Windsurf tier badge

Note:
- WIP commit (de048fad) preserved Windsurf tier access service / NLU
  extractor / ops log stream / Google OAuth login modal et al before merge.
- 3 pre-existing go vet issues in test files (NewOpsHandler, RegisterGatewayRoutes,
  DefaultCLIProductVersion) are unrelated to this merge — leftover from local
  customization refactors; production code (go build ./...) passes.
 2026-05-09 01:42:39 +08:00
win
de048fad25 chore(wip): save Windsurf/Antigravity/ops customizations before upstream merge 
WIP commit保存以下定制工作以便后续合并 upstream v0.1.124-125:
- Windsurf: tier access service, NLU extractor, cold threshold, Google login
- Antigravity: client/oauth 调整
- Ops: log stream handler/broadcaster/middleware, OpsLogStreamView
- Frontend: WindsurfLoginModal Google, GoogleIcon, AccountsView, sidebar/router/i18n
 2026-05-09 00:41:19 +08:00
shaw
57fd7998d3 fix(gateway): stop default redact thinking beta injection 2026-05-07 18:56:11 +08:00
win
3fe228d143 chore: merge upstream v0.1.122-123, keep Windsurf/Antigravity customizations 
New upstream features:
- feat: improve OpenAI messages compatibility for Claude Code
- feat: image generation stream & concurrency controls
- fix(rate-limit): remove 429 cooldown config option
- fix: skip previous_response_id recovery when payload has function_call_output
- feat: support select search in group/account views
- fix: ops cleanup settings
- chore: remove openspec and update axios

Conflict resolutions:
- config.go: kept AntigravityLSWorker+NodeTLSProxy AND added ImageConcurrency
- account_test_service.go: kept windsurf import AND added openai_compat import
- docker-compose.yml: kept Windsurf env vars AND added image concurrency env vars
 2026-05-06 11:50:54 +08:00
lyen1688
0584305e5a feat: improve OpenAI messages compatibility for Claude Code 2026-05-05 19:36:33 +08:00
Wesley Liddick
dc09b367dc
Merge pull request #2143 from alfadb/fix/openai-apikey-cc-default-routing 
修复：APIKey 账户上游不支持 OpenAI Responses API 时的 Chat Completions 路由回退
 2026-05-03 22:58:26 +08:00
shaw
72d5ee4cd1 fix: drain OpenAI compat streams for usage 2026-05-03 17:11:27 +08:00
win
c5eb305f7f chore: merge upstream v0.1.119-121, keep Windsurf/Antigravity customizations 
Upstream changes merged:
- fix(scheduler): resolve SetSnapshot race conditions with Lua CAS script
- fix: improve sticky session scheduling (debug logs + layer 1.5 checks)
- feat: Anthropic cache TTL injection toggle
- fix(gateway): stream EOF failover + sanitize stream errors
- feat(httputil): zstd/gzip/deflate request decompression + bomb guard
- feat(openai): OpenAI Fast/Flex Policy (HTTP + WebSocket + Admin)
- feat(vertex): Vertex Service Account support
- feat: account bulk edit scope and compact settings
- feat(affiliate): rebate freeze migration
- fix(openai): various fixes (passthrough fields, compact payload, etc.)

Conflict resolutions:
- domain/constants.go: keep both AccountTypeWindsurfSession + AccountTypeServiceAccount
- scheduler_cache_unit_test.go: keep both test functions
- gateway_service.go: remove dead code (claudeCodeUserAgentRe, isClaudeCodeRequest)
- wire_gen.go: keep Windsurf service chain + add upstream claudeTokenProvider param
- frontend/src/types/index.ts: keep windsurf + service_account types
- frontend CreateAccountModal.vue: keep Windsurf login + Vertex service_account blocks
- frontend PlatformTypeBadge.vue: keep both Session + Vertex cases
- account_test_service.go: fix createTestPayload call to pass empty prompt arg
 2026-05-02 16:52:21 +08:00
alfadb-bot
4e4cc80971 fix(openai-gateway): route APIKey accounts to /v1/chat/completions when upstream lacks Responses API 
OpenAI APIKey accounts with base_url pointing to third-party OpenAI-compatible
upstreams (DeepSeek, Kimi, GLM, Qwen, etc.) were failing because the gateway
unconditionally converted Chat Completions requests to Responses format and
forwarded to {base_url}/v1/responses, which only exists on OpenAI's official
endpoint.

Detection-based routing:
- Probe upstream capability on account create/update via a minimal POST to
  /v1/responses; HTTP 404/405 means 'unsupported', any other response means
  'supported'.
- Persist result as accounts.extra.openai_responses_supported (bool).
- ForwardAsChatCompletions branches at function entry: APIKey accounts with
  explicit support=false go through new forwardAsRawChatCompletions which
  passthrough-forwards CC body to /v1/chat/completions without protocol
  conversion.

Default behavior for accounts without the marker preserves the legacy
'always Responses' path — existing OpenAI APIKey accounts that were working
before this change continue to work without modification (the 'reality is
evidence' principle: an account that has been running implies upstream
capability).

Probe is fired async after Create / Update / BatchCreate; failures only log,
never block the admin flow. BulkUpdate omitted (low signal of base_url
changes; can be added if needed).

Implementation:
- New pkg internal/pkg/openai_compat: marker key + ShouldUseResponsesAPI
- New service file openai_apikey_responses_probe.go: probe + persist
- New service file openai_gateway_chat_completions_raw.go: CC pass-through
- Account test endpoint short-circuits with explicit message for
  probed-unsupported accounts (full CC test path is a TODO)

Zero schema changes, zero migrations, zero frontend changes, zero wire
modifications — all wired through existing AccountTestService injection.

Closes: DeepSeek-OpenAI account (id=128) production failure
 2026-04-30 19:25:45 +08:00
shaw
40feb86ba4 fix(httputil): add decompression bomb guard and fix errcheck lint 2026-04-29 22:11:45 +08:00