Commit Graph
327 Commits
Author SHA1 Message Date
Finessse a31a7eac0c feat(executor): parse Retry-After headers in antigravity executor
Currently, the executor ignores upstream retry-after headers, causing the manager to default to a 1s backoff. This change parses standard HTTP Retry-After headers (seconds or date) and passes them to the error handler.
2025-12-20 03:59:41 +07:00
Finessse e76aacf5ec feat(executor): enhance Antigravity executor and handlers
- Add improvements to antigravity_executor.go for better CLI communication
- Update handlers.go with additional helper functions
2025-12-20 02:25:56 +07:00
Finessse 2e21fb75d5 fix(gemini-vertex): append api key to url query params 2025-12-20 00:29:57 +07:00
Finessse 88224aab1a fix(gemini): append api key to url query params 2025-12-20 00:25:08 +07:00
Luis Pater 843316ea7a Merge branch 'router-for-me:main' into main 2025-12-19 22:24:26 +08:00
hkfires 2039062845 fix(gemini): add optional skip for gemini3 thinking conversion 2025-12-19 22:07:43 +08:00
Finessse b8f6781bb8 feat(runtime): update executor and runtime internals 2025-12-19 18:18:40 +07:00
evann 404546ce93 refactor(antigravity): regarding production endpoint caching 2025-12-19 16:36:54 +07:00
evann 9058d406a3 feat(antigravity): enhance prompt caching support and update agent version 2025-12-19 16:33:41 +07:00
이대희 b6ba15fcbd fix(runtime/executor): Antigravity executor schema handling and Claude-specific headers 2025-12-19 10:28:23 +09:00
Finessse 15a2aa92ad models: add gemini-3-flash across providers 2025-12-18 19:31:20 +07:00
Finessse cd1a61a8a5 antigravity: populate token limits for /v1/models 2025-12-18 15:56:40 +07:00
Finessse 99eb979353 models: add gemini-3-flash and gemini-3-pro-low (antigravity) 2025-12-18 14:17:57 +07:00
Luis Pater 0f646800f6 Merge branch 'router-for-me:main' into main 2025-12-18 08:36:59 +08:00
Luis Pater 13eb5268de Merge pull request #582 from ben-vargas/fix-gemini-3-thinking-level
feat: use thinkingLevel for Gemini 3 models per Google documentation
2025-12-18 07:19:37 +08:00
Ben Vargas 598f0af19b fix: apply thinkingLevel from model suffix metadata for Gemini 3
The previous commit added thinkingLevel support but didn't apply it
when the reasoning effort came from model name suffix (e.g., model(minimal)).

This was because ResolveThinkingConfigFromMetadata returns nil for
level-based models, bypassing the metadata application.

Changes:
- Add ApplyGemini3ThinkingLevelFromMetadata for standard Gemini API
- Add ApplyGemini3ThinkingLevelFromMetadataCLI for CLI API format
- Update gemini_cli_executor to apply Gemini 3 thinkingLevel from metadata
- Update antigravity_executor to apply Gemini 3 thinkingLevel from metadata
- Update aistudio_executor to apply Gemini 3 thinkingLevel from metadata
- Add comprehensive test coverage for Gemini 3 thinkingLevel functions
2025-12-17 16:08:38 -07:00
Ben Vargas a33f5d31fc feat: use thinkingLevel for Gemini 3 models per Google documentation
Per Google's official documentation, Gemini 3 models should use
thinkingLevel (string) instead of thinkingBudget (number) for
optimal performance.

From Google's Gemini Thinking docs:
> Use the thinkingLevel parameter with Gemini 3 models. While
> thinkingBudget is accepted for backwards compatibility, using
> it with Gemini 3 Pro may result in suboptimal performance.

Changes:
- Add model family detection functions (IsGemini3Model, IsGemini25Model,
  IsGemini3ProModel, IsGemini3FlashModel)
- Add ApplyGeminiThinkingLevel and ApplyGeminiCLIThinkingLevel functions
  for applying thinkingLevel config
- Add ValidateGemini3ThinkingLevel for model-specific level validation
- Add ThinkingBudgetToGemini3Level for backward compatibility conversion
- Update NormalizeGeminiThinkingBudget to convert budget to level for
  Gemini 3 models
- Update ApplyDefaultThinkingIfNeeded to not set a default level for
  Gemini 3 (lets API use its dynamic default "high")
- Update ConvertThinkingLevelToBudget to preserve thinkingLevel for
  Gemini 3 models
- Add Levels field to all Gemini 3 model definitions:
  - Gemini 3 Pro: ["low", "high"]
  - Gemini 3 Flash: ["minimal", "low", "medium", "high"]

Backward compatibility:
- Gemini 2.5 models continue to use thinkingBudget as before
- If thinkingBudget is provided for Gemini 3, it's converted to the
  appropriate thinkingLevel
- Existing configurations continue to work
2025-12-17 15:28:20 -07:00
Ravens2121 54acd69e9d Merge branch 'router-for-me:main' into master 2025-12-18 04:39:17 +08:00
Ravens2121 d687ee2777 feat(kiro): implement official reasoningContentEvent and improve metadat 2025-12-18 04:38:22 +08:00
Luis Pater f7b17ee6ec Merge pull request #36 from rezhajulio/feat/gpt-5.2
Add GPT-5.2 model support for GitHub Copilot
2025-12-18 03:16:25 +08:00
Luis Pater 408614c74c Merge branch 'router-for-me:main' into main 2025-12-18 03:13:48 +08:00
Luis Pater 68a27772b3 feat(antigravity): enable token counting via API with resilient routing
Introduces the capability to count tokens for Antigravity-backed requests. This implementation leverages the `countTokens` endpoint of the Antigravity API, replacing the prior unsupported stub.

Key aspects of this update include:

- **API Integration**: Direct integration with the Antigravity `countTokens` API, including necessary request payload translation and authentication.
- **Resilient Infrastructure**: A fallback mechanism has been established, allowing the system to attempt connections across multiple Antigravity base URLs to ensure request success even in the event of temporary service interruptions.
- **Model Aliasing**: Added mappings for `gemini-3-flash` and `gemini-3-flash-preview` to ensure compatibility with the latest model variants.
- **Robust Error Handling**: Comprehensive error handling and logging are in place to manage failures during API interactions.
2025-12-18 03:12:46 +08:00
Luis Pater 10e0ea1309 Merge main into pr-39 2025-12-18 00:36:51 +08:00
Luis Pater 5fda6f8ef3 feat(antigravity): implement non-streaming execution for Claude model requests 2025-12-17 23:17:11 +08:00
Luis Pater 09923f654c feat(antigravity): add streaming support for Claude model requests 2025-12-17 22:16:57 +08:00
이대희 1b8e538a77 feature: Improves Gemini JSON schema compatibility
Enhances compatibility with the Gemini API by implementing a schema cleaning process.

This includes:
- Centralizing schema cleaning logic for Gemini in a dedicated utility function.
- Converting unsupported schema keywords to hints within the description field.
- Flattening complex schema structures like `anyOf`, `oneOf`, and type arrays to simplify the schema.
- Handling streaming responses with empty tool names, which can occur in subsequent chunks after the initial tool use.
2025-12-17 17:10:53 +09:00
Finessse e7e1b26f11 fix(antigravity): keep scalar text parts 2025-12-17 14:19:45 +07:00
Rezha Julio 92c62bb2fb Add GPT-5.2 model support for GitHub Copilot 2025-12-17 02:15:10 +07:00
Luis Pater 1efade8bdb Merge branch 'main' into plus 2025-12-17 02:50:14 +08:00
hkfires 28a428ae2f fix(thinking): align budget effort mapping across translators
Unify thinking budget-to-effort conversion in a shared helper, handle disabled/default thinking cases in translators, adjust zero-budget mapping, and drop the old OpenAI-specific helper with updated tests.
2025-12-16 18:34:43 +08:00
hkfires b326ec3641 feat(iflow): add thinking support for iFlow models 2025-12-16 18:34:43 +08:00
Ravens2121 f3d1cc8dc1 chore: change debug logs from INFO to DEBUG level 2025-12-16 05:32:03 +08:00
Ravens2121 0a3a95521c feat: enhance thinking mode support for Kiro translator
Changes:
2025-12-16 05:01:40 +08:00
Luis Pater 407020de0c Merge branch 'router-for-me:main' into main 2025-12-15 10:36:39 +08:00
hkfires 367a05bdf6 refactor(thinking): export thinking helpers
Expose thinking/effort normalization helpers from the executor package
so conversion tests use production code and stay aligned with runtime
validation behavior.
2025-12-15 09:16:15 +08:00
hkfires 712ce9f781 fix(thinking): drop unsupported none effort
When budget 0 maps to "none" for models that use thinking levels
but don't support that effort level, strip thinking fields instead
of setting an invalid reasoning_effort value.
Tests now expect removal for this edge case.
2025-12-15 09:16:14 +08:00
hkfires 27a5ad8ec2 Fixed: #534
fix(aistudio): correct JSON string boundary detection for backslash sequences
2025-12-15 09:00:14 +08:00
Finessse 4b4cd108ce Fix tool result ordering for Droid CLI 2025-12-14 23:34:54 +07:00
Ravens2121andAmp de0ea3ac49 fix(kiro): Always parse thinking tags from Kiro API responses
Amp-Thread-ID: https://ampcode.com/threads/T-019b1c00-17b4-713d-a8cc-813b71181934
Co-authored-by: Amp <amp@ampcode.com>
2025-12-14 16:46:17 +08:00
Ravens2121 12116b018d Merge branch 'router-for-me:main' into master 2025-12-14 16:42:30 +08:00
Ravens2121 c3ed3b40ea feat(kiro): Add token usage cross-validation and simplify thinking mode handling 2025-12-14 16:40:33 +08:00
Luis Pater b80c2aabb0 Merge branch 'router-for-me:main' into main 2025-12-14 16:19:29 +08:00
Luis Pater 14ce6aebd1 Merge pull request #449 from sususu98/fix/gemini-cli-429-retry-delay-parsing
fix(gemini-cli): enhance 429 retry delay parsing
2025-12-14 14:04:14 +08:00
Ravens2121 9c04c18c04 feat(kiro): enhance request translation and fix streaming issues
English:
- Fix <thinking> tag parsing: only parse at response start, avoid misinterpreting discussion text
- Add token counting support using tiktoken for local estimation
- Support top_p parameter in inference config
- Handle max_tokens=-1 as maximum (32000 tokens)
- Add tool_choice and response_format parameter handling via system prompt hints
- Support multiple thinking mode detection formats (Claude API, OpenAI reasoning_effort, AMP/Cursor)
- Shorten MCP tool names exceeding 64 characters
- Fix duplicate [DONE] marker in OpenAI SSE streaming
- Enhance token usage statistics with multiple event format support
- Add code fence markers to constants

中文:
- 修复 <thinking> 标签解析:仅在响应开头解析,避免误解析讨论文本中的标签
- 使用 tiktoken 实现本地 token 计数功能
- 支持 top_p 推理配置参数
- 处理 max_tokens=-1 转换为最大值(32000 tokens)
- 通过系统提示词注入实现 tool_choice 和 response_format 参数支持
- 支持多种思考模式检测格式(Claude API、OpenAI reasoning_effort、AMP/Cursor)
- 截断超过64字符的 MCP 工具名称
- 修复 OpenAI SSE 流中重复的 [DONE] 标记
- 增强 token 使用量统计,支持多种事件格式
- 添加代码围栏标记常量
2025-12-14 11:57:16 +08:00
Ravens2121 81ae09d0ec Merge branch 'kiro-refactor-backup' 2025-12-14 07:03:24 +08:00
Ravens2121 01cf221167 feat(kiro): 代码优化重构 + OpenAI翻译器实现 2025-12-14 06:58:50 +08:00
Luis Pater 79033aee34 Merge branch 'main' into plus 2025-12-14 00:07:46 +08:00
Ravens2121 1ea0cff3a4 fix: add missing import declarations for net and time packages 2025-12-13 12:57:47 +08:00
Ravens2121 75793a18f0 feat(kiro): Add Kiro OAuth login entry and auth file filter in Web UI
为Kiro供应商添加WEB UI OAuth登录入口和认证文件过滤器

## Changes / 更改内容

### Frontend / 前端 (management.html)
- Add Kiro OAuth card UI with support for AWS Builder ID, Google, and GitHub login methods
- 添加Kiro OAuth卡片UI,支持AWS Builder ID、Google和GitHub三种登录方式
- Add i18n translations for Kiro OAuth (Chinese and English)
- 添加Kiro OAuth的中英文国际化翻译
- Add Kiro filter button in auth files management page
- 在认证文件管理页面添加Kiro过滤按钮
- Implement JavaScript methods: startKiroOAuth(), openKiroLink(), copyKiroLink(), copyKiroDeviceCode(), startKiroOAuthPolling(), resetKiroOAuthUI()
- 实现JavaScript方法:startKiroOAuth()、openKiroLink()、copyKiroLink()、copyKiroDeviceCode()、startKiroOAuthPolling()、resetKiroOAuthUI()

### Backend / 后端
- Add /kiro-auth-url endpoint for Kiro OAuth authentication (auth_files.go)
- 添加/kiro-auth-url端点用于Kiro OAuth认证 (auth_files.go)
- Fix GetAuthStatus() to correctly parse device_code and auth_url status
- 修复GetAuthStatus()以正确解析device_code和auth_url状态
- Change status delimiter from ':' to '|' to avoid URL parsing issues
- 将状态分隔符从':'改为'|'以避免URL解析问题
- Export CreateToken method in social_auth.go
- 在social_auth.go中导出CreateToken方法
- Register Kiro OAuth routes in server.go
- 在server.go中注册Kiro OAuth路由

## Files Modified / 修改的文件
- management.html
- internal/api/handlers/management/auth_files.go
- internal/api/server.go
- internal/auth/kiro/social_auth.go
2025-12-13 11:39:22 +08:00
Ravens2121 58866b21cb feat: optimize connection pooling and improve Kiro executor reliability
## 中文说明

### 连接池优化
- 为 AMP 代理、SOCKS5 代理和 HTTP 代理配置优化的连接池参数
- MaxIdleConnsPerHost 从默认的 2 增加到 20,支持更多并发用户
- MaxConnsPerHost 设为 0(无限制),避免连接瓶颈
- 添加 IdleConnTimeout (90s) 和其他超时配置

### Kiro 执行器增强
- 添加 Event Stream 消息解析的边界保护,防止越界访问
- 实现实时使用量估算(每 5000 字符或 15 秒发送 ping 事件)
- 正确从上游事件中提取并传递 stop_reason
- 改进输入 token 计算,优先使用 Claude 格式解析
- 添加 max_tokens 截断警告日志

### Token 计算改进
- 添加 tokenizer 缓存(sync.Map)避免重复创建
- 为 Claude/Kiro/AmazonQ 模型添加 1.1 调整因子
- 新增 countClaudeChatTokens 函数支持 Claude API 格式
- 支持图像 token 估算(基于尺寸计算)

### 认证刷新优化
- RefreshLead 从 30 分钟改为 5 分钟,与 Antigravity 保持一致
- 修复 NextRefreshAfter 设置,防止频繁刷新检查
- refreshFailureBackoff 从 5 分钟改为 1 分钟,加快失败恢复

---

## English Description

### Connection Pool Optimization
- Configure optimized connection pool parameters for AMP proxy, SOCKS5 proxy, and HTTP proxy
- Increase MaxIdleConnsPerHost from default 2 to 20 to support more concurrent users
- Set MaxConnsPerHost to 0 (unlimited) to avoid connection bottlenecks
- Add IdleConnTimeout (90s) and other timeout configurations

### Kiro Executor Enhancements
- Add boundary protection for Event Stream message parsing to prevent out-of-bounds access
- Implement real-time usage estimation (send ping events every 5000 chars or 15 seconds)
- Correctly extract and pass stop_reason from upstream events
- Improve input token calculation, prioritize Claude format parsing
- Add max_tokens truncation warning logs

### Token Calculation Improvements
- Add tokenizer cache (sync.Map) to avoid repeated creation
- Add 1.1 adjustment factor for Claude/Kiro/AmazonQ models
- Add countClaudeChatTokens function to support Claude API format
- Support image token estimation (calculated based on dimensions)

### Authentication Refresh Optimization
- Change RefreshLead from 30 minutes to 5 minutes, consistent with Antigravity
- Fix NextRefreshAfter setting to prevent frequent refresh checks
- Change refreshFailureBackoff from 5 minutes to 1 minute for faster failure recovery
2025-12-13 10:21:40 +08:00