i/Aether
1
0
Fork 0
mirror of https://github.com/fawney19/Aether.git synced 2026-09-02 09:20:22 +08:00
Code Issues Packages Projects Releases Wiki Activity
1,377 Commits 5 Branches 174 Tags
627eb4f3356639b7489b31fbf3e4c734ca66566a
Commit Graph

142 Commits

Author SHA1 Message Date
fawney19
627eb4f335 Merge pull request #371 from Kayphoon/feature/embedding-model-support 
feat: add embedding and rerank support
 2026-05-04 12:56:11 +08:00
fawney19
099653f732 Cap Codex pool cooldowns and add key circuit breaker 2026-05-04 02:15:32 +08:00
fawney19
1d62722d47 Improve Codex model fetching 2026-05-04 01:16:56 +08:00
fawney19
fb642f7d62 feat: improve OAuth provider configuration 2026-05-04 00:18:05 +08:00
fawney19
cb5b8ee1ee Use Rust build metadata for system version 2026-05-03 22:06:16 +08:00
Kayphoon
5abe664d65 feat: add embedding and rerank support 2026-05-03 17:32:41 +08:00
fawney19
c4ea042eb4 feat: add model directive management 2026-05-03 14:48:25 +08:00
Kayphoon
fe27fb17fb feat: add provider-key concurrent limit (#352) 
* feat: add provider-key concurrent limit

* Fix provider key concurrent limit checks

---------

Co-authored-by: fawney19 <elky0401@gmail.com>
 2026-05-03 01:55:23 +08:00
fawney19
e3ea2d1451 feat: configure auth channel mismatch formats 2026-05-03 00:49:22 +08:00
fawney19
3a770306cc refactor: separate request auth channel from route kind 
Extract request_auth_channel as a distinct routing dimension to distinguish
between api_key and bearer_like authentication methods. This improves routing
clarity by decoupling authentication mechanism from route classification.
 2026-05-02 21:23:33 +08:00
jiuwovo-ai
bfd1ea72b3 fix: codex free accounts incorrectly marked as quota exhausted (#366) 
* fix: codex free accounts incorrectly marked as quota exhausted when skip_exhausted_accounts enabled

- Cross-validate stored exhausted flag against window used_ratio in snapshot reader
- Guard has_credits:false check with window data presence in upstream_metadata fallback
- Add windows.is_empty() precondition to exhausted_by_credits in snapshot builder

Free codex accounts have has_credits:false (they use window-based quotas, not credits),
but the old logic treated this as credits depleted, skipping actual window usage checks.
This caused all free accounts to be incorrectly marked exhausted.

* test: cover codex free quota windows

---------

Co-authored-by: root <root@ser488556369540.local>
Co-authored-by: fawney19 <elky0401@gmail.com>
 2026-05-02 14:11:29 +08:00
fawney19
c130d0e2c9 refactor ai serving modules and crates 2026-05-02 13:23:54 +08:00
fawney19
4fc7cecf30 Fix OAuth token import and table filters 2026-05-01 02:14:49 +08:00
fawney19
33aa70c22b fix scheduler affinity candidate selection 2026-04-30 16:27:24 +08:00
fawney19
7fcb9e6292 feat(usage): record request origin metadata (#362) 
Record client IP and User-Agent in usage request_metadata for local execution requests without adding schema, API, or table fields.

References #357 and supersedes #358.

Co-authored-by: RWDai <27391645+RWDai@users.noreply.github.com>
 2026-04-30 11:26:56 +08:00
RWDai
2b5247b9a8 fix(data): repair request candidate epoch created_at (#343) 
Co-authored-by: fawney19 <elky0401@gmail.com>
 2026-04-30 09:23:21 +08:00
Entropy.Xu
4d59d518d1 fix: correct API key expiry and dashboard savings (#361) 2026-04-30 00:35:32 +08:00
fawney19
9e2faa7e5b Tighten local auth allow-list matching 2026-04-30 00:26:33 +08:00
fawney19
a16550249b fix: allow empty provider secrets to attempt requests 2026-04-29 18:55:52 +08:00
fawney19
e751289dfb Support per-format provider key auth 2026-04-29 15:46:50 +08:00
fawney19
07a319259b Normalize canonical API formats 2026-04-29 10:20:41 +08:00
fawney19
230e7d6259 Normalize management token prefixes 2026-04-28 19:46:38 +08:00
fawney19
3d20c05ef7 Fix OAuth refresh through tunnel proxy 2026-04-28 19:45:34 +08:00
fawney19
70f747d406 Add shared OAuth flows 2026-04-28 15:46:21 +08:00
fawney19
29fc0be121 Tighten OAuth refresh consistency 2026-04-28 09:36:42 +08:00
fawney19
ed17147281 fix: align provider oauth refresh redirects 2026-04-27 22:08:02 +08:00
fawney19
a7a3c9f023 Fix OAuth import proxy handling 2026-04-27 20:05:43 +08:00
fawney19
4c16b11cb4 Align runtime miss tests with skipped candidates 2026-04-27 18:09:56 +08:00
fawney19
17f09fc8c1 Expose ranking metadata in request trace 2026-04-27 17:24:32 +08:00
fawney19
3991d47166 Fix format checks and local error messages 2026-04-27 00:27:03 +08:00
fawney19
4ec591fbf2 centralize openai responses alias handling 2026-04-26 23:59:53 +08:00
fawney19
5b914aa78c migrate ai format conversion to responses adapters 2026-04-26 23:59:53 +08:00
AAEE86
f3a9fd4c82 feat(gateway): 补齐号池主动探测后台任务 (#342) 
- 解析 pool_advanced 主动探测开关与间隔配置
- 对齐 Python 版本的探测间隔默认值和范围限制
- 新增号池配额主动探测 worker
- 支持 codex、kiro、antigravity provider
- 使用 Redis 时间戳和 provider 级锁避免重复探测
- 接入 gateway 后台任务并补充测试
 2026-04-26 23:48:31 +08:00
fawney19
baa0ddd787 Improve runtime miss usage diagnostics 2026-04-26 01:44:54 +08:00
fawney19
429fdb47e6 fix(usage): 简化本地执行错误提示 2026-04-25 20:46:22 +08:00
fawney19
912a92cd1a refactor(rules): 规则引擎容错优化，无效规则条目跳过而非中止整个规则集 
- header/body rules 的 _are_locally_supported 简化为仅检查是否为数组
- apply 逻辑中遇到格式错误/不支持的规则条目改为 continue 跳过，而非 return false
- 允许非字符串 header value，自动序列化为 JSON 字符串
- 宽松处理无效 regex flag，不再拒绝整条规则

fix(gateway): Claude CLI 路由仅检查 bearer 头，不排斥同时携带 x-api-key 的请求

feat(observability): 监控链路候选展示解密 auth_config 的账号标签和 OAuth 计划类型
 2026-04-25 19:46:52 +08:00
Entropy.Xu
bc97e383d3 fix(gateway): 对齐号池调度预设行为 (#332) 
- 保持 sticky session 优先于分配模式排序，并支持 sticky_session_ttl_seconds=0 禁用粘滞绑定
- 移除旧版 free_team_first 预设入口，统一使用 free_first / team_first 新模式
- 对齐 plus_first 对 Plus/Pro 账号的优先级处理
- 调整负载均衡排序种子，避免同一分钟内固定命中同一 Key
- multi-score 预设下记录延迟样本，并补充号池调度/配置/运行时测试
- 收敛 request candidate extra_data 构造入参，修复 clippy too_many_arguments
 2026-04-25 16:41:51 +08:00
fawney19
65e915fd1d refactor(api-keys): 网关 API Key 改为 32 位字母数字随机串 
由 UUID 拼接（48 位 hex）改为 base62 字母数字，缩短长度并提升可读性
 2026-04-25 10:04:04 +08:00
Entropy.Xu
67b092253d fix(gateway): 兼容 OpenAI 图像 multipart boundary 大小写 (#331) 2026-04-24 21:41:20 +08:00
fawney19
db1059e73b fix(usage): 修复 Gemini total 重复累加 cache_read 并补充测试 
- write.rs: Gemini usage 提取 total 时不再叠加 cachedContentTokenCount
- 补充 Claude 大 cache_read 场景下 input_tokens 不被扣减的测试
- 补充前端 getEffectiveInputTokens 对 Claude 格式不减 cache_read 的测试
 2026-04-24 21:20:45 +08:00
AAEE86
4618184516 fix: 修复账号批量操作快捷多选不生效和今日 Token 不准确 (#328) 
* fix(pool): 修复账号批量操作快捷多选不生效

- 解析号池账号列表接口的 quick_selectors 查询参数
- 列表查询复用现有快捷筛选匹配逻辑
- 支持冷却列表与普通列表的快捷筛选后分页
- 更新架构测试中的 pool admin helper 边界断言

* fix(dashboard): 今日 Token 改为按分项汇总口径展示

- /api/dashboard/stats 的 today.tokens 改为 输入+输出+写缓存+读缓存
- “今日 Token”卡片 value 与 subValue 口径保持一致
- 同步更新 dashboard 相关测试断言
 2026-04-24 18:36:38 +08:00
Entropy.Xu
fb46fcd80f feat(rust): 对齐 SMTP 测试与注册邮件 TLS 链路 (#325) 
* feat(rust): 对齐 SMTP 连接测试链路

* fix(rust): 修复注册邮件 TLS provider 初始化
 2026-04-24 14:59:00 +08:00
fawney19
e0c928fbe8 feat(auth): /me 接口返回 has_password 字段，前端 Profile 类型同步调整 2026-04-24 14:52:49 +08:00
fawney19
f3c9835759 feat(pool): 引入 pro_first 调度预设、Pool 候选持久化跳过与诊断信息优化 
- 新增 pro_first 调度预设（Pro 优先），更新 plus_first 仅针对 Plus 计划，移除 free_team_first
- Pool 内部候选（pool_key_index 不为空）跳过 DB 持久化（available/skipped/unused 均适用）
- LRU 排序新增 catalog_lru_score 回退：runtime 无记录时使用 last_used_at_unix_secs
- 执行路径 miss 诊断消息细化为中文，按 reason 分类输出可读说明
- build_local_request_candidate_status_record 补充 extra_data 和 created_at_unix_ms 字段
- OpenAI CLI 计划构建流程补充候选评估进度跟踪与 terminal reason 设置
- 前端 PoolSchedulingDialog 增加 pro_first 预设展示，修复 LRU 默认预设检测逻辑
 2026-04-24 13:29:05 +08:00
fawney19
5d710d9d10 feat(rust): 支持 image 同步转流式 SSE、free_team_first 调度预设及账户错误自动重试 
- openai:image 同步响应桥接为流式 SSE（image_generation.completed / image_edit.completed 事件）
- image 请求解析与前置校验移除模型白名单，支持任意自定义模型名
- 调度器新增 free_team_first 预设（mode: both / free_only / team_only）
- pool config 解析重构：新增 POOL_ALLOWED_SCHEDULING_PRESETS 白名单，规范化 mode 字段
- 错误分类器新增账户/账单错误模式集，升级为 RetryUpstreamFailure 而非 StopSemanticClientError
- 修复 stream execution 中上游 headers 与输出 headers 混用导致 content-type 判断错误的问题
- codex image 工具始终写入 action 字段（generate/edit），仅 generate 操作填充默认 size/quality
- 前端：pool 节点组只展示实际执行过的候选节点，全部 skipped 时折叠为最后一个节点
- Redis 测试就绪检测从 TCP 连接改为 PING/PONG 协议验证
 2026-04-24 10:07:24 +08:00
AAEE86
a9d10163af fix(oauth): 修复 OAuth 刷新后 Token 有效期不更新问题 (#324) 
- 持久化刷新后的 `expires_at` 到 Provider Key SQL 更新链路
- 手动刷新接口优先返回本次刷新得到的过期时间
- 按当前 Key 字段重建 OAuth 状态快照，避免旧快照覆盖
- 前端刷新后防止旧列表数据回退覆盖新有效期
 2026-04-24 09:35:25 +08:00
Entropy.Xu
31e871fe1b fix(kiro): 修复流响应识别与号池配额刷新 Token 续期问题 (#322) 
* fix(kiro): 兼容 application/json 头下的 eventstream 流响应

- stream_pump: 对带 kiro:generateAssistantResponse envelope 的上游响应按流处理，不再仅依赖 text/event-stream 头判断\n- tests: 补充 application/json 头下仍识别为 Kiro stream 的回归测试

* fix(kiro): 修复号池配额刷新误判封禁
 2026-04-24 09:34:34 +08:00
fawney19
5148370253 fix(usage): 修复含失败信号的 pending 请求被误判为活跃的问题 
- pending/streaming 状态含 status_code >= 400 或 error_message 时视为 failed
- 候选状态派生改为 failed 优先于 stale pending/streaming
- 活跃请求轮询时保留失败信号字段的更新
- 图片流支持 partial_image 事件转发及从 response.completed 读取最终图片
 2026-04-24 03:07:11 +08:00
fawney19
581bc03d4e feat(ai-pipeline): 支持 reasoning signature 及媒体内容块的跨格式流式转换 
- 在 CanonicalStreamEvent 中新增 ReasoningSignature 和 ContentPart 变体，CanonicalContentPart 枚举覆盖图片/文件/音频
- Gemini 流解析器提取 thoughtSignature，并对 inlineData/fileData 等非文本 part 生成 ContentPart 事件
- Claude 流聚合支持 thinking_delta / signature_delta，输出 thinking block 携带 signature 字段
- Gemini 同步响应聚合重写，支持媒体 part 和 reasoning signature 的完整还原
- 跨格式矩阵（gemini↔claude↔openai）补全图片块双向转换及 reasoning signature 传递
- 请求转换层（to/from openai_chat）补全 Claude/Gemini 的 thinking、图片、工具调用字段映射
- 新增/扩展测试：inline image 双向重写、thinking signature 聚合、跨格式 sync product 媒体字段
 2026-04-24 02:18:34 +08:00
Entropy.Xu
0f94f92c37 fix(provider): 将rust分支的gemini cli端点行为对齐到python分支 (#321) 
* fix(provider): 对齐 Vertex/Gemini 上游发包与 Python master

- provider-transport: 为 custom+aiplatform 推断 Vertex API key 上下文并统一 URL 构建顺序，复用共享 request_url 构建最终上游地址
- ai-pipeline/gateway: Vertex Gemini 路径改为仅使用 URL query key，不再向上游附带 x-goog-api-key header；同步对齐 standard/admin/test-connection/runtime miss 摘要中的最终 URL
- gemini conversion: 按 Python master 输出 Gemini 请求体，补齐 system_instruction / generation_config / tool_config / function_declarations 形态，并移植 Gemini schema 清洗逻辑
- scheduler/executor: 将最终 upstream_url、mapped_model、key_name 写入候选 extra_data，运行时 miss 诊断优先展示真实展开后的上游 URL 便于服务器排障

* fix(provider): 修复 Vertex provider 测试与本地调度链路

* fix(provider): 对齐 Vertex 本地执行与 Rust CI
 2026-04-23 23:01:06 +08:00