Repository navigation
Codex Tools v2.10.0
English
- Restore macOS account switching for current Codex app bundles, use the managed file credential store, stop the verified desktop before replacing credentials, preserve shared settings, and verify the relaunched CODEX_HOME. Disabling launch-after-switch leaves the desktop closed. Merge #211 and fix #209.
- Support GPT-6 Sol and Luna across the proxy model catalog, aliases, permissions and Responses Lite transports with the required Codex client identity. Add standard token cost estimates and rebuild cached costs. Merge #207 and fix #208. Existing API-key model allowlists remain unchanged.
- Use an HTTP-only custom provider for Relay profiles to avoid first-message WebSocket handshake delays (#202). Preserve the legacy openai_base_url route for existing sessions and remove Relay routing when returning to ChatGPT.
- Add editing of saved API accounts: URL, model, label and optional key replacement (#201). A blank key preserves the stored key. Switch to the account again to apply edited desktop settings.
- Keep account rows in stable import order across quota refreshes and account switches (#203), and make quota icons follow the selected five-hour or weekly window (#204).
- Preserve Chat Completions stream termination and upstream errors for response.done, incomplete, failed, cancelled and truncated streams (#205). Client validation errors do not trigger retries across every account.
- Bundle every standalone-proxy module through a shared installer/deployment manifest and compile that packaged source in CI, fixing missing-module failures during remote deployment.
- Parse complete structured warm-up events, stop treating unused zero-percent windows as active, and distinguish a completed request from quota-confirmed activation (#199). Warm-up stays opt-in, bounded and protected by persisted cooldowns.
中文
- 修复新版 macOS Codex 账号切换:识别当前应用包内的 CLI,使用托管文件凭据,替换授权前关闭经过验证的桌面进程,保留共享设置,并验证重启进程的 CODEX_HOME。关闭“切换后启动”时,切换后桌面保持关闭。合并 #211,修复 #209。
- 反代支持 GPT-6 Sol 和 Luna,覆盖模型目录、别名、权限和 Responses Lite 协议,使用所需客户端标识;增加标准 Token 费用估算并重建旧成本缓存。合并 #207,修复 #208;已有 API Key 的模型白名单不自动扩大。
- Relay profile 改用仅 HTTP 的自定义 provider,避免新会话首条消息等待 WebSocket 握手失败(#202);保留旧会话的 openai_base_url 路由,切回 ChatGPT 时清理 Relay 路由。
- 支持编辑已保存 API 账号的地址、模型、名称及 Key(#201);Key 留空保留原值,保存后再次切换到该账号即可应用桌面配置。
- 账号列表按导入顺序稳定展示,刷新额度和切换账号不再重排(#203);额度图标跟随所选 5h 或 1 周窗口(#204)。
- 补全 Chat Completions 对 response.done、未完成、失败、取消和截断流的处理,返回结束原因或原始上游错误(#205);普通请求参数错误不再遍历所有账号重试。
- 安装资源与远程部署共用完整源码清单,并在 CI 中直接编译打包后的独立代理,修复远程构建缺少子模块的问题。
- 预热改为解析完整结构化完成事件,不把未使用的 0% 窗口误判为已激活,并区分“请求完成”和“额度已确认激活”(#199);保留默认关闭、请求上限及持久化冷却。
Distribution / 发布范围
This release publishes the macOS Apple Silicon, macOS Intel and Windows desktop installers. The npm/CLI update is deferred.
本次发布 macOS Apple Silicon、macOS Intel 和 Windows 桌面安装包;npm/CLI 更新暂缓。