Repository navigation
Releases: icebear0828/codex-proxy
Releases · icebear0828/codex-proxy
Release list
v2.1.21-beta.852249c
✨ 本次更新
- 桌面端（Windows 与 Linux）新增应用内更新功能，支持自动检测新版本并在后台静默下载。
- 实时进度显示：Dashboard 顶栏现在会直接显示更新状态及下载百分比。
- 无缝安装体验：下载完成后，点击顶栏提示即可一键重启并安装更新，无需手动前往 GitHub 下载安装包。
- 优化交互逻辑：移除了旧版的更新弹窗，并隐藏了设置页面中冗余的自动下载开关。
What's New
- Added in-app update support for Windows and Linux, allowing the app to automatically detect and download new versions in the background.
- Real-time update tracking: The Dashboard top bar now displays the current update status and download percentage.
- Seamless installation: Once a download is complete, you can restart and install the update directly from the top bar without visiting the GitHub Release page.
- Improved user experience by removing legacy update dialogs and hiding redundant update settings in desktop mode.
完整提交记录 / Full commit list
- feat(electron): add in-app desktop updates (#878)
v2.1.20
✨ 本次更新
- 优化了响应截断的处理逻辑，并修复了流式请求失败时错误返回 HTTP 200 的问题，确保客户端能正确识别错误并触发重试。
- 显著提升了不同模型供应商（如 OpenAI、Anthropic、Gemini）之间的兼容性，修复了工具调用、图片上传及系统消息处理中的多项问题。
- 增强了会话稳定性，修复了连接中断后可能导致会话永久卡死在 502 错误状态的问题。
- 修复了界面显示问题，现在能准确统计并显示 Antigravity 账号的状态与数量。
- 优化了代理转发逻辑，自动绕过本地回环地址（localhost），解决了开启代理时无法连接本地模型的问题。
- 修复了图片生成接口（/v1/images/generations）被错误屏蔽的问题，恢复了账号模式下的生图功能。
- 解决了 Docker 部署中的多项问题，包括特定系统下的启动崩溃、配置卷文件无法自动更新以及计费表加载失败等。
What's New
- Improved handling of truncated responses and fixed an issue where streaming errors were incorrectly reported as successful (HTTP 200), allowing clients to correctly trigger retries.
- Enhanced compatibility between different AI providers (OpenAI, Anthropic, Gemini), fixing multiple issues related to tool calling, image uploads, and system message handling.
- Improved session resilience by fixing a bug where interrupted connections could cause a session to get stuck in a persistent 502 error state.
- Fixed UI bugs to ensure Antigravity accounts are correctly counted and displayed in the account list and statistics.
- Optimized proxy routing to automatically bypass local addresses (localhost), resolving connection failures to local models when a system proxy is active.
- Restored the image generation API (/v1/images/generations) functionality which was previously being incorrectly blocked.
- Resolved several Docker deployment issues, including startup crashes on certain systems, missing configuration files after updates, and pricing table loading failures.
完整提交记录 / Full commit list
- fix(translation): report truncation from response.incomplete as max_tokens/refusal
v2.1.20-beta.0543eaa
✨ 本次更新
- 优化了响应截断的处理逻辑，当 AI 回复因达到长度限制而中断时，现在能更准确地向客户端反馈“达到最大 Token”或“被拒绝”状态。
- 修复了在开启系统代理或断网时，发往本地模型（如 Ollama/localhost）的请求被错误拦截导致连接失败的问题。
- 修复了账户列表无法正确统计和显示 Antigravity 账户的问题，避免在仅有此类账户时错误显示“暂无已连接账户”。
- 修复了流式请求失败时错误地返回“成功”状态码（200）的问题，确保客户端（如 Claude Code）能正确识别错误并触发重试。
- 显著提升了 Anthropic、Gemini 与 OpenAI 协议之间的转换兼容性，修复了图片输入、工具调用（Tool Use）及结构化输出在跨平台转发时的多项异常。
- 修复了在账号登录模式下，使用图片生成功能（DALL-E）可能返回 400 错误的问题。
- 修复了 Docker 部署中新版本配置文件（如价格表）无法自动同步到旧配置卷的问题，解决了 Token 计费估算显示为 0 的故障。
- 修复了长连接会话在网络波动导致 WebSocket 断开后，后续请求可能陷入永久 502 错误的问题，现在支持自动尝试恢复会话。
What's New
- Improved handling of truncated responses; when an AI response is cut off due to length limits, it now correctly reports "max_tokens" or "refusal" status to the client.
- Fixed an issue where connections to local models (e.g., Ollama on localhost) would fail when a system proxy was active or the network was disconnected.
- Fixed the account list UI to correctly count and display Antigravity accounts, preventing the "No accounts connected" message when only these accounts are present.
- Fixed a bug where failed streaming requests incorrectly returned a "Success" (200) status, ensuring clients like Claude Code can properly trigger retries.
- Enhanced compatibility for protocol translation between Anthropic, Gemini, and OpenAI, fixing issues with image inputs, tool calls, and structured outputs.
- Fixed an issue where the image generation (DALL-E) endpoint might return a 400 error when using the app in account login mode.
- Fixed a Docker deployment issue where new default configuration files (like pricing tables) were not updated in existing volumes, which caused token cost estimates to stay at zero.
- Fixed an issue where long-running sessions could get stuck in a permanent 502 error state after a WebSocket disconnection; the app now attempts to recover the session automatically.
完整提交记录 / Full commit list
- fix(translation): report truncation from response.incomplete as max_tokens/refusal
v2.1.19
✨ 本次更新
- 修复了 GitHub Release 页面缺少 codex-proxy-npm.zip 离线包的问题，现在可以正常通过链接下载。
- 优化了本地回环地址的处理，修复了在开启系统代理或断网环境下无法连接本地模型（如 Ollama、LM Studio）的问题。
- 修复了 Antigravity 账户在首页列表不显示或统计不准的问题，现在能正确显示所有已连接的账户状态。
- 改进了日志窗口的显示，出口模型列现在显示真实的供应商模型 ID，不再带有内部处理后缀（如 -high/-fast）。
- 大幅提升了跨供应商（OpenAI、Gemini、Anthropic）的兼容性，修复了工具调用、图片输入和结构化输出在转换时可能导致的报错。
- 修复了 Docker 用户更新镜像后无法自动获取新增默认配置文件（如价格表）的问题，现在会自动补齐缺失的配置。
- 修复了流式请求失败时错误地返回“成功”状态码的问题，确保客户端（如 Claude Code）能正确识别错误并触发重试。
- 修复了在账号登录模式下无法使用图片生成功能（/v1/images/generations）的问题。
What's New
- Fixed the issue where the codex-proxy-npm.zip offline bundle was missing from GitHub Releases; it is now available for download.
- Improved handling of local loopback addresses, fixing connection failures to local models (e.g., Ollama, LM Studio) when a system proxy is active or when offline.
- Fixed Antigravity accounts not appearing in the account list or being excluded from the total count on the home page.
- The "Egress Model" column in logs now displays the actual upstream model ID, removing confusing internal processing suffixes.
- Enhanced compatibility across providers (OpenAI, Gemini, Anthropic), resolving errors related to tool calls, image inputs, and structured outputs during translation.
- Fixed an issue where Docker deployments wouldn't receive new default configuration files (like pricing tables) when upgrading the image.
- Fixed a bug where failed streaming requests incorrectly returned an HTTP 200 status, allowing clients to properly detect errors and retry.
- Fixed the image generation endpoint (/v1/images/generations) being inaccessible when using account login mode.
完整提交记录 / Full commit list
- fix(npm): attach the offline bundle on tag dispatches
v2.1.19-beta.f824033
✨ 本次更新
- 修复了 GitHub Release 页面缺少 codex-proxy-npm.zip 离线包的问题。
- 修复了在开启系统代理或断网状态下，连接本地模型（如 Ollama/localhost）失败的问题。
- 修复了仅配置 Antigravity 账号时，首页错误显示“暂无已连接账户”及统计不准确的问题。
- 大幅提升了不同模型供应商（Anthropic, Gemini, OpenAI）之间的兼容性，修复了工具调用、图片输入及系统消息格式不匹配导致的请求失败。
- 修复了长对话中因连接临时中断导致的持续 502 错误，现在支持自动尝试恢复会话。
- 修复了流式请求失败时错误返回 HTTP 200 的问题，确保 Claude Code 等客户端在遇到限流或错误时能正确触发重试。
- 修复了 Docker 部署时无法自动更新新增配置文件（如价格表）的问题，并优化了计费统计的准确性。
- 修复了在账号模式下图片生成接口（/v1/images/generations）被错误屏蔽的问题。
What's New
- Fixed the issue where the codex-proxy-npm.zip offline bundle was missing from GitHub Releases.
- Fixed connection failures to local models (e.g., Ollama/localhost) when a system proxy or VPN is active.
- Fixed the account list incorrectly showing "No accounts" and inaccurate statistics when only Antigravity accounts are used.
- Significantly improved compatibility between different AI providers (Anthropic, Gemini, OpenAI), fixing issues with tool calls, image inputs, and system message formatting.
- Fixed persistent 502 errors in long-running sessions after a temporary connection drop by implementing automatic session recovery.
- Fixed an issue where streaming errors were incorrectly returned as HTTP 200, allowing clients like Claude Code to properly handle retries and rate limits.
- Fixed Docker deployments failing to receive new default configuration files (such as price tables) during updates.
- Fixed the image generation API (/v1/images/generations) being unavailable or returning errors in account login mode.
完整提交记录 / Full commit list
- fix(npm): attach the offline bundle on tag dispatches
v2.1.18
✨ 本次更新
- API Key 增加自定义代理路由功能，支持为每个 Key 单独指定代理节点、直连或自动轮询，不再强制全局统一。
- 修复了 API Key 代理设置可能失效的问题，确保未指定代理时能正确回退到全局默认设置。
- 新增 Antigravity OAuth 账号管理，支持通过 Google 登录接入 Claude 和 Gemini 模型，并可直观查看各账号的用量额度与套餐详情。
- 提升弱网环境下的连接稳定性，支持请求失败自动重试及上游流超时自动断开，避免请求卡死或资源浪费。
- 增强模型列表元数据，现在可以在支持的客户端中看到更详细的模型描述、上下文窗口大小及功能特性。
- 账号模式现已支持图片编辑接口 (/v1/images/edits)，兼容 OpenAI 标准的文件上传格式。
- 新增 NPM 分发渠道，支持通过命令行安装和运行，方便在服务器或无界面环境中使用。
What's New
- Added custom proxy routing for API Keys, allowing you to assign specific proxy nodes, direct connections, or auto-polling to individual keys.
- Fixed an issue where API Key proxy assignments could fail, ensuring they correctly fall back to global default settings.
- Introduced Antigravity OAuth account management, enabling Google login for Claude and Gemini models with real-time usage and quota tracking.
- Improved network resilience with automatic retries for failed connections and a watchdog timer to cancel hung requests in poor network conditions.
- Enhanced model metadata, providing detailed information like context window size and model descriptions to your coding clients.
- Added support for the Image Edits API (/v1/images/edits) in account mode, compatible with standard OpenAI file upload formats.
- Launched a new NPM distribution channel, allowing users to install and run the proxy as a lightweight CLI service.
完整提交记录 / Full commit list
- fix(proxy): normalize null proxyId to global default and validate proxy assignments
- feat(proxy): support custom proxy routing for api keys and bypass loopback in fetch dispatcher
v2.1.18-beta.d62e172
✨ 本次更新
- 新增 Antigravity OAuth 账号支持：现在可以通过 Google OAuth 快速接入账号，并直接在应用内管理账号启停、查看 Claude 与 Gemini 等模型的实时用量与额度。
- 支持 API Key 自定义代理路由：您可以为不同的第三方 Key 独立设置代理策略，支持选择全局默认、直连、自动轮询或指定的代理节点。
- 修复了账号列表中 Antigravity 账号数量统计不准确的问题，确保账号概览信息实时同步。
- 优化了代理分配的校验逻辑，修复了部分情况下代理设置无法正确回退到全局默认值的问题，提升了网络请求的稳定性。
What's New
- Added Antigravity OAuth support: You can now connect accounts via Google OAuth and manage them directly from the dashboard, including toggling accounts and tracking real-time usage for Claude and Gemini models.
- Custom proxy routing for API Keys: You can now assign specific proxy strategies to individual API keys, choosing between global defaults, direct connection, automatic polling, or specific proxy nodes.
- Fixed an issue where Antigravity accounts were not being counted correctly in the account list summary.
- Improved proxy assignment validation and normalization, ensuring requests correctly fallback to global defaults and routing remains stable when using custom configurations.
完整提交记录 / Full commit list
- fix(proxy): normalize null proxyId to global default and validate proxy assignments
- feat(proxy): support custom proxy routing for api keys and bypass loopback in fetch dispatcher
- fix(web): count antigravity accounts in account list (#870)
- feat(antigravity): add OAuth account management and usage tracking (#868)
v2.1.17
✨ 本次更新
- Antigravity OAuth 账号现已整合至首页账号管理区，支持用量追踪，并修复了账号总数统计不包含此类账号的问题。
- 新增账号保活调度器，通过自动化的心跳机制维持账号在线状态，减少连接中断。
- 优化日志与统计：出口日志现可准确显示上游实际模型 ID；字数过少的回复（少于 100 token）将不再计入平均速率统计，以避免数据失真。
- 提升操作效率：新增仪表盘键盘快捷键，并将系统托盘图标改为单击即可打开主窗口。
- 改进交互体验：「添加账号」时新增确认对话框以防止误开浏览器，同时为 API 密钥管理页面增加了模型筛选与手动刷新功能。
What's New
- Antigravity OAuth accounts are now fully integrated into the main account management area with usage tracking and accurate counting in the dashboard.
- Added an account keepalive scheduler to automatically maintain active sessions and improve connection stability.
- Improved logs and metrics: Egress logs now show the actual upstream model ID, and short responses (under 100 tokens) are excluded from speed metrics to ensure more accurate dashboard averages.
- Enhanced efficiency: Added keyboard shortcuts for the dashboard and updated the system tray icon to open the main window with a single click.
- Refined UI interactions: Added a confirmation dialog when adding accounts to prevent accidental browser popups, and introduced model filtering and manual refresh for API key management.
完整提交记录 / Full commit list
- fix(web): count antigravity accounts in account list (#870)
- feat(antigravity): add OAuth account management and usage tracking (#868)
- feat(logs): skip speed metric for outputs under 100 tokens (#869)
- feat(web): add dashboard keyboard shortcuts
- feat(keepalive): add account keepalive scheduler and admin routes (#867)
- fix(logs): report the resolved upstream model in egress audit rows (#864)
v2.1.17-beta.1604624
✨ 本次更新
- 托盘图标现支持单击打开主窗口，无需双击，操作更便捷。
- 增强了对新官方模型的兼容性，即使本地尚未收录（如未来的 GPT 新型号），也能直接透传请求至上游。
- 优化了日志统计逻辑，短回复不再计入平均速度以避免数据失真，并修复了日志中“出口模型”名称显示不准确的问题。
- 改进了“添加账号”流程，改为弹出对话框供用户选择打开链接或复制代码，并统一了全站工具栏按钮风格。
- API 密钥管理新增模型列表手动刷新与筛选功能，缓存时间缩短，方便及时获取供应商最新的模型列表。
- 修复了会话在连接断开后可能陷入永久 502 错误的问题，并大幅提升了 Anthropic、Gemini 与 OpenAI 之间工具调用和图片处理的兼容性。
- 修复了 Docker 或 Lite 版本更新后，新增的默认配置文件（如计价表）无法自动同步到已有配置目录的问题。
What's New
- The system tray icon now opens the main window with a single click for quicker access.
- Improved compatibility with new official models; requests for models not yet in the local catalog (e.g., future GPT versions) will now be passed through directly to the upstream.
- Optimized log metrics: short responses no longer skew average speed statistics, and the "Egress Model" column now accurately reflects the actual upstream model name.
- Improved the "Add Account" flow with a new dialog for copying or opening authorization URLs, and unified the toolbar button styles across all pages.
- Added manual refresh and filtering for model lists in API Key management, with shorter cache times to ensure you see the latest models from providers.
- Fixed an issue where sessions could get stuck in permanent 502 errors after a connection drop, and significantly improved compatibility for tool calls and image handling across Anthropic, Gemini, and OpenAI.
- Fixed an issue in Docker and Lite versions where new default configuration files (such as pricing tables) were not automatically added to existing data volumes after an update.
完整提交记录 / Full commit list
- feat(logs): skip speed metric for outputs under 100 tokens (#869)
v2.1.17-beta.87a8f10
✨ 本次更新
- 现在只需单击系统托盘图标即可快速打开主窗口，无需再双击。
- 新增账户活跃保持（Keepalive）定时任务，支持按固定时刻或时间间隔自动刷新配额，确保账号状态稳定。
- 控制台新增键盘快捷键支持，提升日常操作效率。
- 优化日志显示，出口模型列现在会准确显示请求到上游的实际模型 ID，不再带有本地处理后缀（如 -high）。
- 修复了 Gemini 和 Antigravity 在处理复杂工具调用（Function Calling）时的稳定性问题，并改进了错误诊断。
- 改进了“添加账户”流程，现在会先弹出包含授权链接的对话框，方便用户选择复制链接或手动打开浏览器。
What's New
- You can now open the main window with a single click on the system tray icon, eliminating the need for double-clicking.
- Added a new account keepalive scheduler that supports refreshing quotas at fixed times or intervals to maintain account stability.
- Introduced keyboard shortcuts for the dashboard to improve navigation and efficiency.
- Improved log accuracy: the "Egress Model" column now displays the actual upstream model ID instead of local internal suffixes (like -high).
- Fixed stability issues with Gemini and Antigravity tool calls (Function Calling) and improved error diagnostics for these requests.
- Enhanced the "Add Account" workflow by introducing a confirmation dialog with "Copy" and "Open Link" options for authorization URLs.
完整提交记录 / Full commit list
- feat(web): add dashboard keyboard shortcuts
- feat(keepalive): add account keepalive scheduler and admin routes (#867)
- fix(logs): report the resolved upstream model in egress audit rows (#864)
- feat(electron): open main window on single tray click (#862)
- fix(antigravity): stabilize Gemini tool requests and error diagnostics
