English

If you have any questions or suggestions about the app, please contact us through our QQ group or GitHub Issues. We will respond as soon as possible.

Fixed

This release includes the log-folder fix missing from the initial v0.3.48 packages. If the error persists on v0.3.48, upgrade to this version and restart the app.
Fixed a stack overflow error (Maximum call stack size exceeded) when opening the log folder from Advanced settings.
Restored an opaque main window on Linux to address partial blank or transparent content after hovering, as reported on AMD / KDE and NVIDIA / Hyprland. Native translucent themes on Windows and macOS remain supported. #491

Reduced background UI work: silent startup, minimizing, and hiding to the tray now synchronize main and plugin WebView visibility and pause UI polling and countdowns. Restoring the window refreshes current data; the proxy core and usage collection continue running. #493
Usage updates now coalesce events, with fallback polling reduced from once per second to once every 10 seconds. Usage update events are no longer sent to the hidden UI, reducing background queries and rendering work.
Restored default browser background throttling. UI refresh scheduling uses native window visibility and minimized state, so losing focus alone does not pause updates.
Updated Tauri to stable version 2.12.4 and refreshed related runtime and build dependencies.

Added an Approval model setting to Codex configuration management. Choose Codex default or a specific model, save it to the managed catalog, and let newly discovered models inherit the setting. The selected model must be callable through the current proxy.
Added an experimental Remote compaction toggle for Codex, off by default, to enable server-side context compaction when supported by the upstream service.
Added an option to use the openai provider identity in Codex while continuing to route through CPA, with optional backup and synchronization of local session history when switching.

Quota cards now support model groups with separate usage windows and reset countdowns, with updated plan information and refresh button placement.
The app now restores its previous window position after restarting.
Plugin installation requests can now be cancelled using Cancel, Close, or Esc while installation is in progress.

Usage cost calculation now prefers automatic pricing for the actual original model before falling back to other prices, preventing manual prices from taking precedence unexpectedly.
Fixed reasoning token counts being squeezed or cut off when the cost column is enabled.
Fixed input controls squeezing setting labels and descriptions on configuration pages.
Fixed detection of the DeepSeek Harness desktop app and its bundled CLI on macOS.

Restart the relevant Codex CLI service or desktop app after changing these settings. The remote compaction toggle changes the provider name; if historical sessions disappear, resynchronize them in Session management.

本次重新发布包含日志文件夹修复的版本，补齐 v0.3.48 首批发布包中遗漏的修复。若使用 v0.3.48 仍出现该错误，请升级到本版本并重启应用。
修复在高级功能中打开日志文件夹时出现堆栈溢出错误（Maximum call stack size exceeded）的问题。
针对 Linux 上鼠标悬停后部分内容白屏或透明的重绘问题，恢复主窗口的不透明模式，覆盖 AMD / KDE 和 NVIDIA / Hyprland 等反馈场景。Windows 和 macOS 的原生半透明主题支持不受影响。#491

优化后台资源使用：后台启动、最小化或缩到托盘时，同步主界面与插件 WebView 的可见性，暂停界面轮询和倒计时；恢复窗口后自动刷新最新数据。代理核心与用量采集继续运行。#493
用量页改为合并事件刷新，兜底轮询由每秒一次调整为每 10 秒一次；窗口隐藏时停止向界面发送用量更新事件，减少后台查询与渲染开销。
恢复浏览器默认后台节流策略。界面刷新根据原生窗口是否显示及是否最小化判断，不因单纯失去焦点而暂停。
将 Tauri 升级至 2.12.4 稳定版，并同步更新相关运行时与构建依赖。

Codex「配置管理」新增自动审批模型设置，可选择 Codex 默认或指定模型；保存后同步到模型目录，新发现的模型自动继承设置。所选模型需能通过当前代理调用。
Codex 新增**远程压缩（实验性）**开关，默认关闭，可在上游支持时启用服务端上下文压缩。
Codex 支持切换为 openai 提供商标识，仍通过 CPA 代理；切换时可选择备份并同步本机历史会话。