CherryHQ/cherry-studio v2.1.2

on GitHub

Cherry Studio 2.1.2 - Stability and User Experience Improvements

✨ New Features

• [Model] Add oMLX as a built-in local provider, with its base URL preset and no API key required.
• [Model] Add Moonshot AI global endpoint for users of the international Moonshot console.
• [Settings] Renamed Recycle Bin to Archive and promoted it to a standalone settings page with improved layout and navigation.
• [MCP] The built-in mcp-auto-install server now keeps its Registry cache and config writes inside Cherry's config directory.
• [Sign-in] CherryIN sign-in now uses a local HTTP callback instead of the system custom-protocol handler.
• [Agents] Return to the original chat from a link at the fork boundary.
• [WebView] The browser pane's “Annotate page” picker now uses a consistent icon and follows your primary color.
• [Painting] Honor model-declared maxInputImages for generate_image references.

🐛 Bug Fixes

• [Sidebar] Clicking an app icon in the sidebar replaces the active tab again instead of jumping to another tab.
• [Sidebar] All resource shortcuts can now be removed via context menus.
• [Settings] Fixed settings, authorization, billing, and feedback links opening in the internal browser instead of the system browser.
• [Settings] English search terms now find the V2 serif-font setting.
• [Agents] Fixed timeouts when forking long DSH conversations.
• [Models] Fix outdated DeepSeek model listings and refresh the bundled catalog and pricing.
• [Models] Keep valid LM Studio models when one catalog entry is malformed.
• [Tools] Fixed skill installation, Skill tool dependency check, and Windows binary management failing when command lookup times out or errors.
• [Input] AskUserQuestion answers now wrap and grow, and keep selected options when a custom note is typed.
• [AI] Fixed Gemini requests failing with HTTP 400 when a tool's parameters declare numeric or boolean enum values.
• [AI] Fix Gemini requests failing with HTTP 400 when the built-in browser tools are enabled.
• [Markdown] Fixed KaTeX-stretched symbols (e.g., square roots) disappearing from rendered formulas.
• [Markdown] Keep the UI responsive during long agent streams by avoiding full markdown re-parse on each tick.
• [Markdown] Fixed emphasis formatting being lost after saving notes with adjacent punctuation.
• [Windows] Fixed agent-session trace history never being saved.
• [Chat] Fixed AI replies involving tool or MCP calls having no Edit option; their text is now editable.
• [Privacy] Fixed analytics events still being sent after disabling data collection.
• [macOS] Warn Apple Silicon users running the Intel build and link to the correct edition's download page.

Cherry Studio 2.1.2 - 稳定性与用户体验改进

✨ 新功能

• [模型] 新增 oMLX 内置本地服务商，预设 base URL，无需 API 密钥。
• [模型] 新增 Moonshot AI 全球端点，方便使用国际版 Moonshot 控制台的用户连接。
• [设置] 将回收站重命名为归档，并提升为独立设置页面，改进布局和导航。
• [MCP] 内置 mcp-auto-install 服务器将 Registry 缓存和配置写入保持在 Cherry 的配置目录。
• [登录] CherryIN 登录使用本地 HTTP 回调，不再依赖系统自定义协议处理程序。
• [智能体] 从分叉边界链接返回原始对话。
• [WebView] 浏览器窗格的”标注页面”选择器现在使用一致的图标并跟随主题色。
• [绘图] 生成图像时遵守模型声明的最大输入图片数限制。

🐛 问题修复

• [侧边栏] 点击侧边栏的应用图标再次替换当前标签页，而非跳转到同一应用的其他标签页。
• [侧边栏] 所有资源快捷方式现在可以通过右键菜单移除。
• [设置] 修复设置、授权、账单和反馈链接在内置浏览器打开而非系统浏览器的问题。
• [设置] 英文搜索词现在可以找到 V2 衬线字体设置。
• [智能体] 修复分叉长时间 DSH 对话时的超时问题。
• [模型] 修复 DeepSeek 模型列表过时的问题，刷新内置目录和定价。
• [模型] 当单个目录条目格式错误时保留有效的 LM Studio 模型。
• [工具] 修复技能安装、Skill 工具依赖检查和 Windows 二进制管理在命令查找超时或出错时失败的问题。
• [输入] AskUserQuestion 回答现在可以换行和增长，并在输入自定义说明时保留已选选项。
• [AI] 修复工具参数声明数字或布尔枚举值时 Gemini 请求返回 400 错误的问题。
• [AI] 修复启用内置浏览器工具时 Gemini 请求返回 400 错误的问题。
• [Markdown] 修复 KaTeX 拉伸符号（如平方根）在渲染公式中消失的问题。
• [Markdown] 长 agent 流式输出时避免已稳定段落每 tick 全量重解析，降低界面卡顿。
• [Markdown] 修复保存笔记时相邻标点导致强调格式丢失的问题。
• [Windows] 修复智能体会话跟踪历史从未保存的问题。
• [对话] 修复涉及工具或 MCP 调用的 AI 回复没有编辑选项的问题，现在可以编辑文本。
• [隐私] 修复禁用数据收集后仍发送分析事件的问题。
• [macOS] 提醒在 Apple Silicon 上运行 Intel 版本的用户，并链接到正确版本的下载页面。