v2.0.12

Downloads / 下载 (v2.0.12)

Platform	Architecture	Global	China Edition
Windows	x64	Installer · Portable	Installer · Portable
Windows	arm64	Installer · Portable	Installer · Portable
macOS	x64	DMG · ZIP	DMG · ZIP
macOS	arm64	DMG · ZIP	DMG · ZIP
Linux	x64	AppImage · DEB · RPM	AppImage · DEB · RPM
Linux	arm64	AppImage · DEB · RPM	AppImage · DEB · RPM

Release Notes / 发布说明

English

Cherry Studio 2.0.12 - Features, Fixes, and Improvements

✨ New Features

• [Settings] Settings gains a sidebar search: find any setting row across all sections (title/description/aliases, Chinese pinyin included), jump to it with scroll + highlight, full keyboard navigation.
• [Settings] Settings search now covers all sections with static settings rows, including MCP tabs.
• [Translation] The translate model now has its own temperature, top-p and reasoning effort settings, editable from the translation settings panel and the translate page toolbar.
• [Markdown] Markdown previews outside chat now use the same rich renderer as chat messages. In agent session messages and rendered artifact Markdown previews, workspace links open files in the artifact pane and directories in the system file manager; relative links resolve from the agent workspace root.

🐛 Bug Fixes

• [Cherry Cloud] Cherry Cloud models now honor server-declared capabilities, support configured OpenAI Chat Completions routes, and remain compatible with older model-list responses. Message token details now also show time to first token without requiring the expanded runtime breakdown.
• [Chat] Preserve partial chat responses and show a clear, retryable error when a provider stream ends without a finish reason.
• [Chat] Fix middle-click (autoscroll) scrolling in the chat message list so it no longer vibrates or jumps.
• [Chat] Sent image attachments now appear as compact filename chips with real image thumbnails, including after reopening a conversation.
• [Model Health] Model health check results no longer linger after leaving the provider page.
• [Filesystem MCP] Allow grep searches to use regular expression syntax supported by ripgrep.
• [Knowledge Base] Empty knowledge bases now appear in the knowledge list.
• [Chat Export] Prevent fraction denominators from being clipped when exporting conversations as images.
• [OpenClaw] Prevent the OpenClaw gateway from failing to start when Cherry Studio uses a socks5 proxy.
• [Web Search] Web fetch no longer sends localhost, LAN, or locally-unresolvable URLs to Jina Reader when the built-in fetch fails, regardless of the “Allow fetching local network addresses” setting.
• [Topics] Pinned topics and agent sessions now keep their text aligned with unpinned items in owner-grouped lists.
• [Agent] The task progress capsule now always tracks the main agent's task plan; todo lists written by spawned subagents no longer replace it.
• [Agent] Fixed the permission-mode popup showing two highlighted rows at once, and tooltips no longer pop up on their own when the panel opens.
• [PDF Preview] Preserve colors in colored PDF documents.
• [Models] Fix reasoning mode not being available for Xunfei MaaS (讯飞星辰) custom models — their model IDs are now recognized and expose the reasoning toggle/effort controls.
• [UI] Improved Skills detail scrolling and Sidebar avatar and title-bar alignment.

💄 Improvements

• [Input Fields] Number fields across settings, model options, MCP servers, assistants and painting parameters no longer show the browser's native spinner buttons, and no longer change value when the mouse wheel scrolls over a focused field. They step with the arrow keys, announce their allowed range to screen readers, and the API gateway port now tells you when a value is out of range instead of quietly applying a different one.

⚡ Performance

• [Paintings] Improved painting history refresh performance by reusing unchanged hydrated records.

简体中文

Cherry Studio 2.0.12 - 新功能、修复与改进

✨ 新功能

• [设置] 设置侧边栏新增搜索功能：可搜索所有设置项（标题/描述/别名，支持中文拼音），点击跳转并高亮显示，支持完整的键盘导航。
• [设置] 设置搜索现覆盖所有包含静态设置行的部分，包括 MCP 标签页。
• [翻译] 翻译模型现在可以单独设置温度、Top-P 和推理强度，在翻译设置面板和翻译页工具栏均可调整。
• [Markdown] 聊天外的 Markdown 预览现在使用与聊天消息相同的富渲染器。在智能体会话和渲染的产物 Markdown 预览中，工作区链接在产物面板中打开文件，目录在系统文件管理器中打开；相对链接从智能体工作区根目录解析。

🐛 问题修复

• [樱桃云] 樱桃云模型现在遵循服务器声明的功能，支持配置的 OpenAI Chat Completions 路由，并与旧版模型列表响应保持兼容。消息令牌详情现在也显示首个令牌时间，无需展开运行时明细。
• [对话] 当服务商流在没有结束原因的情况下终止时，保留部分对话响应并显示清晰的可重试错误。
• [对话] 修复聊天消息列表中的中键自动滚动不再抖动或跳跃。
• [对话] 发送的图片附件现在显示为紧凑的文件名芯片，带有真实的图片缩略图，包括重新打开对话后。
• [模型健康] 离开服务商页面后，模型健康检查结果不再残留。
• [文件系统 MCP] 允许 grep 搜索使用 ripgrep 支持的正则表达式语法。
• [知识库] 空知识库现在显示在知识库列表中。
• [对话导出] 防止导出对话图片时分数分母被裁剪。
• [OpenClaw] 防止 Cherry Studio 使用 socks5 代理时 OpenClaw 网关启动失败。
• [联网搜索] 当内置获取失败时，Web 获取不再将 localhost、局域网或本地无法解析的 URL 发送到 Jina Reader，无论”允许获取本地网络地址”设置如何。
• [话题] 固定的话题和智能体会话现在在按所有者分组的列表中与未固定项目保持文本对齐。
• [智能体] 任务进度胶囊现在始终跟踪主智能体的任务计划；子智能体编写的待办列表不再替换它。
• [智能体] 修复权限模式弹出窗口同时显示两个高亮行的问题，工具提示不再在面板打开时自动弹出。
• [PDF 预览] 在彩色 PDF 文档中保留颜色。
• [模型] 修复讯飞星辰自定义模型无法使用推理模式的问题 —— 其模型 ID 现在被识别并显示推理开关/强度控制。
• [界面] 改进技能详情滚动和侧边栏头像及标题栏对齐。

💄 改进

• [输入字段] 设置、模型选项、MCP 服务器、助手和绘画参数中的数字字段不再显示浏览器的原生微调按钮，鼠标滚轮滚动聚焦字段时不再改变值。它们使用方向键步进，向屏幕阅读器宣布允许的范围，API 网关端口现在会告诉您值何时超出范围，而不是悄悄应用不同的值。

⚡ 性能优化

• [绘画] 通过重用未更改的水合记录改进绘画历史刷新性能。

---

What's Changed

• fix(chat-messages): stop reading-mode freeze from fighting middle-click autoscroll by @404-Page-Found in #19225
• fix(pdf-preview): preserve PDF document colors by @ActiveSky in #19918
• chore(release): sync v2.0.11 metadata by @kangfenmao in #19925
• fix(provider-registry): detect Xunfei MaaS reasoning models by @404-Page-Found in #17811
• chore(selection): upgrade selection-hook to 2.1.1 by @0xfullex in #19928
• fix(quick-panel): open single-select submenus on the current selection by @ousugo in #19697
• fix(agents): keep the progress capsule on the main agent's plan by @ousugo in #19684
• feat(settings): sidebar search for settings rows (core engine + 5 sections) by @GeorgeDong32 in #19345
• feat(settings): index remaining settings sections for search by @GeorgeDong32 in #19346
• feat(settings): carry focus intent in the URL by @GeorgeDong32 in #19752
• fix(message-rendering): render sent image attachments by @SiinXu in #19900
• perf(paintings): cache unchanged history hydration by @userInner in #18720
• fix: refine skill details and navigation chrome by @kangfenmao in #19932
• fix(resource-list): keep pinned item text aligned by @AtomsH4 in #19934
• feat(markdown): unify previews on a shared host-injected renderer by @DeJeune in #17135
• fix(web-search): keep private targets out of the Jina fetch fallback by @kovsu in #19920
• refactor(ui): consolidate numeric entry into an InputNumber primitive by @EurFelux in #19129
• fix(openclaw): strip proxy env from gateway spawn by @SiinXu in #18911
• perf(chat): load citations panel on first open by @SiinXu in #19869
• fix(chat): classify streams missing finish reason by @SiinXu in #19870
• fix(i18n): add missing Turkish finish-reason error string by @DeJeune in #19937
• fix(chat-export): preserve formula bounds in exported images by @hburaktasyurek in #19895
• fix: allow documentCount 0 for empty knowledge bases by @gokay-ai in #19874
• perf(model-health): isolate row status updates by @SiinXu in #19876
• perf(mini-app): isolate card subscriptions by @SiinXu in #19884
• feat(translate): add temperature, top-p and reasoning effort by @EurFelux in #19700
• docs(gh-pr-review): distinguish open product decisions by @DeJeune in #19657
• fix(local-model): sweep stale downloads and settle installers on shutdown by @kovsu in #19911
• fix(preference): repair invalid JSON during startup by @SiinXu in #19641
• fix(file): compare inodes as bigint in isSameFile by @kovsu in #19961
• fix(filesystem-mcp): defer regex validation to fallback by @kovsu in #19960
• fix(model-health): clear cached check results on unmount by @kovsu in #19959
• revert(preference): drop invalid-JSON startup repair seeder by @DeJeune in #19969
• fix(cherry-cloud): support model capabilities and OpenAI chat by @zhibisora in #19941

New Contributors

• @gokay-ai made their first contribution in #19874

Full Changelog: v2.0.11...v2.0.12