github CherryHQ/cherry-studio v2.0.13

4 hours ago

Downloads / 下载 (v2.0.13)

Platform Architecture Global China Edition
Windows x64 Installer · Portable Installer · Portable
Windows arm64 Installer · Portable Installer · Portable
macOS Intel DMG · ZIP DMG · ZIP
macOS Apple M Series 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 2.0.13 - New Features, Fixes, and Improvements

✨ New Features

  • [Models] Add GPT-6 Astra support with context limits, reasoning controls, Fast mode, native web search, tiered pricing, and the official model icon.
  • [Models] Add TokenDance as a built-in model provider with browser-based API key authorization.
  • [Scheduled Tasks] Daily, weekdays, and weekly presets now support multiple times per day (e.g. 09:00 and 18:00).
  • [Chat] Show topic and session titles, switch branches from the header, and export individual branches as Markdown.
  • [Skills] Add installed-skill source filters and a single/double-column layout switch in Skills settings.
  • [Quick Panel] Management actions now stay in an accessible footer across prompts, skills, MCP, Notes, Knowledge Base, and toolbar customization.
  • [Composer] Add a configurable Ctrl+I / Cmd+I shortcut to focus the input.
  • [Image Export] Message image export now uses Chromium's compositor for pixel-perfect output matching what you see on screen.
  • [Translate] Render streamed output with rich Markdown formatting.

🐛 Bug Fixes

  • [Models] Fix custom model providers that reject Cherry Studio's automatic attribution headers.
  • [Models] Fix Moonshot Kimi K2.5+ requests failing when custom sampling settings are enabled.
  • [Models] Fix Kimi K3 requests failing when an unsupported reasoning effort was selected.
  • [Models] Retired the Yi (01.AI) provider following its service shutdown.
  • [Models] Profiles migrated from v1 now retain access to all providers and models regardless of the installed edition.
  • [Chat] Fix a topic permanently failing with a 400 error after a tool call when the provider returns a non-standard item id.
  • [Chat] Fix switching between multi-model answers while a response is streaming.
  • [Chat] Fix normal chat authentication errors failing to link to provider settings after automatic retries.
  • [Chat] Native web search citations returned by providers are now displayed in chat responses.
  • [Chat] Completed Agent sessions now show unavailable token usage instead of a misleading zero.
  • [Chat] Agent client-certificate authentication failures now show proxy and SSL recovery guidance.
  • [Chat] Claude Code Agent sign-in failures now link to the provider settings needed for recovery.
  • [Chat] Add a provider settings recovery action to generic HTTP 400 errors.
  • [Chat] Fix message quotes routing to the wrong window.
  • [Scheduled Tasks] Fixed the task run log losing its “view session” link when a scheduled agent run failed, timed out, or was cancelled.
  • [Agent] Fix taskless agents disappearing from the agent list when sessions are displayed on the right.
  • [Agent] Fix Bash tool failures labeled as errors instead of command output.
  • [Agent] Fix tool calls so repaired and replayed arguments keep their expected JSON shape.
  • [Agent] Restore Claude Code's repeat-command loop guard when RTK rewrites Bash commands.
  • [Agent] Prevent agents from modifying Cherry Studio user-data SQLite files through native tools.
  • [Agent] Claude Code process failures now show an actionable cause and diagnostic reference.
  • [Agent] The context-usage indicator now advances live during a task instead of only at completion.
  • [DeepSeek] DeepSeek Harness Web sessions now start correctly when the runtime requires its token-authenticated URL.
  • [Pi] Resume Pi tasks after context-clamped length responses with a bounded compact-and-retry attempt.
  • [OpenCode] OpenCode Go automatic conversation titles now preserve session identity.
  • [Context] Fixed the context compression model selector so users can switch back to “Follow current model”.
  • [Settings] Fix the provider model list freezing the app when a provider has many models.
  • [Settings] Fix Assistant settings breaking auto-save with unsafe Max tokens values.
  • [Settings] China-edition onboarding now uses CherryIN login by default again.
  • [Settings] AMD GPU Cloud no longer shows promotional “Free” or daily-credit messaging.
  • [Quick Assistant] Fix built-in selection quick actions to call MCP tools when a quickAssistant with MCP is configured.
  • [Translate] Fixed the copy button showing its feedback on the wrong pane.
  • [Translate] Fixed translation history stars opening details instead of toggling when activated with Enter or Space.
  • [Web Search] Keep valid Querit results when individual items omit optional metadata.
  • [Word Export] Word export now renders four-space-indented code blocks, numbered lists, image alt text, strikethrough, and fixes empty blockquote errors.
  • [Markdown] Horizontal scrollbars are now visible on overflowing content inside markdown.
  • [Browser MCP] Prevent hidden or inactive built-in browser pages from playing audible media.
  • [Backup] Fix misleading “Backup file format error” when restoring a backup created by a newer version.
  • [Backup] Fixed Qiniu Cloud S3 backup, restore, management, and retention cleanup when using regional endpoints.
  • [Region] Region-dependent routing now falls back to global endpoints when IP region detection fails instead of treating failure as mainland China.
  • [Region] Fix region detection reusing an in-flight request from a different proxy.
  • [Mini Apps] Fixed Mini Apps search failing to find apps by their translated display names.
  • [Mini Apps] Remove raw permission key tooltips that showed untranslated strings.
  • [Notes] Fixed note and folder rename fields closing before users could enter a new name.
  • [PDF Preview] Fixed PDF page input retaining an out-of-range number when submitting.
  • [Cherry Cloud] Bring Cherry Studio to the foreground after sign-in completes.
  • [Portable] Fix Windows portable deep links opening a second app instance that can lose packaged resources.
  • [Windows] Fix garbled localized default application names in the Open with menu.
  • [macOS] Repair macOS keychain signing.

💄 Improvements

  • [Image Generation] Disable automatic retries to avoid duplicate billable requests.
  • [Image Generation] Probe edit-only image models via edit mode.
  • [Agent Channels] Begin processing incoming messages after a one-second batching delay.

⚡ Performance

  • [Code Editor] CodeMirror theme catalog is now loaded only when a code editor is opened.
  • [HTML Artifacts] HTML artifact previews skip full-element layout sweeps for scrollable content.
简体中文

Cherry Studio 2.0.13 - 新功能、修复与改进

✨ 新功能

  • [模型] 新增 GPT-6 Astra 支持,包括上下文限制、推理控制、快速模式、原生联网搜索、分层定价和官方模型图标。
  • [模型] 新增 TokenDance 内置模型服务商,支持浏览器 API 密钥授权。
  • [定时任务] 每天、工作日和每周预设现在支持每天多个时间点(如 09:00 和 18:00)。
  • [对话] 显示话题和会话标题,从标题栏切换分支,并将单个分支导出为 Markdown。
  • [技能] 技能设置中新增已安装技能来源过滤和单/双列布局切换。
  • [快捷面板] 管理操作现在固定在可访问的页脚,涵盖提示词、技能、MCP、笔记、知识库和工具栏自定义。
  • [输入框] 新增可配置的 Ctrl+I / Cmd+I 快捷键聚焦输入框。
  • [图片导出] 对话消息图片导出现在使用 Chromium 合成器,输出像素完美匹配屏幕显示。
  • [翻译] 翻译输出使用富 Markdown 格式渲染。

🐛 问题修复

  • [模型] 修复拒绝 Cherry Studio 自动归属头部的自定义模型服务商请求。
  • [模型] 修复启用自定义采样设置时 Moonshot Kimi K2.5+ 请求失败的问题。
  • [模型] 修复选择不支持的推理强度时 Kimi K3 请求失败的问题。
  • [模型] 因服务关闭,已停用零一万物 (Yi) 服务商。
  • [模型] 从 v1 迁移的配置文件现在无论安装版本如何都能访问所有服务商和模型。
  • [对话] 修复服务商返回非标准项目 ID 时工具调用导致话题永久 400 错误的问题。
  • [对话] 修复响应流式传输时切换多模型答案的问题。
  • [对话] 修复自动重试后普通对话认证错误无法链接到服务商设置的问题。
  • [对话] 服务商返回的原生联网搜索引用现在显示在对话响应中。
  • [对话] 完成的智能体会话现在正确显示不可用的令牌用量,而非误导性的零值。
  • [对话] 智能体客户端证书认证失败现在显示代理和 SSL 恢复指导。
  • [对话] Claude Code 智能体登录失败现在链接到恢复所需的服务商设置。
  • [对话] 为通用 HTTP 400 错误添加服务商设置恢复操作。
  • [对话] 修复消息引用路由到错误窗口的问题。
  • [定时任务] 修复定时智能体运行失败、超时或取消后任务运行日志丢失”查看会话”链接的问题。
  • [智能体] 修复会话显示在右侧时无任务智能体从列表消失的问题。
  • [智能体] Bash 工具失败现在正确标记为错误而非命令输出。
  • [智能体] 修复工具调用中修复和重放的参数保持预期的 JSON 格式。
  • [智能体] 恢复 RTK 重写 Bash 命令时的重复命令循环保护。
  • [智能体] 防止智能体通过原生工具修改 Cherry Studio 用户数据 SQLite 文件。
  • [智能体] Claude Code 进程失败现在显示可操作的成因和诊断参考。
  • [智能体] 上下文用量指示器现在在任务期间实时推进,而非仅在完成时更新。
  • [DeepSeek] 当运行时需要令牌认证 URL 时,DeepSeek Harness Web 会话现在正确启动。
  • [Pi] 上下文截断长度响应后通过有界压缩重试恢复 Pi 任务。
  • [OpenCode] OpenCode Go 自动对话标题现在保留会话身份。
  • [上下文] 修复上下文压缩模型选择器,用户可以切换回”跟随当前模型”。
  • [设置] 修复服务商模型数量较多时模型列表卡死界面的问题。
  • [设置] 修复不安全的 Max tokens 值导致助手自动保存失败的问题。
  • [设置] 中国版引导现在默认再次使用 CherryIN 登录。
  • [设置] AMD GPU Cloud 不再显示”免费”或每日积分促销信息。
  • [快捷助手] 修复配置 MCP 的快捷助手调用内置选择快捷操作时正确使用 MCP 工具。
  • [翻译] 修复复制按钮反馈显示在错误面板的问题。
  • [翻译] 修复使用 Enter 或 Space 激活翻译历史星标时打开详情而非切换的问题。
  • [联网搜索] Querit 单条结果缺少可选元数据时仍保留其他有效搜索结果。
  • [Word 导出] Word 导出现在正确渲染四空格缩进代码块、有序列表、图片替代文本、删除线,并修复空引用块错误。
  • [Markdown] Markdown 中溢出内容现在显示水平滚动条。
  • [浏览器 MCP] 防止隐藏或非活动的内置浏览器页面播放音频。
  • [备份] 修复从更新版本恢复备份时显示误导性的”备份文件格式错误”。
  • [备份] 修复使用区域或存储桶前缀端点时的七牛云 S3 备份、恢复、管理和保留清理。
  • [区域] IP 区域检测失败时,依赖区域的路由现在回退到全球默认端点,而非将失败视为中国大陆。
  • [区域] 修复区域检测复用来自不同代理的进行中请求。
  • [小程序] 修复小程序搜索无法按翻译后的显示名称查找应用的问题。
  • [小程序] 移除显示未翻译字符串的原始权限键工具提示。
  • [笔记] 修复笔记和文件夹重命名字段在输入前关闭的问题。
  • [PDF 预览] 修复 PDF 页码输入在最后一页提交时保留越界数字的问题。
  • [樱桃云] 登录完成后将 Cherry Studio 带到前台。
  • [便携版] 修复 Windows 便携版深度链接打开第二个应用实例导致丢失打包资源的问题。
  • [Windows] 修复”打开方式”菜单中本地化的默认应用程序名称显示乱码。
  • [macOS] 修复 macOS 钥匙串签名。

💄 改进

  • [图片生成] 禁用自动重试以避免重复计费请求。
  • [图片生成] 通过编辑模式探测仅编辑的图片模型。
  • [智能体通道] 现在在一秒批量延迟后开始处理传入消息。

⚡ 性能优化

  • [代码编辑器] CodeMirror 主题目录现在仅在打开代码编辑器时加载。
  • [HTML 产物] HTML 产物预览跳过可滚动内容的全元素布局扫描。

What's Changed

  • fix(release-workflow): restore generated notes and add downloads by @DeJeune in #19970
  • ci(github-actions): add workflow lint gate by @DeJeune in #19972
  • chore(release): sync v2.0.12 metadata by @kangfenmao in #19977
  • fix(quote): route message quotes to the originating window by @NatsUIJM in #19060
  • feat(chat): add topic branch switcher and branch export by @DeJeune in #19736
  • fix(cherry-cloud): focus app after sign-in by @zhibisora in #19980
  • fix(ai): probe edit-only image models via edit mode by @GeorgeDong32 in #19953
  • fix(external-app): preserve localized Windows app names by @SiinXu in #19983
  • fix(topic-naming): preserve OpenCode session identity by @DeJeune in #19939
  • fix(mini-app): drop raw permission key from leaf tooltips by @tommyzhang100504 in #19995
  • fix(notes): keep rename editor focused by @hburaktasyurek in #19997
  • feat(claude-code-runtime): deny Bash calls that repeat without new output by @NatsUIJM in #19906
  • fix(provider-settings): remove AMD promos and set default order by @SiinXu in #20013
  • fix(protocol): register portable launcher for deep links by @SiinXu in #20015
  • feat(quick-panel): add scoped footer actions by @SiinXu in #19779
  • fix(deepseek-harness): support authenticated ready URLs by @SiinXu in #20012
  • fix(provider): retire Yi service integration by @ousugo in #20016
  • fix(agent-runtime): guard user data SQLite writes by @DeJeune in #18864
  • fix(assistant-editor): clamp unsafe max token values by @SiinXu in #20038
  • fix(kb_manage): persist absolute path as source for file items (fixes #19954) by @xxiaoxiong in #19979
  • fix(chat-errors): recover from wrapped authentication errors by @SiinXu in #20046
  • fix(ai-sdk): preserve provider-native citation sources by @SiinXu in #20047
  • fix(backup): explain newer-version backup as downgrade instead of file_format by @404-Page-Found in #19767
  • fix(chat-errors): recognize Claude Code sign-in failures by @SiinXu in #20050
  • perf(code-style): defer CodeMirror theme resolution to active editor boundaries by @GeorgeDong32 in #19727
  • fix: keep multi-model selection during streaming by @SiinXu in #20051
  • fix(chat-errors): classify client certificate failures by @SiinXu in #20052
  • feat(claude-code): advance the context-usage ring on each message_start by @NatsUIJM in #20053
  • fix(web-search): tolerate partial Querit results by @SiinXu in #19936
  • fix(markdown): render scrollbars on nested content scrollers by @tommyzhang100504 in #20058
  • fix(provider-registry): prevent unsupported Kimi K3 reasoning effort by @SiinXu in #20036
  • feat(claude-code): inject pending steers at the PostToolBatch boundary by @NatsUIJM in #20008
  • feat(skills): add source tabs and column layout switch by @ousugo in #20060
  • fix(agent-sessions): show agents without tasks by @SiinXu in #20083
  • feat(translate): render streamed output with shared StreamingMarkdown by @he-yufeng in #17627
  • fix(tool-repair): preserve canonical argument shape by @SiinXu in #20043
  • refactor(router): read page search from the route instead of a cast by @EurFelux in #20108
  • refactor(conversation-view): remove the message-only conversation view by @EurFelux in #20102
  • fix(chat-errors): recover from generic API 400 responses by @SiinXu in #20087
  • fix(image-generation): disable automatic retries by @SiinXu in #20033
  • perf(html-artifacts): skip full-element height sweeps for scrollable previews by @GeorgeDong32 in #18927
  • fix(translate): route copy feedback to the pane that was copied by @EurFelux in #20115
  • fix(browser-mcp): mute hidden and inactive pages by @SiinXu in #20075
  • fix(claude-code): key the bash loop guard on the original command for rtk rewrites by @kovsu in #20082
  • feat(composer): add focus shortcut hint by @ousugo in #20079
  • fix(agent-tools): label failed Bash results as errors by @SiinXu in #20086
  • fix(model-retry): fail over API keys on auth and rate limits by @SiinXu in #19978
  • fix(claude-code): surface process exit causes by @SiinXu in #20042
  • fix(paintings): skip image generation when the signal is already aborted by @kovsu in #20136
  • refactor(message-list): isolate per-message stream activity subscriptions by @SiinXu in #19875
  • fix(word-export): preserve formatted hyperlink text by @kovsu in #20138
  • fix(ai-sdk): stop replaying function_call item ids to the Responses API by @DeJeune in #20130
  • feat(provider-registry): add Jalapeno Cloud as an OpenAI-compatible p… by @Hidoikoto in #19058
  • fix(pdf-preview): normalize page input after submission by @ousugo in #20141
  • fix(translate): preserve keyboard activation of history stars by @ousugo in #20148
  • fix(region-service): do not treat detection failures as China by @squarepots in #20117
  • fix(provider-registry): add muse-spark 1.3 catalog by @GeorgeDong32 in #20129
  • fix(region-service): scope in-flight lookup to proxy by @squarepots in #20150
  • feat(utility-process): add core utilityProcess V1 (RFC + implementation) by @eeee0717 in #19660
  • fix(selection): inherit quickAssistant MCP for built-in quick actions by @404-Page-Found in #19765
  • fix(mini-apps): search localized display names by @ousugo in #20153
  • fix(azure-openai): preserve custom gateway URL semantics by @DeJeune in #20157
  • fix(pi-runtime): resume context-clamped length responses after compaction by @dadcolin in #20163
  • refactor(inference): run local embedding and OCR in utility processes by @eeee0717 in #19896
  • fix(export): handle empty and nested blockquotes in Word export by @kovsu in #20143
  • fix(agent-prompt): decouple agent reply language from UI locale and defer to persona instructions by @404-Page-Found in #19160
  • fix(provider-settings): bound the model list so virtualization works by @DeJeune in #20133
  • fix(context-management): restore follow-current-model option by @WhiteZeek in #20169
  • fix(onboarding): restore CherryIN login in China edition by @kangfenmao in #20155
  • fix(message-tokens): distinguish unavailable usage by @SiinXu in #20011
  • feat(data-migration): record v1 migration origin by @kangfenmao in #20171
  • fix(agent-runtime): align gateway timeout budgets by @kangfenmao in #20176
  • fix(s3-backup): normalize Qiniu bucket endpoints by @SiinXu in #20186
  • feat(provider): add tokendance provider by @jianyi-gronk in #19882
  • fix(agent-channels): reduce message batch latency by @SiinXu in #20180
  • feat(openai): support GPT-6 Astra by @SiinXu in #20056
  • fix(word-export): render indented code, ordered lists, images and strikethrough by @kovsu in #20197
  • fix(provider-registry): own Kimi sampling constraints by @lihuiyang1024 in #20205
  • fix(provider-service): show all providers for v1 upgrades by @kangfenmao in #20172
  • fix(provider): omit app attribution for custom providers by @userInner in #19531
  • test(provider-registry): cover Muse Spark Responses endpoint by @ActiveSky in #20182
  • fix(agent-task): keep run log session link for failed and cancelled runs by @kovsu in #20200
  • fix(registry-sync): restore catalog publishing and refresh older schema dirs by @DeJeune in #20208
  • feat(scheduled-tasks): support multiple times per day in schedule presets by @NatsUIJM in #20104
  • feat(image-export): pixel-faithful message image export via Chromium compositor by @NatsUIJM in #20089

New Contributors

Full Changelog: v2.0.12...v2.0.13

Don't miss a new cherry-studio release

NewReleases is sending notifications on new releases.