{"tag_name":"v2.0.9","target_commitish":"4e988d2a14260926e9981daec016bb1903dbcc6e","prerelease":false,"name":"v2.0.9","body":"\u003c!--LANG:en--\u003e\nCherry Studio 2.0.9 - Features, Fixes, and Experience Improvements\n\n✨ New Features\n- [WeChat] Agents can now send documents and videos through WeChat channels, while incoming video and voice messages are saved to the Agent workspace.\n- [Skills] Installed Skill details once again show the source file tree and content preview, with optional one-click Chinese translation for Markdown files.\n- [Skills] Installed Skill details now include an action to open the managed skill folder.\n- [Agent] Agent sessions now automatically discover workspace skills stored under .agents/skills.\n- [Appearance] Chat conversation list and work task list positions can now be changed from Settings → Appearance.\n- [Sidebar] Sidebar items can now be opened in a new tab with middle-click or the context menu.\n- [Models] Added DeepSeek V4 Flash Vision Exp with image input support.\n- [Export] Notion export now converts GitHub-style Markdown alerts into native Notion callout blocks.\n- [Files] The Agent workspace file tree now offers a context-menu action to copy a file path.\n- [Prompts] Prompts can now be global or limited to selected contexts, with context-specific ordering and management in Settings.\n- [MCP] Agents can now register an MCP server in one step; new servers remain inactive and untrusted until enabled unless activation was explicitly requested.\n- [Usage] Redesigned usage analytics with a clearer overview, independent request sorting, and interactive charts.\n- [Mini Apps] Mini apps can now be shown side by side in a split view.\n- [Sidebar] Assistants and Agents can now be pinned to the sidebar for quick access to their latest conversation.\n- [Providers] Provider and model catalog updates can now be delivered without waiting for an app release.\n\n🐛 Bug Fixes\n- [Providers] Model deletion now keeps the model list synchronized when a request reports failure after the model was already removed.\n- [Providers] Updating provider settings now preserves untouched nested options and correctly removes explicitly cleared values.\n- [Screenshot] Capture overlays now remain on the active macOS Space.\n- [Composer] Fixed the first paperclip attachment not appearing immediately after switching to an unused assistant.\n- [Agent] Fixed an immediate retry after stopping a session returning an empty or stalled response or losing Claude Code resume state.\n- [UI] Fixed the left sidebar resize handle becoming misaligned after returning from Settings.\n- [Code Editor] Fixed artifact file previews not filling or scrolling correctly and restored syntax highlighting for Markdown, TSX, TOML, Vue, PHP, and other file types.\n- [Painting] Painting size selections now keep their full outline visible.\n- [Settings] Generic Settings entry points now consistently open Model Services.\n- [LM Studio] Fixed LM Studio models not appearing in Pi and DeepSeek Harness Agent model pickers.\n- [Gemini] Fixed Gemini 3 latest aliases failing after the first tool call because replayed calls lacked thought signatures.\n- [Painting] Image count controls now reject fractional values while continuous controls retain decimal precision.\n- [Markdown] Fixed streamed inline code appearing abruptly and occasionally causing a rendering error.\n- [Agent] Global search no longer exposes hidden model reasoning, and existing search data is cleaned automatically.\n- [Agent] Referenced conversations are now treated as historical context and cannot authorize Agent actions.\n- [Composer] Fixed pasted Windows screenshots being dropped when the clipboard also contains text.\n- [Gemini] Fixed Gemini 3.x requests failing with HTTP 400 when web search and MCP tools are used together.\n- [Agent] Deleting an Agent now also removes its scheduled tasks instead of leaving them running and failing in the background.\n- [Migration] An unreadable legacy Agent database no longer blocks v2 migration; it is preserved and reported while remaining data continues migrating.\n- [Input] Confirming a Chinese, Japanese, or Korean IME candidate no longer saves raw composition text when renaming a conversation.\n- [Onboarding] Fixed a fresh installation crashing on the model-selection step.\n- [Markdown] Copying a table now returns the original Markdown table instead of unrelated message content.\n- [CodeMate] Fixed DeepSeek Harness failing to launch when old managed API-key entries remain in its credentials file.\n- [MCP] Fixed workspace file search failing to parse ripgrep results correctly.\n- [Prompts] Prompt and translation placeholders now preserve literal replacement text, including special replacement characters.\n- [Artifact Pane] Fixed the elevated composer remaining visible after maximizing the artifact pane.\n- [QuickPanel] Fixed hidden actions remaining clickable after switching back to a tab.\n- [Models] Recognized custom models now inherit missing registry descriptions and input/output capabilities.\n- [PDF] Fixed BabelDOC installation failing in mainland China by using accessible Python, font, and model mirrors.\n- [Tools] Python-based tool installation in mainland China now retries alternate package mirrors when the default mirror is not ready.\n- [Composer] Markdown attachments can once again be opened in the file preview from the Composer or sent messages.\n- [Models] Existing custom models now inherit context limits and pricing when a registry update recognizes them.\n- [Screenshot] The IME candidate window now remains visible while annotating screenshots with text.\n- [API Gateway] Fixed reasoning effort being lost when Chat Completions requests are converted to OpenAI Responses requests.\n- [Screenshot] Escape now dismisses a capture overlay before a selection is made, including on another hovered display.\n- [API Gateway] Fixed Agent sessions failing on OpenAI-compatible endpoints that require system messages to appear first.\n- [Providers] Fixed Anthropic-compatible providers failing when thinking blocks omit a signature.\n- [Screenshot] Fixed the annotation toolbar appearing off-screen after a full-screen capture.\n- [Startup] Launch on Startup now registers with the operating system and reapplies the saved preference when Cherry Studio starts.\n- [Agent] Pi and DeepSeek Harness Agents can now use models without a configured context window by falling back to 256K.\n- [Ollama] Local Ollama models now use their reported context window, preventing tool-based Agent sessions from being truncated and failing.\n- [CodeMate] Fixed DeepSeek Harness 0.1.1 and newer rejecting the credentials layout written by Cherry Studio.\n- [Providers] Fixed Volcengine Ark multi-turn Responses and web search, Hugging Face reasoning and tool use, compatible third-party reasoning settings, OpenAI reasoning summaries, and Fast mode for supported Doubao models.\n- [Quick Assistant] Quick Assistant now uses the configured Quick Model instead of silently falling back to the default model.\n- [MCP] MCP prompts with arguments now collect values in a form before inserting the rendered prompt.\n- [Mini Apps] Keyboard shortcuts such as Ctrl+Tab now work while focus is inside a Mini App.\n- [API Gateway] Fixed Agent requests to OpenAI reasoning models failing because compatible providers sent max_tokens instead of max_completion_tokens.\n- [Composer] Fixed the input box restoring an already-sent message after sending.\n- [Agent] Pi Agent citations now render as citation badges with a source list instead of literal markers.\n- [Search] Opening another conversation while message location is in progress no longer loads the entire history into the newly opened conversation.\n- [Migration] V1-to-v2 migration now skips IndexedDB records whose large backing value can no longer be read.\n- [Tools] Dependency installs now validate the executables a recipe actually provides, fixing false “installed but not runnable” errors.\n- [Artifacts] Fixed inline HTML previews continuously jumping between two heights when content ends with bare text.\n- [Chat] Fixed arrow-key scrolling jitter in message lists.\n- [Chat] Topic rename success now appears only after the title is saved, and automatic rename failures are reported.\n- [Citations] Switching conversations now closes the citations panel so sources from the previous conversation are not shown.\n- [OVMS] Downloaded OVMS models now remain visible in Model Manager even when server-side loading fails.\n- [Shortcuts] The clear-messages shortcut once again clears the active conversation.\n- [Shortcuts] Restored Ctrl+L as the clear-messages shortcut, with a confirmation step.\n- [Tabs] Message navigation and selection now target the correct tab when the same conversation is open more than once.\n- [Composer] Drafts, attachments, and structured tokens are preserved when the main process blocks a send request.\n- [Navigation] Actions no longer target the previous assistant or Agent conversation while a newly selected conversation is loading.\n- [Chat] Clearing messages from a non-active conversation now affects the selected conversation instead of the active one.\n- [Migration] Data-directory migration on Windows now copies symbolic-link targets as ordinary files and directories.\n- [Notes] The first click after renaming a note now switches to the selected note correctly.\n- [Windows] Restored Bash command output; if CLAUDE_CODE_GIT_BASH_PATH points to git-bash.exe, update it to Git Bash's bin/bash.exe.\n- [Chat] Conversation and Agent session titles no longer briefly revert after an inline rename.\n- [Agent] Reused empty Agent sessions now appear in the correct recent date group.\n- [Agent] V2 Agent identity and memory files are preserved when their old v1 source workspace no longer exists.\n- [Ollama] Fixed Claude Agent SDK system updates being dropped by some Ollama model templates.\n- [Tasks] Scheduled tasks now report accurate next-run times after pause, resume, manual runs, and restarts.\n- [CodeMate] Fixed stale provider selections interfering with the currently selected Code Mate configuration.\n- [Navigation] A chat window no longer remains on an assistant conversation that was deleted elsewhere.\n- [Search] Message-search navigation now stops loading old history after switching conversations.\n- [Agent] Continuing after an unanswered tool approval no longer fails with “No tool output found for tool call”.\n- [Code] Code blocks no longer flash as unhighlighted text before syntax highlighting appears.\n- [API Gateway] Agent sessions on non-Anthropic models no longer receive internal Agent SDK context as an assistant reply.\n- [Skills] Missing subagent dependencies now show a clear error, and unresolved CLI executables show a warning instead of silently producing unrelated output.\n- [API Gateway] Anthropic requests containing server tools without an input schema no longer fail the whole request on providers that cannot run those tools.\n- [Agent] Pi Agent citations now resolve for web-search results returned by third-party MCP tools.\n- [Tabs] Transient Mini App tabs no longer offer an unsupported Pin Tab action.\n- [Qwen] Fixed the thinking toggle for locally deployed Qwen models on OpenAI-compatible providers.\n- [Skills] Fixed GitHub Skill installation from raw links, repository-root descriptors, lowercase skill.md files, and explicit branches or tags.\n\n💄 Improvements\n- [Agent] Improved Agent creation and editing with consistent navigation, clearer runtime choices, integrated avatar editing, and reliable permission defaults.\n- [Agent] Pi Agent now uses its official icon.\n- [Memory] Added a memory overview and clarified the roles of Agent File Memory, Knowledge Base, and MCP Memory after removal of v1 Global Memory.\n- [Agent] Added a back button to subagent flow details.\n- [Agent] Agent creation tools now refresh the Agent list and offer a Go to chat action for the newly created Agent.\n- [QuickPanel] Improved row layout so command and Skill names remain readable while descriptions use the available space.\n- [Launchpad] Refreshed built-in apps with dedicated multicolor icons and consistent sizing.\n- [Prompts] Aligned prompt editor action buttons for consistent sizing, shape, color, and icon scale.\n- [Models] Removed native number spinners from numeric model-setting fields.\n- [QuickPanel] Improved Skill-name readability.\n- [Models] Model selectors now identify duplicate names by provider and model ID.\n- [Composer] Truncated pasted links now show an ellipsis, reveal the full URL on hover, and display preview cards faster.\n- [Artifact Pane] Maximizing and restoring the right pane now animates smoothly without content jumps, overlap, or dark-mode transparency.\n- [Files] File-opening menus now show bounded application choices and remember the preferred app for each file type.\n- [PDF] PDF translation now shows accurate progress while required resources are downloaded.\n- [Agent] Agent questions now show their full title instead of truncating it to one line.\n- [Permissions] Improved permission-mode readability and restored warning tooltips.\n- [Assistant] Selection Assistant and Quick Assistant settings now show the active model and link directly to its global configuration.\n- [Providers] Action required: the retired GitHub Models provider has been removed; configure another provider if you previously relied on it.\n- [MCP] Built-in flomo and Nowledge Mem servers now use their real Streamable HTTP endpoints; existing servers migrate automatically, but flomo users must authorize once more.\n- [Privacy] The August 2026 privacy policy adds model-relay and third-party billing disclosures; users with anonymous data collection enabled must review it before analytics resumes.\n- [Permissions] Full Access now skips per-call approval for knowledge-base edits, image generation, CLI installation, and Assistant settings/file tools; choose Approve for Me or Ask Before Acting if these actions should still prompt.\n\n⚡ Performance\n- [Installation] Reduced unpacked DeepSeek Harness runtime files to improve Cherry Studio installation speed.\n- [Painting] Painting history now loads independent file metadata concurrently.\n- [Screenshot] Capture overlays now appear immediately.\n- [Local Models] Hardware acceleration is now enabled by default for local inference.\n\n\u003c!--LANG:zh-CN--\u003e\nCherry Studio 2.0.9 - 新功能、修复与体验优化\n\n✨ 新功能\n- [微信] Agent 现在可以通过微信渠道发送文档和视频，收到的视频和语音消息会保存到 Agent 工作区。\n- [技能] 已安装技能详情重新支持查看源文件树和内容，并可一键将 Markdown 文件翻译为中文。\n- [技能] 已安装技能详情新增打开托管技能文件夹的操作。\n- [Agent] Agent 会话现在会自动发现工作区 .agents/skills 目录中的技能。\n- [外观] 可在“设置 → 外观”中调整对话列表和工作任务列表的位置。\n- [侧边栏] 支持通过鼠标中键或右键菜单在新标签页中打开侧边栏项目。\n- [模型] 新增支持图片输入的 DeepSeek V4 Flash Vision Exp。\n- [导出] 导出到 Notion 时会将 GitHub 风格的 Markdown 提示块转换为原生 Notion Callout。\n- [文件] Agent 工作区文件树新增右键复制文件路径。\n- [提示词] 提示词可设为全局可用或仅在指定场景可用，并支持按场景排序和在设置中管理。\n- [MCP] Agent 现在可一步注册 MCP 服务器；除非用户明确要求启用，否则新服务器会保持未启用和不受信任状态。\n- [用量] 全新用量统计页面提供更清晰的概览、独立请求排序和可交互图表。\n- [小程序] 支持分屏并排显示两个小程序。\n- [侧边栏] 助手和 Agent 可固定到侧边栏，快速打开各自最近的会话。\n- [服务商] 服务商和模型目录现在无需等待应用发版即可热更新。\n\n🐛 问题修复\n- [服务商] 修复删除模型时请求报告失败但模型实际已删除，导致模型列表不同步的问题。\n- [服务商] 更新服务商设置时会保留未修改的嵌套选项，并正确移除明确清空的值。\n- [截图] 截图遮罩在捕获期间会保持在当前 macOS 空间。\n- [输入框] 修复切换到未使用的助手后，首次通过回形针添加的附件未立即显示的问题。\n- [Agent] 修复停止会话后立即重试可能返回空白或卡住、或丢失 Claude Code 恢复状态的问题。\n- [界面] 修复从设置返回后左侧边栏缩放手柄位置错位的问题。\n- [代码编辑器] 修复产物文件预览无法正确铺满或滚动，并恢复 Markdown、TSX、TOML、Vue、PHP 等文件的语法高亮。\n- [绘画] 绘画尺寸选项现在会完整显示选中描边。\n- [设置] 通用设置入口现在会稳定打开“模型服务”。\n- [LM Studio] 修复 LM Studio 模型未出现在 Pi 和 DeepSeek Harness Agent 模型选择器中的问题。\n- [Gemini] 修复 Gemini 3 latest 别名在首次工具调用后因缺少思考签名而失败的问题。\n- [绘画] 图片数量参数不再接受小数，连续型参数仍保留小数精度。\n- [Markdown] 修复流式内联代码突然出现，并可能在流式渲染期间触发错误的问题。\n- [Agent] 全局搜索不再暴露模型隐藏思考内容，升级时会自动清理已有搜索数据。\n- [Agent] 引用的历史会话仅作为上下文，不再能够授权 Agent 执行动作。\n- [输入框] 修复 Windows 截图在剪贴板同时包含文本格式时未发送给模型的问题。\n- [Gemini] 修复 Gemini 3.x 同时使用联网搜索和 MCP 工具时返回 HTTP 400 的问题。\n- [Agent] 删除 Agent 时会一并删除其定时任务，避免任务继续在后台运行并反复失败。\n- [迁移] 无法读取的旧版 Agent 数据库不再阻断 v2 迁移；数据库会被保留并报告，其余数据继续迁移。\n- [输入] 使用中日韩输入法重命名会话时，确认候选词不再错误保存为未上屏的拼音或组合文本。\n- [引导] 修复全新安装在模型选择步骤崩溃的问题。\n- [Markdown] 修复复制表格时得到无关消息内容而非原始 Markdown 表格的问题。\n- [CodeMate] 修复凭据文件残留旧版托管 API Key 时 DeepSeek Harness 无法启动的问题。\n- [MCP] 修复工作区文件搜索无法正确解析 ripgrep 结果的问题。\n- [提示词] 提示词和翻译占位符现在会保留包含特殊替换字符的原始文本。\n- [产物面板] 修复最大化产物面板后悬浮输入框仍然显示的问题。\n- [QuickPanel] 修复切换回标签页后隐藏操作仍可被点击的问题。\n- [模型] 已识别的自定义模型现在会补齐缺失的目录描述和输入输出能力。\n- [PDF] 修复中国大陆安装 BabelDOC 失败的问题，改用可访问的 Python、字体和模型镜像。\n- [工具] 在中国大陆安装 Python 工具时，如果默认包镜像尚未同步，会自动尝试其他镜像。\n- [输入框] 修复无法从输入框或已发送消息中打开 Markdown 附件预览的问题。\n- [模型] 自定义模型被目录更新识别后，会自动补齐上下文长度和价格信息。\n- [截图] 使用文字标注截图时，输入法候选窗口会正常显示。\n- [API 网关] 修复 Chat Completions 请求转换为 OpenAI Responses 请求时丢失推理强度的问题。\n- [截图] 选区创建前可用 Escape 关闭截图遮罩，也支持关闭鼠标所在的其他显示器遮罩。\n- [API 网关] 修复要求系统消息必须置顶的 OpenAI 兼容端点上 Agent 会话失败的问题。\n- [服务商] 修复部分 Anthropic 兼容服务商因思考块缺少签名而失败的问题。\n- [截图] 修复全屏截图后标注工具栏出现在屏幕外的问题。\n- [启动] “开机启动”现在会正确注册到操作系统，并在 Cherry Studio 启动时重新应用已保存设置。\n- [Agent] Pi 和 DeepSeek Harness Agent 可使用未配置上下文长度的模型，默认按 256K 处理。\n- [Ollama] 本地 Ollama 模型现在使用服务端上报的真实上下文长度，避免带工具的 Agent 会话被截断并失败。\n- [CodeMate] 修复 DeepSeek Harness 0.1.1 及更高版本拒绝 Cherry Studio 写入的凭据格式的问题。\n- [服务商] 修复火山方舟多轮 Responses 与联网搜索、Hugging Face 推理与多轮工具、第三方推理设置、OpenAI 推理摘要，以及部分豆包模型的快速模式。\n- [快速助手] 快速助手现在会使用已配置的“快速模型”，不再静默回退到默认模型。\n- [MCP] 带参数的 MCP 提示词现在会先通过表单收集参数，再插入渲染后的提示词。\n- [小程序] 焦点位于小程序内时，Ctrl+Tab 等键盘快捷键现在可以正常工作。\n- [API 网关] 修复 OpenAI 推理模型因兼容服务商发送 max_tokens 而非 max_completion_tokens 导致 Agent 请求失败的问题。\n- [输入框] 修复消息发送后输入框重新出现已发送文本的问题。\n- [Agent] Pi Agent 的引用标记现在会显示为引用徽章和来源列表，不再显示原始标记。\n- [搜索] 消息定位过程中打开其他会话，不再把完整历史错误加载到新会话。\n- [迁移] v1 到 v2 迁移现在会跳过大型后备值已无法读取的 IndexedDB 记录。\n- [工具] 依赖安装会按配方实际提供的可执行文件校验，修复“已安装但不可运行”的误报。\n- [产物] 修复内容以纯文本结尾时内联 HTML 预览在两个高度之间持续跳动的问题。\n- [对话] 修复使用方向键滚动消息列表时的抖动。\n- [对话] 话题重命名成功提示只会在标题保存后出现，自动重命名失败也会正确报告。\n- [引用] 切换会话时会关闭引用面板，不再显示上一个会话的来源。\n- [OVMS] 即使服务端加载失败，已下载的 OVMS 模型也会继续显示在模型管理器中。\n- [快捷键] 清空消息快捷键恢复为清空当前会话。\n- [快捷键] 恢复 Ctrl+L 清空消息快捷键，并保留确认步骤。\n- [标签页] 同一会话打开多个标签页时，消息导航和选择会定位到正确标签页。\n- [输入框] 主进程阻止发送请求时，会保留草稿、附件和结构化 Token。\n- [导航] 新会话加载期间，操作不再错误作用于上一个助手或 Agent 会话。\n- [对话] 从列表清空非当前会话时，会正确清空所选会话而非当前会话。\n- [迁移] Windows 数据目录迁移会将符号链接目标复制为普通文件和目录。\n- [笔记] 重命名笔记后的第一次点击现在会正确切换到所选笔记。\n- [Windows] 恢复 Bash 命令输出；如 CLAUDE_CODE_GIT_BASH_PATH 指向 git-bash.exe，请改为 Git Bash 的 bin/bash.exe。\n- [对话] 对话和 Agent 会话标题在行内重命名后不再短暂恢复旧名称。\n- [Agent] 复用空 Agent 会话时，会将其显示在正确的最近日期分组中。\n- [Agent] 当旧版 v1 源工作区不存在时，仍会保留 v2 Agent 的身份和记忆文件。\n- [Ollama] 修复部分 Ollama 模型模板丢弃 Claude Agent SDK 系统更新的问题。\n- [任务] 定时任务在暂停、恢复、手动运行和重启后会显示准确的下次运行时间。\n- [CodeMate] 修复过期的服务商选择干扰当前 Code Mate 配置的问题。\n- [导航] 助手会话在其他位置被删除后，聊天窗口不再停留在已删除会话。\n- [搜索] 切换会话后，消息搜索导航会停止继续加载旧会话历史。\n- [Agent] 工具审批未响应时继续对话，不再报“No tool output found for tool call”。\n- [代码] 代码块不再先闪现未高亮文本再显示语法高亮。\n- [API 网关] 非 Anthropic 模型的 Agent 会话不再把 Agent SDK 内部上下文当作助手回复发送给模型。\n- [技能] 缺少子 Agent 依赖时会显示明确错误，找不到 CLI 可执行文件时会给出警告，不再静默输出无关结果。\n- [API 网关] Anthropic 请求包含无输入 Schema 的服务端工具时，不支持这些工具的服务商不再让整个请求失败。\n- [Agent] Pi Agent 现在可以解析第三方 MCP 工具返回的网页搜索引用。\n- [标签页] 临时小程序标签页不再显示不受支持的“固定标签页”操作。\n- [Qwen] 修复通过 OpenAI 兼容服务商使用本地部署 Qwen 模型时思考开关无效的问题。\n- [技能] 修复从 GitHub Raw 链接、仓库根描述文件、小写 skill.md 和指定分支或标签安装技能的问题。\n\n💄 改进\n- [Agent] 优化 Agent 创建和编辑流程，统一导航、明确运行时选项、集成头像编辑，并可靠应用权限默认值。\n- [Agent] Pi Agent 改用官方图标。\n- [记忆] 新增记忆功能概览，明确 v1 全局记忆移除后 Agent 文件记忆、知识库和 MCP 记忆各自的用途。\n- [Agent] 子 Agent 流程详情新增返回按钮。\n- [Agent] Agent 创建工具会刷新 Agent 列表，并提供直接进入新 Agent 对话的操作。\n- [QuickPanel] 优化行布局，命令和技能名称保持清晰，描述充分利用剩余空间。\n- [启动台] 内置应用改用独立的多彩图标并统一尺寸。\n- [提示词] 统一提示词编辑器操作按钮的尺寸、形状、颜色和图标比例。\n- [模型] 模型数值设置项移除浏览器原生数字微调按钮。\n- [QuickPanel] 提升技能名称的可读性。\n- [模型] 模型选择器会用服务商和模型 ID 区分同名模型。\n- [输入框] 截断的粘贴链接会显示省略号，悬停可查看完整 URL，链接预览卡片也会更快出现。\n- [产物面板] 最大化和还原右侧面板时动画更平滑，内容不再跳动、遮挡，深色模式下也不会透出对话内容。\n- [文件] 文件打开菜单提供有限且清晰的应用选择，并按文件类型记住首选应用。\n- [PDF] 下载 PDF 翻译所需资源时会显示准确且持续前进的进度。\n- [Agent] Agent 提问会显示完整标题，不再截断为一行。\n- [权限] 提升权限模式选项的可读性，并恢复警告提示。\n- [助手] 划词助手和快速助手设置会显示当前模型，并可直接跳转到对应的全局模型配置。\n- [服务商] 需要操作：已移除停止服务的 GitHub Models；此前依赖该服务的用户需要配置其他服务商。\n- [MCP] 内置 flomo 和 Nowledge Mem 改用真实 Streamable HTTP 端点；已有服务器会自动迁移，但 flomo 用户需要重新授权一次。\n- [隐私] 2026 年 8 月隐私政策新增模型中转和第三方服务计费说明；开启匿名数据收集的用户需确认后才会恢复分析数据上报。\n- [权限] “完全访问”现在会跳过知识库编辑、图片生成、CLI 安装以及助手设置/文件工具的逐次审批；如仍需确认，请改用“自动批准”或“操作前询问”。\n\n⚡ 性能优化\n- [安装] 减少 DeepSeek Harness 运行时的解包文件数量，提升 Cherry Studio 安装速度。\n- [绘画] 绘画历史现在会并行加载互不依赖的文件信息。\n- [截图] 截图遮罩现在会立即显示。\n- [本地模型] 本地推理默认启用硬件加速。\n\u003c!--LANG:END--\u003e\n","author":{"id":"4643061","login":"kangfenmaox","name":"亢奋猫","avatar_url":"https://cdn-img.gitcode.com/da/ba/200d90a891ba012123960861c745882c96c0e46013a83014f4d985ded857f67d.png","html_url":"https://gitcode.com/kangfenmaox","type":"User","url":"https://api.gitcode.com/api/v5/users/kangfenmaox"},"created_at":"2026-08-24T12:25:04Z","published_at":"2026-08-24T12:25:04Z","assets":[{"id":"164291","name":"Cherry-Studio-2.0.9-aarch64.rpm","type":"attach","browser_download_url":"https://gitcode.com/CherryHQ/cherry-studio/releases/download/v2.0.9/Cherry-Studio-2.0.9-aarch64.rpm","updated_at":"0001-01-01T00:00:00Z"},{"id":"164292","name":"Cherry-Studio-2.0.9-amd64.deb","type":"attach","browser_download_url":"https://gitcode.com/CherryHQ/cherry-studio/releases/download/v2.0.9/Cherry-Studio-2.0.9-amd64.deb","updated_at":"0001-01-01T00:00:00Z"},{"id":"164306","name":"Cherry-Studio-2.0.9-arm64.AppImage","type":"attach","browser_download_url":"https://gitcode.com/CherryHQ/cherry-studio/releases/download/v2.0.9/Cherry-Studio-2.0.9-arm64.AppImage","updated_at":"0001-01-01T00:00:00Z"},{"id":"164307","name":"Cherry-Studio-2.0.9-arm64.deb","type":"attach","browser_download_url":"https://gitcode.com/CherryHQ/cherry-studio/releases/download/v2.0.9/Cherry-Studio-2.0.9-arm64.deb","updated_at":"0001-01-01T00:00:00Z"},{"id":"164308","name":"Cherry-Studio-2.0.9-arm64.dmg","type":"attach","browser_download_url":"https://gitcode.com/CherryHQ/cherry-studio/releases/download/v2.0.9/Cherry-Studio-2.0.9-arm64.dmg","updated_at":"0001-01-01T00:00:00Z"},{"id":"164300","name":"Cherry-Studio-2.0.9-arm64-portable.exe","type":"attach","browser_download_url":"https://gitcode.com/CherryHQ/cherry-studio/releases/download/v2.0.9/Cherry-Studio-2.0.9-arm64-portable.exe","updated_at":"0001-01-01T00:00:00Z"},{"id":"164302","name":"Cherry-Studio-2.0.9-arm64-setup.exe","type":"attach","browser_download_url":"https://gitcode.com/CherryHQ/cherry-studio/releases/download/v2.0.9/Cherry-Studio-2.0.9-arm64-setup.exe","updated_at":"0001-01-01T00:00:00Z"},{"id":"164309","name":"Cherry-Studio-2.0.9-arm64.zip","type":"attach","browser_download_url":"https://gitcode.com/CherryHQ/cherry-studio/releases/download/v2.0.9/Cherry-Studio-2.0.9-arm64.zip","updated_at":"0001-01-01T00:00:00Z"},{"id":"164310","name":"Cherry-Studio-2.0.9-arm64.zip.blockmap","type":"attach","browser_download_url":"https://gitcode.com/CherryHQ/cherry-studio/releases/download/v2.0.9/Cherry-Studio-2.0.9-arm64.zip.blockmap","updated_at":"0001-01-01T00:00:00Z"},{"id":"164314","name":"Cherry-Studio-2.0.9-x64.dmg","type":"attach","browser_download_url":"https://gitcode.com/CherryHQ/cherry-studio/releases/download/v2.0.9/Cherry-Studio-2.0.9-x64.dmg","updated_at":"0001-01-01T00:00:00Z"},{"id":"164311","name":"Cherry-Studio-2.0.9-x64-portable.exe","type":"attach","browser_download_url":"https://gitcode.com/CherryHQ/cherry-studio/releases/download/v2.0.9/Cherry-Studio-2.0.9-x64-portable.exe","updated_at":"0001-01-01T00:00:00Z"},{"id":"164313","name":"Cherry-Studio-2.0.9-x64-setup.exe","type":"attach","browser_download_url":"https://gitcode.com/CherryHQ/cherry-studio/releases/download/v2.0.9/Cherry-Studio-2.0.9-x64-setup.exe","updated_at":"0001-01-01T00:00:00Z"},{"id":"164315","name":"Cherry-Studio-2.0.9-x64.zip","type":"attach","browser_download_url":"https://gitcode.com/CherryHQ/cherry-studio/releases/download/v2.0.9/Cherry-Studio-2.0.9-x64.zip","updated_at":"0001-01-01T00:00:00Z"},{"id":"164316","name":"Cherry-Studio-2.0.9-x64.zip.blockmap","type":"attach","browser_download_url":"https://gitcode.com/CherryHQ/cherry-studio/releases/download/v2.0.9/Cherry-Studio-2.0.9-x64.zip.blockmap","updated_at":"0001-01-01T00:00:00Z"},{"id":"164317","name":"Cherry-Studio-2.0.9-x86_64.AppImage","type":"attach","browser_download_url":"https://gitcode.com/CherryHQ/cherry-studio/releases/download/v2.0.9/Cherry-Studio-2.0.9-x86_64.AppImage","updated_at":"0001-01-01T00:00:00Z"},{"id":"164318","name":"Cherry-Studio-2.0.9-x86_64.rpm","type":"attach","browser_download_url":"https://gitcode.com/CherryHQ/cherry-studio/releases/download/v2.0.9/Cherry-Studio-2.0.9-x86_64.rpm","updated_at":"0001-01-01T00:00:00Z"},{"id":"164319","name":"latest-linux-arm64.yml","type":"attach","browser_download_url":"https://gitcode.com/CherryHQ/cherry-studio/releases/download/v2.0.9/latest-linux-arm64.yml","updated_at":"0001-01-01T00:00:00Z"},{"id":"164320","name":"latest-linux.yml","type":"attach","browser_download_url":"https://gitcode.com/CherryHQ/cherry-studio/releases/download/v2.0.9/latest-linux.yml","updated_at":"0001-01-01T00:00:00Z"},{"id":"164321","name":"latest-mac.yml","type":"attach","browser_download_url":"https://gitcode.com/CherryHQ/cherry-studio/releases/download/v2.0.9/latest-mac.yml","updated_at":"0001-01-01T00:00:00Z"},{"id":"164322","name":"latest.yml","type":"attach","browser_download_url":"https://gitcode.com/CherryHQ/cherry-studio/releases/download/v2.0.9/latest.yml","updated_at":"0001-01-01T00:00:00Z"},{"id":"164323","name":"release-history.json","type":"attach","browser_download_url":"https://gitcode.com/CherryHQ/cherry-studio/releases/download/v2.0.9/release-history.json","updated_at":"0001-01-01T00:00:00Z"}]}