✨ 100% local · open source · free forever ✨ 100% 本地 · 开源 · 永久免费

A pixel pet AI
that lives where
your cursor is.

一只在光标边住的
像素 AI 桌宠。

Hold a shortcut. Speak. Release.
MouseClaw screenshots, hears you, and runs your favorite AI CLI to do the thing.

按住快捷键 → 说话 → 松开。
截屏 + 语音 + 你已经用熟的那个 AI CLI(Claude/Codex/OpenClaw/Hermes)= 真动手。

⬇ Download for macOS ⬇ 下载 macOS 版 ★ GitHub

v0.7.0 · macOS · Apple Silicon · Developer-ID signedv0.7.0 · macOS · Apple Silicon · 开发者签名

3 modes, 1 pet.

三种模式,一只宠物。

🦞

AI summon

AI 召唤

Space

Hold, speak, release. Screenshot + voice get a markdown reply in a speech bubble. Streaming.

按住、说话、松开。截图 + 语音 → markdown 流式回到气泡。

🎙️

Voice IME

语音输入法

hold fn

100% local sherpa-onnx transcription typed at your cursor. Chinese + English, no IME switching, no cloud.

本地 sherpa-onnx 流式识别 → 直接打到光标。中英文混合,不切输入法,不上云。

📋

Clipboard hub

剪贴板 Hub

V

Everything you copied — searchable, pin-able, encrypted. Drag out, paste back, or right-click for more.

所有复制过的内容 — 可搜、可标星、加密保存。拖出、粘回、右键更多。

Pick your mouse 🐭

挑一只老鼠陪你🐭

9 pixel-art skins — mice, plus a cat, a fox and a frog. Switch anytime from the pet menu. 9 款像素风皮肤 —— 老鼠之外还有猫、狐狸、树蛙。点桌宠菜单随时换。

How to actually use it

具体怎么用

🦞 Summon the AI召唤 AI

  1. Press and hold Space (you can change it during setup)
  2. Speak naturally — "summarize this page in three sentences", "draft a reply to my boss about taking the weekend off", "explain this error message"
  3. Let go. Your speech is recognized locally, your screen is captured, and the AI streams its reply into a speech bubble next to the pet
  4. If the AI suggests a piece of text, you can tell it to type that into your current cursor position — no copy-paste needed
  5. 按住 空格(首次启动时可以改)
  6. 说话就行 —— "三句话总结这个页面"、"帮我婉拒老板加班"、"这个报错什么意思"
  7. 松开。你说的话在本地识别完,AI 看到你屏幕上的内容,回复就流到桌宠头顶的气泡里
  8. 如果你觉得 AI 的回答可以直接用,让它帮你打到光标位置 —— 不用复制粘贴

💡 Tip: hold the shortcut + drag left mouse to draw a circle on screen. The AI sees your highlight baked into the screenshot. 💡 小技巧:按住快捷键时按左键拖动 → 屏幕上画粉红圈圈 → AI 看到的截图带你的高亮。圈个报错让它解释,圈个网页区域让它总结。

🎙 Type with your voice用嘴打字

  1. Click into any input box (a Slack message, a search bar, a code editor, an email reply)
  2. Press and hold fn (the function key, bottom-left of most Mac keyboards) — you can switch this to ⌥ / ⌃ / right-shift if you don't have fn
  3. Speak. Characters appear in the input box in real-time as you talk. Punctuation gets added automatically
  4. Let go of fn. The text stays. Use it like any other typed text — backspace, copy, send
  5. 把光标放到任何输入框里(微信、搜索、邮件、代码编辑器都行)
  6. 按住 fn 键(Mac 键盘左下角那个)—— 没 fn 键可以改成 ⌥/⌃/右 shift
  7. 说话。字一边说一边出现在光标位置。中英文混合可以、标点自动加
  8. 松开 fn。字就留在那。和你手打的字一样,可以删、可以发

💡 Tip: works in Chinese, English, or mixed. Try saying "今晚 6 点半 dinner together?" — both languages come out right. 💡 小技巧:中英文混合识别。试试说"今晚 6 点半 dinner 吗"—— 中英文都出来对的。不用切输入法。

🍽 Feed it a file喂它一份文件

  1. Find a file in Finder — a PDF, a screenshot, a web page saved as HTML, a code file, even a Word doc
  2. Drag it onto the pet (the pet's mouth opens to "eat" it). You'll see "🍽 Eaten: filename.pdf" appear above the pet
  3. Now talk to the pet — "summarize in three sentences", "what's the conclusion?", "is there anything I missed?". The AI reads the file and answers
  4. 在 Finder 找一份文件 —— PDF / 截图 / 网页 / 代码 / Word 文档都行
  5. 拖到桌宠身上(老鼠会张嘴吞下)。桌宠头顶气泡显示 "🍽 已吃下:文件名.pdf"
  6. 然后就跟它说话 —— "三句话总结"、"重点是什么"、"我漏看了什么"。AI 读完文件再回你

💡 Tip: drop multiple files at once. The pet eats them all and answers based on everything. 💡 小技巧:一次拖多个文件也行,老鼠全吃下,AI 综合所有文件回答你。

📜 Continue a past chat接着上次的话题聊

  1. Click the menu bar mouse icon → "View history". Or click the pet → "History"
  2. Browse past conversations — they're searchable by what you asked
  3. Found the one you want to revive? Click "💬 Continue this topic". A panel opens with the conversation context already loaded
  4. Type or speak your follow-up. The AI remembers everything from before
  5. 点菜单栏的老鼠图标 → "查看历史记录"。也可以点桌宠 → "历史记录"
  6. 翻找以前的对话 —— 都记着,可以搜索
  7. 找到想接着聊的那条?点 "💬 继续这个话题"。会开一个新窗口,AI 已经知道你们之前聊了啥
  8. 用嘴或者打字接着问。它记得所有上下文,不用重复说

🦞 Pet follows your cursor桌宠跟着光标走

By default the pet sleeps in the corner. But during AI summon (only AI summon — not voice typing), the pet follows your mouse around. You see it land on whatever you're pointing at while you talk. It looks alive. Toggle in the tray menu. 默认情况下老鼠在屏幕角落睡觉。但当你召唤 AI 时(只有 AI 召唤、不是语音输入),老鼠会跟着你鼠标走 —— 你指哪它就过去哪。看起来像活的。托盘菜单里可以开关。

Why bother?

为什么值得装?

Zero context switch.零切换成本。

No window to open, no app to focus. The pet is already there. Hold-speak-done.

不用开窗口、不用切应用。桌宠已经在那。按住-说话-完事。

🎯Circle the bug.圈选 bug。

Hold shortcut, hold left mouse + drag — your highlight is baked into the screenshot. AI knows exactly what you mean.

按住快捷键、按住左键拖 — 你画的粉红圈直接烘进截图。AI 一眼看懂你说哪。

🧠14 AI brains.14 个 AI 后端。

Claude · Codex · Gemini · Copilot · Qwen Code · OpenCode · Cline · OpenClaw · Hermes & more. Pick the CLI you've already authed; switch from the tray.

Claude · Codex · Gemini · Copilot · 通义 Qwen Code · OpenCode · Cline · OpenClaw · Hermes 等 14 个。挑你登过的那个,托盘随时切换。

📁Workspace-aware.工作区感知。

Point at your project once → "fix this bug" actually edits the right files. Read/Edit/Bash all run with cwd set.

指一次项目目录 → 「修这个 bug」是真的改文件。AI 的工具全以那为 cwd。

🖱️Computer use.真的能动手。

"Go here" pointing at a place → opens Maps. "Convert this page to Word" → does it. macOS URL schemes + Claude's Bash tool.

指地名说「去这里」 → 开 Maps。「把这页转 Word」 → 转。macOS URL schemes + Bash 工具。

🌐Real-Chrome bridge.桥接你的真 Chrome。

One click enables chrome-devtools-mcp against a dedicated debug profile. AI drives your tabs with your logins.

一键启用 chrome-devtools-mcp 对接专用 debug profile。AI 用你登录态操作 tab。

New in v0.6 — voice typing, leveled up

v0.6 新功能 —— 语音输入法大升级(2026-05)

🗣️Chinese + English, finally right.中英混说,终于听对了。

Say "把代码 push 上去" and it writes push, not a sound-alike Chinese word. We taught the recognizer the real BPE units behind 100+ dev terms (push / pull / commit / merge / deploy…), so English words inside Chinese sentences come out as English.

说"把代码 push 上去",写出来就是 push,不再是音近的"铺石"。给识别器接上了 100+ 程序员术语的真实 BPE 建模单元(push / pull / commit / merge / deploy…),中文句子里的英文词就按英文出。

↩️Just correct yourself mid-sentence.说错了,改口就行。

Talk naturally, then fix it out loud: "…drank a bowl of soup. Oh, remove the soup part." It edits before the text lands — no backspacing. Works in both languages, all local, zero delay.

自然地说,然后口头纠正:"…喝了一碗汤。哦,把喝汤去掉。" 字落下前就改好,不用退格。中英文都行、全本地、零延迟。

🔢Rambling → a clean list.一长串口述 → 清爽清单。

Dictate "first write the report then email the client oh and book a room" and it turns into a numbered 1. 2. 3. list with line breaks. Pure rules — instant, no model, opt-in toggle.

口述"先写周报然后给客户回邮件对了还要订会议室",自动变成带换行的 1. 2. 3. 有序清单。纯规则、即时、零模型,设置里可开关。

📝Teach it your words.教它你的词。

Add a custom term in one box — no weights, no fuss. It also quietly learns from your corrections, so the more you use it, the fewer mistakes it makes.

一个小框加词就进 —— 不用填权重、不折腾。它还会从你的纠错里默默学词,越用越准。

📊See how much typing you saved.看你省了多少打字。

A banner in History shows total words spoken, time, average WPM, and the minutes of typing you saved. All counted locally.

历史窗顶部 banner:累计字数 / 时长 / 平均 WPM / 帮你省下多少分钟打字。全部本地统计。

🪶Light tasks, zero extra download.轻任务,零额外下载。

Cleanup & list-organizing run on fast local rules; translate / explain / draft-reply go to the AI CLI you already use. No 580 MB model to download — instant and lean.

清理、整理清单走快速本地规则;翻译 / 解释 / 写回信交给你已经在用的 AI CLI。不用下任何 580MB 模型 —— 即时、不臃肿。

Earlier in v0.4.6

v0.4.6 新功能(2026-05)

🧠It remembers.它会记事了。

Fully-local SQLite memory (profile / episodic / knowledge graph). It distills what you do together when idle, and recalls relevant bits on summon. A window to view / delete / pause — all on your machine.

全本地 SQLite 三层记忆(画像 / 情景 / 知识图谱)。空闲时蒸馏对你的认识,召唤时检索相关记忆。「桌宠记得的事」窗口可看 / 删 / 暂停 —— 全部存你本地。

Scheduled tasks.定时任务 / 心跳。

Say "summarize AI news every morning" — it just works. Runs in the background, delivers via a gentle bubble. A unified results page shows every run, newest-first, with markdown. Anti-hallucination harness: real web/file tools, never fabricates.

召唤说一句「每天早上整理 AI 新闻」就建好。到点后台跑、轻气泡投递。统一结果页倒序看全部执行(markdown 渲染、外链走浏览器)。反幻觉 harness:放开联网/读文件工具取真实数据,绝不编造。

🐭A name + a personality.起名 + 9 种性格。

Give it a name (🎲 random + suggestions) and pick its tone — warm, snarky, minimal, chatty, tsundere & more. Personality only changes the voice, never the answer's correctness.

给它起个名字(🎲 随机 + 建议名),选语气 —— 暖心 / 毒舌 / 极简 / 话痨 / 傲娇 等 9 种。性格只改语气,绝不影响答案正确性。

🖥️Floats over fullscreen.浮在全屏之上。

The pet now uses a real NSPanel, so it stays visible even when another app goes fullscreen. Plus per-skin procedural sound effects.

桌宠改用真正的 NSPanel,某个 app 全屏后也不会消失。外加 9 款皮肤各有音色的程序化音效。

Install in 30 seconds.30 秒装好。

Tiny installer, one drag to Applications. First-time onboarding walks you through everything — pick a shortcut, an AI, and a pet skin.

超小安装包,拖一下到应用程序就完事。首次启动桌宠带你 30 秒搞定快捷键、AI 后端、皮肤选择。

🌐Mandarin + English.中文 · 英文 · 都能听。

Onboarding asks which language you mostly speak. The pet learns to listen in your language. Mix Chinese and English in one sentence — it gets it right.

首次启动选你的常用语言。老鼠就听这种语言。一句话里中英混着说也行 —— 它都能识别对。

🇨🇳No VPN. No subscription.不用 VPN · 不用订阅。

Auto-routes to the fastest source for your location. Crash, sleep, or disconnect mid-setup — it just picks up where it left off.

自动走最快的下载源 —— 国内用户开箱即用。中途睡机器、断网、关电脑都没事,下次启动接着来。

🎓First-run tour.首次使用引导。

After models download, pet pops up with a 5-step walkthrough. "📖 Teach me" in pet menu to redo anytime.

模型下完桌宠主动 5 步教学。PetMenu「📖 教我用」可重看。

✏️Voice confirmation.语音确认编辑。

3-second countdown after transcription. Esc cancel · send · click to edit. Stops misrecognized voice burning tokens.

转写完 3 秒倒数确认。Esc 取消 · 立即 · 点气泡编辑。防止误识别浪费 token。

🍽Feed files.拖文件喂桌宠。

Drag PDF / web page / screenshot / code onto the pet. AI reads the file with your follow-up question.

PDF / 网页 / 截图 / 代码拖到老鼠嘴里 → AI 读懂文件 + 听你的提问。

🐭It has a personality now.它有性格了。

Eyes follow your cursor, nods while you type, perks up when you hover, dozes off when idle, floats a heart when you pat it, and grows closer the more you interact — across all 9 skins.

眼球追你的鼠标、打字时点头陪你、贴近抬头、闲置渐睡、摸头浮爱心、互动越多越亲密(眼睛渐大)—— 9 款皮肤全适配。

🧱Bounces off the edges.撞墙会回弹。

Follow the cursor, drag it around, or let a bubble hit a screen edge — the pet stays on-screen and does a soft squash-and-recoil against the wall. No more drifting off.

跟随光标 / 拖动 / 气泡撞到屏幕边时,桌宠被挡在屏内 + 软果冻挤压回弹,再也不会飘出屏幕。

📋Reacts to your clipboard.对剪贴板有反应。

Copy something and the pet twitches, then a row of quick actions pops above its head — translate, explain, plain-text. Click one or it fades in 5s.

复制内容 → 桌宠抖耳 + 头顶弹一排快捷动作(翻译 / 解释 / 📄 纯文本)。点一个就做,不点 5 秒消失。

🧠Never looks frozen.不再像卡死。

During long AI tasks the bubble shows what it's doing — thinking, reading a file, running a command. And tasks queue one at a time, each notified on done.

AI 处理时气泡实时显示"正在思考 / 读文件 / 跑命令"。多个任务排队逐个做完都通知,语音听写永远即时不排队。

📖Gets your jargon right.术语不听错。

Voice typing has a built-in programmer vocabulary (API / Tauri / Rust / commit…) plus your own custom words, and a fast local rule-based fixer for common misrecognitions — no LLM round-trip.

语音输入内置程序员术语库(API / Tauri / Rust / commit…)+ 可加自定义词,再加 3 秒规则的本地纠错修常见误识别 —— 全程不调 LLM。

Privacy by default.

隐私默认开。

🔐
  • Clipboard encrypted — AES-256-GCM, key in macOS Keychain, file mode 0600.
  • Transient content ignored — 1Password, Bitwarden, terminal sudo prompts never recorded.
  • Voice + screenshot stay local until you call AI. Temp files you can inspect.
  • Workspace path is opt-in. No automatic file scanning.
  • Pause anytime from the tray menu — flips a single AtomicBool.
  • 剪贴板加密 — AES-256-GCM,密钥存 macOS Keychain,文件权限 0600。
  • 遵守 transient 标志 — 1Password / Bitwarden / 终端 sudo 永远不记。
  • 语音 + 截图本地保存,只在你召唤 AI 时才发出去。
  • 工作区路径 opt-in,没设就不扫文件。
  • 托盘菜单一键暂停 — 切个 atomic 就好。

Install in 30 seconds.

30 秒装好。

  1. Click the big Download button above
  2. Open the .dmg, drag MouseClaw to Applications
  3. First launch: right-click → "Open" → "Open" again (one-time macOS check)
  4. Walk through 5-step onboarding (pick shortcut, AI backend, skin, voice key, grant permissions)
  5. Hold ⌘⇧Space anywhere and speak — try "what's on my screen"
  1. 点上面那个大大的 下载 按钮
  2. 打开 .dmg,把 MouseClaw 拖到 Applications
  3. 第一次启动:右键 → 「打开」→ 再点「打开」(macOS 一次性检查)
  4. 走 5 步 Onboarding(选快捷键、AI 后端、皮肤、语音键、给权限)
  5. 在任意 app 按住 ⌘⇧Space 说话 — 试一句「看下我屏幕上有啥」

Need an AI CLI? Run one of:

还没装 AI CLI?跑下面任一条:

npm i -g @anthropic-ai/claude-code && claude login
npm i -g @openai/codex && codex login
npm i -g openclaw
curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash