The Execution Layer for AI Agents AI Agent 的管家 — 替你动手,不替你做主
Connect your AI Agent to your browser and phone. Receive structured instructions, execute with one click or automatically. Open source. 连接你的 AI Agent 与浏览器、手机。接收结构化指令,一键执行或自动完成。完全开源。
No more copy-paste. Your agent sends instructions, your device executes: tweet, email, fill forms, click, extract.不再复制粘贴 — Agent 发指令,你的设备直接执行:发推、写邮件、填表单、点击、抓取。
Chrome extension + mobile PWA. One token connects all your devices. Scan QR to connect instantly.浏览器插件 + 手机 PWA,一个 Token 连接所有设备。扫码即连。
Semi-automatic: confirm each instruction before execution, or enable auto-execute. Token-authenticated.半自动执行:每条指令你确认后再执行,也可开启自动模式。Token 授权。
Fully open source. Self-host or use our service. Your data stays yours.代码完全开源,社区共建。自部署或用托管服务,你的数据你做主。
Chain instructions in sequence with progress visualization and abort support.一组指令按顺序自动执行,支持进度可视化和中止。
Not locked to any agent — OpenClaw, custom agents, or any HTTP-capable AI can integrate.不绑定特定 Agent — OpenClaw、自研 Agent 或任何 HTTP AI 都可接入。
Chrome extension for browser, or add this page to home screen (PWA) for mobile.Chrome 安装浏览器插件,或手机打开本页添加到主屏幕(PWA)。
Scan QR code or enter Backend URL + Token. Connected in seconds.扫描二维码或填入 Backend URL 和 Token,一秒连通。
Your agent sends instructions → your device gets notified → tap to execute or auto-complete.Agent 发送指令 → 设备收到通知 → 点击执行或自动完成。
Simple REST API. Any AI agent that can make HTTP requests can integrate in minutes. 简单 REST API,任何能发 HTTP 请求的 AI Agent 都能在几分钟内接入。
POST /v1/instructions Header: X-ClawMe-Token: <your-token> { "target": "browser", "instruction": { "type": "compose_tweet", "payload": { "text": "Hello from my AI agent!" } } }
remind → Show a reminder (title, body) open_url → Open a URL in browser compose_tweet → Open Twitter with pre-filled text compose_email → Open Gmail/mailto with to, subject, body fill_form → Auto-fill form fields by CSS selector click → Click an element by CSS selector extract → Extract text/attribute from page element
// Add meta.workflow_id and meta.step to group instructions { "target": "browser", "instruction": { "type": "open_url", "payload": { "url": "..." } }, "meta": { "workflow_id": "my-flow", "step": 1 } }
GET /v1/instructions/pending?target=browser Header: X-ClawMe-Token: <your-token> // Client reports result after execution: POST /v1/instructions/:id/result { "status": "ok", "result": "Filled 3 fields" }