{"slug":"agent-desktop","name":"agent-desktop","nameZh":"agent-desktop（AI 代理桌面操作 CLI）","category":"cli","categoryName":"CLIs & execution basics","audience":["for-ai"],"primary":"for-ai","openClosed":"open","pricing":"open-source","license":"Apache-2.0","maintained":"yes","officialUrl":"https://agent-browser.dev/desktop","repo":"https://github.com/vercel-labs/agent-desktop","latestVersion":null,"latestRelease":null,"stars":18372,"starsAsOf":"2026-11-07","description":"agent-desktop is Vercel Labs' Apache-2.0 computer-use CLI that lets AI agents see and operate desktop apps on macOS, Windows and Linux. Instead of screenshots or pixel matching it reads the OS accessibility tree (AX API, UI Automation, AT-SPI) and returns it as JSON with a ref for every actionable element; the agent acts on a ref and snapshots again to verify. It has no model inside, ships as a CLI and a C library, runs headless by default, and installs via npm or Homebrew.","descriptionZh":"agent-desktop 是 Vercel Labs 以 Apache-2.0 许可证开源的「电脑操作」（computer use）命令行工具，供 AI 代理查看并操作 macOS、Windows 和 Linux 上的桌面应用。它不依赖截图或像素识别，而是读取系统无障碍树（macOS 的 AX API、Windows 的 UI Automation、Linux 的 AT-SPI），以 JSON 输出界面结构，并为每个可操作元素分配引用编号；代理按编号点击或输入，再重新读取界面核对结果。工具本身不内置模型，决策循环由调用它的代理负责，以 CLI 和 C 库两种形式提供，默认无头执行、不移动鼠标。可通过 npm 或 Homebrew 安装。官方提示快照、剪贴板与截图可能含隐私内容，不宜上传到公开日志。","callMethods":[],"platforms":["macOS","Windows","Linux","CLI"],"addedAt":"2026-11-05","updatedAt":"2026-11-07","topics":[],"language":null,"lastPush":"2026-11-05","formerly":null,"packages":[],"recentVersions":[],"alternatives":["node","deno","bun","code-server","cpython","moby","git","powershell"],"stacks":[],"scenarios":["connect-tools-to-agents"],"problems":[],"reasonZh":null,"sources":["editorial","github-pool"],"record":"https://aimap.page/api/v1/software/agent-desktop.json","pages":{"en":"https://aimap.page/en/software/agent-desktop/","zh-hans":"https://aimap.page/zh-hans/software/agent-desktop/"}}