Skip to main content
agent-desktop

agent-desktop

github.com/vercel-labs/agent-desktop

Free Apache-2.0 CLI from Vercel Labs. Coding agents read and operate desktop apps through the OS accessibility tree.

DevelopmentOpen SourceArtificial Intelligence

About agent-desktop

agent-desktop is a free Apache-2.0 CLI and C library from Vercel Labs. Chris Tate introduced it on October 31, 2026 for coding agents on macOS, Windows, and Linux. Install with npm install -g agent-desktop.

A snapshot returns JSON refs. The agent acts, then snapshots again, using the accessibility tree and no pixel matching. Headless is the default.

An opt-in overlay can show one cursor per agent on macOS, Windows, and Linux X11 without replacing the system pointer. Docs are at agent-browser.dev.

Pricing: Free. The repository is Apache-2.0. The pages checked do not list a paid plan.

Pricing model: Free

Features

  • CLI and C library for macOS, Windows, and Linux
  • JSON snapshot with refs, then act and snapshot again
  • Headless by default, no pixel matching
  • Optional cursor per agent on macOS, Windows, and Linux X11
  • Apache-2.0
  • npm package agent-desktop
  • macOS, Windows, and Linux
  • Apache-2.0 CLI and C library
  • Same product as the agent-desktop tool listing
  • Docs at agent-browser.dev/desktop
  • Snapshot, ref, and verify loop
  • Skill install for a coding agent

FAQs

Common questions about agent-desktop

What is agent-desktop?
A CLI and C library from Vercel Labs. A coding agent reads a desktop app through the operating system accessibility tree, acts on a ref, and snapshots again. It does not match pixels, and it has no model inside it.
How much does agent-desktop cost?
It is free under the Apache-2.0 license. Install it with npm install -g agent-desktop. The pages checked do not list a paid plan.
Can several agents use the computer at once?
An opt-in overlay gives each agent its own cursor on macOS, Windows, and Linux X11. The overlay does not replace the system pointer. Headless actions still take turns on that shared pointer, while the drawn cursors can move at the same time.
Does it drive websites?
No. The docs say it is for desktop apps. Web pages are left to agent-browser. For the web contents of a Chromium or Electron app you launch, agent-desktop can open a DevTools port that agent-browser connects to.
Where does it run?
macOS 13 or later, Windows 10 1809 or later, or Ubuntu 24.04 or later with AT-SPI.
How does an agent load the guide?
Run agent-desktop skills get desktop, or npx skills add https://github.com/vercel-labs/agent-desktop --skill agent-desktop.

Similar products