Which operating systems does agent-desktop support?

agent-desktop runs on macOS 13 or later (Intel or Apple Silicon) through the AX API, Windows 10 1809 or Windows Server 2019 and later through UI Automation, and Ubuntu 24.04 or later through AT-SPI. It ships native binaries for x64 and ARM64. macOS needs Accessibility permission; Linux needs a graphical session with at-spi2-core.

Does agent-desktop include an AI model?

No. agent-desktop is a tool that agents call, not an agent, and it contains no model. The reasoning loop stays in your own agent, which runs a snapshot, acts on a ref and snapshots again to verify. The binary ships version-matched skill docs, loaded with agent-desktop skills get desktop, so the agent knows how to use the commands.

How is agent-desktop different from agent-browser?

agent-desktop handles native desktop apps only, reading them through the operating system's accessibility tree. Vercel Labs points users to its sibling project agent-browser for web pages. Both give agents structured element refs instead of pixel matching, but agent-desktop cannot see an app that publishes no accessibility tree.

What changed in agent-desktop 1.1.0?

agent-desktop 1.1.0 fixes Windows overlay crashes at exit, double coordinate scaling and a black startup flash, and makes the macOS Dock, menu bar status items and system panels reachable. It adds a type --interval-ms option, improves ref re-identification on all three platforms, and removes the video feature and the macOS command host.

Vercel Labs · 2026-11-03 · major

agent-desktop 1.1 — Vercel Labs' computer-use CLI for AI agents

In short

agent-desktop lets AI agents read and operate desktop apps on macOS, Windows and Linux through the accessibility tree. Version 1.1.0 fixes the Windows overlay and reaches the macOS Dock and menu bar. It gained 2,275 GitHub stars in one day.

Vercel Labs · 2026-11-03 · 2 sources checked

GitHub card for the vercel-labs/agent-desktop repository: computer use CLI for AI agents

A command-line tool that lets any AI agent see and click native desktop apps, without matching pixels.

Key specs

GitHub stars	10,963
Stars today	2,275

Quick facts

Maker	Vercel Labs
Version	v1.1.0
License	Apache-2.0
Platforms	macOS 13+, Windows 10 1809+, Ubuntu 24.04+
Commands	About 60, same names on every OS
Install	npm install -g agent-desktop

What is it?

Version 1.1.0 of agent-desktop, Vercel Labs' computer-use CLI and C library, shipped on 3 November, days after the 1.0 release. It fixes crashes at exit, double scaling and a black startup flash in the Windows overlay, lets macOS agents reach the Dock, menu bar items and system panels, and adds `type --interval-ms` for steady typing. The repository was the top trending project on GitHub with 2,275 new stars in a day.

How does it work?

Instead of reading screenshots, the tool reads each app through the operating system's accessibility APIs: AX on macOS, UI Automation on Windows and AT-SPI on Linux. A snapshot returns JSON with a ref for every element an agent can act on; the agent clicks or types on a ref, then takes a new snapshot to check the result. In 1.1.0 refs are re-identified more reliably — through the element's accessibility address on macOS and a capture rule on Linux and Windows.

Why does it matter?

Many desktop apps have no API, and pixel-based computer use is slow and fragile. A structured element tree with stable refs lets any agent drive native apps with the same roughly 60 commands on three operating systems. The limit is clear: an app that publishes no accessibility tree stays invisible.

Who is it for?

developers building agents that operate desktop apps

Frequently asked questions

Which operating systems does agent-desktop support?: agent-desktop runs on macOS 13 or later (Intel or Apple Silicon) through the AX API, Windows 10 1809 or Windows Server 2019 and later through UI Automation, and Ubuntu 24.04 or later through AT-SPI. It ships native binaries for x64 and ARM64. macOS needs Accessibility permission; Linux needs a graphical session with at-spi2-core.
Does agent-desktop include an AI model?: No. agent-desktop is a tool that agents call, not an agent, and it contains no model. The reasoning loop stays in your own agent, which runs a snapshot, acts on a ref and snapshots again to verify. The binary ships version-matched skill docs, loaded with agent-desktop skills get desktop, so the agent knows how to use the commands.
How is agent-desktop different from agent-browser?: agent-desktop handles native desktop apps only, reading them through the operating system's accessibility tree. Vercel Labs points users to its sibling project agent-browser for web pages. Both give agents structured element refs instead of pixel matching, but agent-desktop cannot see an app that publishes no accessibility tree.
What changed in agent-desktop 1.1.0?: agent-desktop 1.1.0 fixes Windows overlay crashes at exit, double coordinate scaling and a black startup flash, and makes the macOS Dock, menu bar status items and system panels reachable. It adds a type --interval-ms option, improves ref re-identification on all three platforms, and removes the video feature and the macOS command host.

Try it

npm install -g agent-desktop && agent-desktop snapshot --app Finder -i