Vercel Labs has released agent-desktop, a free, open-source tool that gives AI coding agents computer use across macOS, Windows and Linux. Chris Tate, who also built Vercel's agent-browser, announced the release on October 31 and described it as "agent-browser, for your whole computer."
Instead of having a model read screenshots and guess where to click, agent-desktop reads the operating system's accessibility tree: the AX API on macOS, UI Automation on Windows and AT-SPI on Linux. A snapshot returns JSON with a reference for each actionable element. The agent acts on those references, then takes a fresh snapshot to confirm the change.
References are tied to the snapshot that produced them. If the interface shifts before an action lands, the tool returns errors such as STALE_REF or AMBIGUOUS_TARGET so the agent knows to look again. Output is structured JSON with error codes and a retry disposition, and the project lists about 56 commands covering observation, clicks and typing, windows, notifications, the clipboard and sessions.
Key facts
- 01Vercel Labs released agent-desktop, a free, open-source computer-use tool for AI agents on macOS, Windows and Linux.
- 02Chris Tate announced it October 31; version 1.0 with Linux support shipped October 28 and 1.0.2 followed October 30.
- 03It reads the OS accessibility tree rather than screenshots and contains no language model of its own.
- 04Actions run headless by default, and each subagent in a multi-agent session gets its own colored cursor.
- 05It installs with npm, ships x64 and ARM64 binaries, and is licensed under Apache 2.0.
agent-desktop contains no language model. It is a Rust command-line tool plus a C library, so it works with Anthropic's Claude Code, OpenAI's Codex or any agent that can run shell commands. Developers who want in-process calls can use the C library directly.
Actions run headless by default, so the real mouse pointer does not move and focus is not stolen. A headed flag switches to real mouse and keyboard input. In multi-agent sessions, each subagent gets its own colored overlay cursor, and cursors can move at the same time while actions that need the physical pointer take turns.
“Agents use your computer in parallel, each with its own cursor”— Chris Tate, Creator of Vercel's agent-browser
Vercel distributes agent-desktop through npm, with native binaries for x64 and ARM64. It requires macOS 13 or later with Accessibility permission, Windows 10 version 1809 or later, or Ubuntu 24.04 or later with AT-SPI running. The project is licensed under Apache 2.0.
Version 1.0, which added Linux support, shipped on October 28, followed by version 1.0.2 on October 30. The project started in February, with development led by Lahfir.
agent-desktop cannot see apps that lack an accessibility tree, and it does not automate web pages; Vercel points users to agent-browser for that. On Linux, mouse input needs X11 or the headed flag, and Wayland clipboard access needs wl-clipboard.
The developers warn that snapshots, screenshots and clipboard reads can contain private data and should not go into public logs. An optional helper called JEV sends screen descriptions to a configured model provider.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
