Agent Desktop
Vercel Labs' open-source computer-use layer that lets AI agents read and operate desktop apps on macOS, Windows and Linux through the OS accessibility tree instead of guessing from screenshots. A snapshot returns JSON with a stable ref for every clickable element; the agent acts on a ref and snapshots again to verify, which keeps actions safe to retry, saves tokens by scoping to one app region and lets several agents work in parallel. Rust CLI plus C library, installed with npm install -g agent-desktop. Apache-2.0.
Updated
✨ Try AgentScout's free AI image generator →Good for: coding in the terminal, browsing the web, controlling a computer, building AI agents
Use cases
Boost Agent Desktop with our $15 Fast-Track + 7-Day Launch Boost, pin it to the top of browser/computer use for 30 days + permanent /best/ & /compare/ placement + dual X spotlight (@clampdown19 & @AgentScoutHQ) for $49, or book the exclusive $99/wk sitewide & newsletter slot.
About Agent Desktop
Agent Desktop is listed in our browser/computer use category and is aimed at builders. Vercel Labs' open-source computer-use layer that lets AI agents read and operate desktop apps on macOS, Windows and Linux through the OS accessibility tree instead of guessing from screenshots. A snapshot returns JSON with a stable ref for every clickable element; the agent acts on a ref and snapshots again to verify, which keeps actions safe to retry, saves tokens by scoping to one app region and lets several agents work in parallel. Rust CLI plus C library, installed with npm install -g agent-desktop. Apache-2.0. Pricing, as the maker states it: Open source. Maturity: Active. We last checked the official link on 2026-10-31.
Agent Desktop at a glance
- Category: Browser/computer use
- Built for: Developers and teams building with AI agents
- Pricing: Open source (as the maker states it)
- GitHub: vercel-labs/agent-desktop · 3,416 stars (as of Oct 31, 2026)
- Official site: agent-browser.dev/desktop
- Listed on AgentScout: Oct 31, 2026
Agent Desktop FAQ
What is Agent Desktop?
Vercel Labs' open-source computer-use layer that lets AI agents read and operate desktop apps on macOS, Windows and Linux through the OS accessibility tree instead of guessing from screenshots. A snapshot returns JSON with a stable ref for every clickable element; the agent acts on a ref and snapshots again to verify, which keeps actions safe to retry, saves tokens by scoping to one app region and lets several agents work in parallel. Rust CLI plus C library, installed with npm install -g agent-desktop. Apache-2.0.
How much does Agent Desktop cost?
Agent Desktop's pricing, as the maker states it: Open source. Open-source tools are free to use, though you may still pay for model API usage or a hosted plan. Check the official site for current plans.
Where can I find Agent Desktop's source code?
On GitHub at vercel-labs/agent-desktop, which had 3,416 stars as of Oct 31, 2026.
What are alternatives to Agent Desktop?
Similar tools include Terminator, AutoCua, axuse, FluidUse, only-cli and Pudding.
Facts from our link-checked listing, the GitHub repository and the official site. See something wrong? Suggest an edit.
Agent Desktop ratings and reviews
Be the first to rate Agent Desktop: tap 1 to 5 stars below. A written review is optional.
Agent Desktop alternatives
Other browser/computer use tools in our directory. See all browser/computer use tools →
Terminator
Desktop automation library and MCP server that lets AI assistants control any Windows app through the accessibility tree, like Playwright for the desktop.
AutoCua
AutoCua is an open-source computer use agent that drives real desktop apps with mouse and keyboard on Windows, macOS and Linux, and also works in Chrome, Firefox and other browsers, in a real terminal and on an iPhone, iPad or iOS Simulator. It is built as a multi-agent harness you install with pip, so one agent can carry a task across the desktop, the web, the shell and a phone.
axuse
axuse gives AI agents browser and computer use through a single MCP server that plugs into Claude Code, Codex, OpenCode and other MCP hosts. Instead of a screenshot loop it reads the DOM, the accessibility tree and Chrome DevTools, which makes actions faster and uses far fewer tokens, and it has zero dependencies, with desktop control on Windows, macOS (opt-in) and Linux.
FluidUse
Local computer use on Apple silicon: on-device typed decisions with laya and form filling with CUA-S1-FORMS, driven through the macOS Accessibility API
only-cli
only-cli (oc) turns any website into a compact command-line view for AI agents. It distills a page into short text with numbered actions that agents like Claude Code can follow and click through, so agents browse the web using a small fraction of the tokens raw HTML or screenshots would take.
Pudding
Pudding is a desktop AI workspace from Teatak for macOS and Windows 11 that connects your own model API keys or local Ollama models with a built-in browser, file and command tools, a Studio for Markdown documents and typed tables, scheduled tasks, parallel subtask delegation and MCP plugins. Every tool call is shown in the conversation with per-project permission controls (Ask, Auto or Full access).
Get your badge
Built Agent Desktop? Optional. Showing this badge does not change your rank or earn a Featured spot. It links to this page. About badges
