Most browser automation for AI agents runs through heavyweight frameworks or a remote browser in the cloud. Vercel Labs took a different route: agent-browser is a native Rust command-line tool that lets an AI agent drive a real Chrome browser on the machine, one shell command at a time. The agent sees the page through accessibility-tree snapshots and refers to elements by short refs — click this, type that, read the result — without screenshots or a running server in between. A Node.js fallback is included for environments where the Rust binary is not used. The project is Apache 2.0 licensed, its repository was created on 11 January 2026, and its documentation lives at agent-browser.dev.
The growth has been steady through the autumn: 44,600 stars on 26 October, 47,397 on 8 November, 48,636 on 10 November, 51,915 on 16 November, and 53,093 stars with 3,711 forks as of 21 November 2026 (RepoRadar). Version 0.43.0 was released on 12 November 2026, and the project recorded roughly 1.63 million installs via skills.sh as of 21 November. Star counts measure interest, not production use — but the install figure suggests this is not just a bookmarked curiosity.
Key facts
| Item | Details |
|---|---|
| Repository | vercel-labs/agent-browser (GitHub) |
| From | Vercel Labs |
| What it is | Native Rust CLI letting an AI agent drive a real Chrome browser |
| How it works | Accessibility-tree snapshots + element refs, one shell command at a time |
| Fallback | Node.js fallback included |
| Licence | Apache 2.0 |
| Docs | agent-browser.dev |
| Repo created | 11 January 2026 |
| Current version | v0.43.0 (12 November 2026) |
| Installs | ~1.63 million via skills.sh (as of 21 November 2026) |
| GitHub stars | 53,093 / 3,711 forks as of 21 November 2026 (RepoRadar) |
The security caveat — read this before using it
This tool can reuse the user's Chrome profile and logins. That means an agent driving the browser through agent-browser acts as the user on signed-in sites — it can reach whatever the user is logged into. One review notes that security controls are off by default and that the repository has no SECURITY.md file. That does not make the tool unusable, but it does make it a tool to run with care: use a separate browser profile for agent work, keep it away from banking and sensitive accounts, and do not hand it to an agent you would not trust with your own logged-in session.
What this means
Agent-browser matters because it removes friction. Earlier agent-browser setups needed a server, a driver, or a cloud browser; here the agent just runs shell commands against a real browser on the same machine. That simplicity is a big part of why it has grown so fast — and why the security caveat matters just as much. The easier a tool makes it for an agent to act as you, the more carefully you need to decide where it is allowed to act.
For students learning how agents work, the design is instructive: the accessibility tree is the same structure screen readers use, so the agent "sees" the page as a blind user would — buttons, headings, links and form fields, in order, with labels. That turns out to be a cleaner input for a language model than a screenshot, and it is a technique worth understanding whether or not you use this particular tool.
A note on verification
The star and install figures in this post are snapshots taken on the dates shown, from RepoRadar and skills.sh, and will have moved since. The security caveat is as recorded in the verified research row (a review's observation that security controls are off by default and that the repo has no SECURITY.md). No claim here has been checked against the live repository beyond the 21 November 2026 snapshot.
Source: vercel-labs/agent-browser repository, agent-browser.dev, RepoRadar and skills.sh, via verified research. Last checked: 21 November 2026.
FAQs
What is agent-browser?
agent-browser is a native Rust command-line tool from Vercel Labs (Apache 2.0) that lets an AI agent drive a real Chrome browser one shell command at a time, using accessibility-tree snapshots and element refs. A Node.js fallback is included, and the docs are at agent-browser.dev.
How popular is it?
It had 53,093 stars and 3,711 forks as of 21 November 2026 (RepoRadar), up from 44,600 stars on 26 October, with roughly 1.63 million installs via skills.sh. Stars measure interest, not production use.
What is the security caveat?
The tool can reuse the user's Chrome profile and logins, so an agent using it acts as the user on signed-in sites. One review notes security controls are off by default and the repo has no SECURITY.md. Use a separate browser profile for agent work and keep it away from sensitive accounts.
Which version is current?
Version 0.43.0 was released on 12 November 2026. The repository was created on 11 January 2026.
Comments (0)
Share your question or experience about this syllabus. Your comment will appear after review.
No comments yet — be the first to share your thoughts!