Agent browser infrastructure

BrowserOS neo vs agent-browser

Both are local, open source, speak MCP, and can reuse your Chrome logins. One is an application; one is a runtime.

agent-browser is a native Rust CLI with a background daemon from Vercel Labs. It drives Chrome over the DevTools Protocol using accessibility snapshots and element refs, ships an MCP server, and runs headless by default.

BrowserOS neo is a browser application. You install it, import your Chrome logins in one click, and your agent opens visible tabs while you watch from a dashboard. On capability defaults agent-browser lands close to neo, since both are local, open source, speak MCP and can reuse your Chrome logins, so the differences are narrower and more specific than usual.

[ 01 ]

The short version

Choose BrowserOS neo when

You want an application to open, not a runtime to wire up.
You want visible tabs and a recorded replay of every run, without enabling anything.
You want your everyday logins there by default, not behind a flag.
You are not a command-line person, or do not want to be for this.

Choose agent-browser when

You want a CLI you can pipe, script, and drop into CI or a Makefile.
You need React introspection, Core Web Vitals, accessibility audits or HAR capture.
The work runs headless, on a server, with no desktop involved.
You are embedding it inside a product rather than using it yourself.

[ 02 ]

Side by side

BrowserOS neo compared with agent-browser, row by row.
Dimension	BrowserOS neo	agent-browser
What it is	BrowserOS neoA browser application you install and sign into.	agent-browserA native CLI, a background daemon and an embeddable runtime. No window unless you ask.
Where it runs	BrowserOS neoLocally, on macOS or Windows. That is the whole list.	agent-browserLocally by default, plus CI, serverless, Vercel Sandbox, remote CDP and cloud browser providers.
Default mode	BrowserOS neoVisible. Watching the work is the point.	agent-browserHeadless. Pass --headed for a window; a local dashboard exists either way.
Your logins	BrowserOS neoOne-click Chrome import at setup, then signed in by default.	agent-browserSupported several ways — reuse a named Chrome profile, persist its own, load a state file, or use its local encrypted vault.
Several at once	BrowserOS neoSeveral agents share one browser, each with its own visible tabs.	agent-browserIsolated instances behind a --session flag, each with its own cookies and history.
Watching and reviewing	BrowserOS neoLive tabs and a dashboard, with every session saved as replayable video by default.	agent-browserA local dashboard, plus screenshots, WebM recording, traces, HAR and activity logs — on request rather than automatically.
Developer tooling	BrowserOS neoNone. It is a browser, not a debugger.	agent-browserReact component tree and render profiling, Core Web Vitals, accessibility audits, network interception, HAR.
Price	BrowserOS neoFree.	agent-browserFree.

Almost every row above is a difference of default rather than capability. agent-browser can reuse your Chrome logins, show a window, run a dashboard and record WebM — it is not the blind, logged-out sandbox this category usually is, and we are not going to imply otherwise.

What differs is what happens when you do nothing. With neo, the browser is visible, signed in and recording because that is what an application does. With agent-browser, each of those is a flag you chose — which is exactly right when the browser is a dependency inside something you are building.

[ 03 ]

Defaults are the product, and ours point at real work.

Almost every difference here is a default, and defaults are what you actually live with. agent-browser starts headless and logged out because it was built to inspect pages. neo starts signed in and visible because it was built to finish tasks.

That compounds. A visible browser holding your sessions means several agents can work at once in their own tabs while you keep using yours, and every run is there to replay afterwards.

Both are local, open source and speak MCP. What separates them is what each assumes you are trying to get done.

[ 04 ]

Common questions

What is agent-browser?

An open-source browser-automation CLI from Vercel Labs, written in Rust with a background daemon. It drives Chrome over the DevTools Protocol using accessibility snapshots and element refs, and ships an MCP server so AI agents can call it directly.

Can agent-browser use my existing Chrome logins?

Yes. Its documentation lists reusing a named Chrome profile, keeping its own persistent profile, saving and reloading a state file, auto-connecting to a running Chrome, and a local encrypted auth vault. The difference from neo is the default, not the capability.

Does agent-browser run locally or in the cloud?

Locally by default, headless unless you pass --headed. It can also run in CI, in a serverless function, in Vercel Sandbox, or attach to a remote browser over CDP.

Which should I use with Claude Code or Codex?

Both work and both are free. Pick agent-browser for a command-line tool with deep front-end debugging; pick BrowserOS neo for a browser window you can watch, everyday logins by default, and an automatic replay. An agent can have both connected.

[ 05 ]

Check it yourself

Every claim about agent-browser here was read off these pages on 4 August 2026. If something has gone stale, tell us.

What we could not confirm

We could not confirm that agent-browser saves every run as a durable, scrubbable replay the way neo does. It documents a dashboard, WebM recording, screenshots, traces and HAR; we treat replay parity as unconfirmed.
Its login reuse is limited here to what its docs list. We did not test passkeys or hardware keys.

[ 06 ]

Other comparisons

Give your agent a browser it can sign into.

Free, open source, and it sits beside the browser you already use.