BrowserOS neo

BrowserOS neo is a standalone desktop browser built to be operated by an AI agent rather than by a person. Instead of giving an agent a headless Chromium sandbox with no cookies, neo runs locally as a second browser that imports your existing Chrome profile, so an agent can act inside sessions you are already signed in to — a dashboard, an internal admin tool, a bank portal — without you handing over passwords or waiting on an API that does not exist.

It connects to agents over the Model Context Protocol, so any MCP-capable client can drive it: Claude Code, OpenAI Codex, Cursor, VS Code, Zed, OpenClaw and Hermes are all listed as supported out of the box. Pages are handed to the model as compact structured snapshots rather than full-page screenshots or raw DOM, which keeps long automations inside a reasonable token budget.

The part that distinguishes neo from a scripted automation stack is supervision. Every run is visible while it happens, each agent gets its own isolated tabs so several can work in parallel without stealing focus, and finished sessions can be replayed step by step to see exactly which pages were opened and what was clicked. An audit view records the trail; an undo path exists for actions you did not want.

Everything runs on your machine. There is no cloud relay, no hosted browser farm and no upload of your session cookies, and you bring your own model API key. The project is fully open source under AGPL-3.0 at github.com/browseros-ai/BrowserOS, where the wider BrowserOS browser has passed 12,000 stars.

Who it is for: developers and operators who want an agent to finish real web work — pulling reports, filing forms, checking dashboards — on sites that have no usable API, while keeping the ability to watch and audit every step.

Worth knowing: neo was previously released as BrowserClaw and existing installs keep working under the old name; it ships for macOS and Windows, and it is free, with cost limited to whatever your chosen model provider charges.