BrowserOS: A Local-Only Chromium Fork That Puts an Agent in Your New Tab
🌐 The open-source Agentic browser; alternative to ChatGPT Atlas, Perplexity Comet, Dia.
At a glance
- What is it?
- BrowserOS is an open-source Chromium fork with an AI agent in every new tab, plus a separate 'neo' browser for AI agents that imports your Chrome logins. It runs locally, supports Ollama or your own API keys, and is positioned against cloud browsers and locked-in AI browsers.
- Who is it for?
- Adopt BrowserOS if you want a local, open-source browser that keeps your sessions and data on your machine, works with your own AI keys or Ollama, and supports both human-driven and agent-driven browsing. Do not adopt it if you need a cloud-hosted solution, if you rely on proprietary AI browser integrations, or if you must avoid the AGPL-3.0 license.
- Can I use it commercially?
- Yes, with strict conditions. AGPL-3.0 is a network copyleft licence: if people use a modified version over a network, for example as a hosted service, you must offer them its source code under the same licence.
- Is it still maintained?
- Yes. The repository received new commits within the last day.
- What is it written in?
- Mainly TypeScript, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 15, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
Two Products, One Repository: What BrowserOS Actually Is
The repository contains two distinct products under one name. The first, called BrowserOS, is a free, open-source Chromium fork with an AI agent built into every new tab. The second, called BrowserOS neo, is a separate browser designed for AI agents, not for humans. The README states: 'Two browsers: one for your agents, one for you.' This split is important because the project's identity is not a single tool but a pair of tools that share a codebase. BrowserOS targets people who want an AI assistant in their daily browser, while neo targets developers who use agents like Claude Code or Codex and need a browser those agents can drive. The repository layout includes packages for the browser, an agent extension, and an agent server, which suggests a modular architecture. For an engineer evaluating adoption, the first question is which product you need, because the features and workflows differ significantly.
The Core Mechanism: A Chromium Fork With an Agent in the New Tab
BrowserOS is not a browser extension that adds AI to an existing browser. It is a fork of Chromium, which means the agent is baked into the browser itself. According to the README, the agent lives in every new tab and can 'summarise a page, click through a flow, extract data, or run a scheduled task.' It uses 20+ built-in tools and 40+ app integrations. The agent can work with your own AI keys or run entirely locally with Ollama. This design choice has a clear consequence: you do not need to send data to a third-party cloud. The README emphasizes that other AI browsers require signing into their cloud and handing over data, while BrowserOS does not. The local-first approach is a genuine differentiator, but it also means the browser must handle model inference or API calls on its own, which could affect performance depending on your hardware and chosen model.
BrowserOS neo: A Second Browser for Agents, Not a Chrome Replacement
The README is explicit that BrowserOS neo is 'NOT a Chrome replacement.' It is a secondary browser that sits next to Chrome and is made agent-friendly. The key feature is one-click import of logins, bookmarks, and extensions from Chrome. This solves a real problem: headless drivers like Playwright spin up a fresh Chrome subprocess with no logins, which is useless for tasks that require your logged-in state, such as reading your inbox. BrowserOS neo persists that logged-in state across sessions. Agents run in parallel, each in its own tab, and you can watch them live or replay any session like a video. The replay feature saves every session as a scrubbable video on your disk with a step-by-step action timeline. This is an audit trail that many agent tools lack. The README claims that for the same task, BrowserOS neo consumes significantly fewer tokens than alternatives like Claude's Chrome extension or Codex browser, but that claim is not backed by benchmark data in the material, so you should treat it as a vendor assertion.
Getting It Running: Downloads, Installation, and Setup
The README provides direct download links for macOS and Windows. For BrowserOS neo, there is a .dmg for macOS and an .exe installer for Windows. The setup process is described as a few steps: install BrowserOS neo, import from Chrome in one click, and the browser automatically finds your agents, including Claude Code, Codex, Cursor, VS Code, OpenClaw, and Hermes. You connect with one click, then give a task from your agent, such as 'Book me the cheapest flight to London.' For BrowserOS, the README does not list a separate download link in the provided material, but it states you can bring your own AI keys or run everything locally with Ollama. The repository also shows separate release tags for an agent extension and an agent server, indicating that installation may involve more than just the browser binary. The docs are hosted at docs.browseros.com/neo, which is where you would find detailed setup instructions for the neo product. The lack of explicit command-line installation steps in the README suggests a focus on GUI-based setup, which may be a limitation for users who prefer scripted deployments.
Data Locality and Privacy: Where Your Sessions Live
A central promise of BrowserOS is that sessions, screenshots, and history live under ~/.browserclaw/ and never leave your machine. This is a concrete privacy claim that distinguishes it from cloud browsers like browser-use or browserbase, which run in a datacenter and often face blocks from sites like Twitter and LinkedIn because of datacenter IP addresses. Running on 127.0.0.1 avoids those blocks and keeps your logins usable. However, local-only storage has trade-offs. If your machine fails or you switch devices, your session data may not be portable unless you manually back up the ~/.browserclaw/ directory. The README does not mention any sync or backup mechanism. For an engineer, this means you need to account for data persistence yourself. Also, 'never leave your machine' applies to the browser's own storage, but if you use an external AI API, the prompts and page content you send to that API will leave your machine. The privacy guarantee is about the browser's data, not about the AI provider's data handling.
Licensing and Maintenance Considerations
The project is licensed under AGPL-3.0. This is a strong copyleft license that has implications if you modify the code and deploy it as a network service. The AGPL requires that if you modify the software and make it available over a network, you must offer the corresponding source code to users. For an internal tool that you do not distribute, this may not be an issue, but for a company that provides a hosted service based on this code, the obligations are significant. The README does not discuss licensing, so you should consult the full license text or a legal professional before integrating it into a commercial product. On maintenance, the repository shows recent releases: agent-server v0.0.162 was pushed on 2026-09-09, and the last push to the main branch was the same day. This indicates active development. However, the project is not archived, and the topic list includes hacktoberfest, which suggests community contributions are welcome. The dual-product structure and the presence of separate release tracks for the extension and server mean that updates may come at different cadences, so you should monitor both if you rely on the full stack.
Alternatives and the Real Differences
The README explicitly names alternatives: ChatGPT Atlas, Perplexity Comet, Dia, Playwright, agent-browser, browser-use, and browserbase. The key differences are approach-based. Playwright and agent-browser are headless drivers that create fresh browser sessions without logins, which is fine for CI but not for tasks requiring your authenticated state. Browser-use and browserbase are cloud browsers that run in datacenters, which creates login friction and IP-blocking issues on social platforms. Atlas, Comet, and Dia are AI browsers that only work with their own AI. BrowserOS neo distinguishes itself by working with agents you already use, like Claude Code and Codex, and by running locally with imported logins. BrowserOS itself differs from these by being a full Chromium fork with an embedded agent that can run on Ollama. The practical difference is control: you own the browser, the data, and the model choice. The trade-off is that you must manage the infrastructure yourself, including keeping the fork updated with Chromium security patches, which is a non-trivial maintenance burden for an individual or small team.
Editorial conclusion
Adopt BrowserOS if you want a local, open-source browser that keeps your sessions and data on your machine, works with your own AI keys or Ollama, and supports both human-driven and agent-driven browsing. Do not adopt it if you need a cloud-hosted solution, if you rely on proprietary AI browser integrations, or if you must avoid the AGPL-3.0 license. Before adopting, verify that the Chromium fork stays updated with upstream security patches, test the agent's performance with your specific AI provider, and confirm that the login import works with your Chrome profile. The project is actively maintained with recent releases, but the dual-product structure (BrowserOS and BrowserOS neo) may confuse new users, so check the docs to see which product fits your workflow.
Community notes