On BrowserOS, we provide first-class support to bring your own LLMs either local models or via API keys and run the agent entirely on the client side, so your data stays on your machine!
Today we're launching filesystem access... just like Claude Cowork, our browser agent can read files, write files, run shell commands! But honestly, we didn't plan for this. It turns out the privacy decision we made 9 months ago accidentally positioned us for this moment.
--- The architectural bet we made 9 months ago Unlike other AI browsers (ChatGPT Atlas, Perplexity Comet) where the agent loop runs server-side, we decided early on to run our agent entirely on your machine (client side).
But building everything on the client side wasn't smooth.
We initially built our agent loop inside a Chrome extension. But we kept hitting walls:
1) JS (background service worker) is single-threaded, so we couldn't start multiple agents in parallel.
2) Not having access to a NodeJS-like runtime meant we couldn't use many great npm packages (Vercel AI SDK, Anthropic's MCP SDK, etc)
3) And finally, there was no good way to expose our agent and tools as an API
So we made the hard decision 2 months ago to throw away everything we built and start from scratch.
In the new architecture, we went with a sidecar approach. We put our agent loop in a standalone Bun binary and ship it alongside our Chromium binary. We also decided not to rewrite our own agent loop, but borrowed gemini-cli's loop with some tweaks! We wrote a neat adapter to translate between Gemini format and Vercel AI SDK format. You can look at our entire codebase here: https://git.new/browseros-agent
--- How this helped build filesystem access When Claude Cowork launched, we realized something: because Atlas and Comet run their agent loop server-side, there's no good way for their agent to access your files without uploading them to the server first.
But our agent was already local. Adding filesystem access meant just... opening the door (with your permissions ofc). Our agent can now read and write files just like Claude Code. No uploads, no cloud storage, no sync.
--- What you can actually do today
a) Organize files in my desktop folder https://youtu.be/NOZ7xjto6Uc
b) Open top 5 HN links, extract the details and write summary into a HTML file https://youtu.be/uXvqs_TCmMQ
--- Where we are now If you haven't tried us since the last Show HN, give us another shot. The new architecture unlocked a ton of new features, and we've grown to 8.5K GitHub stars and 100K+ downloads:
c) You can now build more reliable workflows using n8n-like graph https://youtu.be/H_bFfWIevSY
d) You can also use BrowserOS as an MCP server in Cursor or Claude Code https://youtu.be/5nevh00lckM
e) You can also schedule repetitive tasks!
--- Why we think browser is the right platform We are very bullish on browser being the right platform for a Claude Cowork like agent. Browser is the most commonly used app by knowledge workers (emails, docs, spreadsheets, research, etc). And it seems like even Anthropic recognizes this -- for Claude Cowork, they have janky integration with browser via a chrome extension. But owning the entire stack allows us to provide a much smoother experience. It also lets us build differentiated features that wouldn't be possible otherwise. One example: Browser ACLs.
Agents can do dumb or destructive things, so we're adding browser-level guardrails (think IAM for agents): "role(agent): can never click buy" or "role(agent): read-only access on my bank's homepage." We have a prototype already—curious to hear your take on this and the overall thesis.
We’ll be in the comments. Thanks for reading!
GitHub: https://git.new/browseros Download: https://browseros.com (available for Mac, Windows, Linux!)