browser use github
If you came here looking for the repository, it is github.com/browser-use/browser-use - MIT licensed, Python, 114,000 stars when read on 2026-09-13, described by its own authors as "Agents that use the browser." Go there; that is the primary source and this page does not try to replace it.
What this page adds is the part a repository landing page does not tell you: what you are agreeing to when you clone it, which of its three shipping paths you are actually choosing, and which problems its configuration surface can and cannot reach. Written by people who maintain a competing project, which is disclosed here rather than at the bottom.
Requirements, from the README: Python 3.11 or newer with 3.12 recommended,
and an LLM API key. The default setup expects an OpenAI key; the project also
supports Anthropic and Google models. Installation is one line, uv add browser-use, which tells you something useful on its own: the project has
standardised on uv rather than bare pip, so a tutorial that starts with pip install is working from an older version of the instructions.
Three shipping paths, and they are not equivalent. The README presents the
project as a hosted cloud service, a CLI, and a Python library. The library is
the MIT-licensed thing in the repository; Browser Use Cloud is a commercial
product at cloud.browser-use.com with its own API key. Which one you are
reading about matters, because most tutorials blur them, and the
free-tier question is a different answer
for each.
Topics the project claims for itself: ai-agents, ai-tools,
browser-automation, browser-use, llm, playwright, python. That playwright
entry is the load-bearing one for anything to do with detection, and the next
section is about it.
One number this wiki can offer that the repository page cannot, because it requires having looked twice: this project recorded 112,000 stars on 2026-09-03 and 114,000 on 2026-09-10 - roughly two thousand in a week, from our own dated readings a week apart. For a dependency decision that is the useful shape of the signal rather than the absolute: this is not a project that is quietly stalling, and the ecosystem risk is the opposite one, which is a fast-moving surface you will be upgrading against.
The README is candid in a way worth quoting, because a lot of derivative content is not: it states that no browser configuration guarantees that every captcha can be avoided or solved, and that profile sync transfers cookies but not local storage or extensions.
That first sentence is the honest version of the thing people arrive at this repo hoping for, and it matches what we found reading the project's own configuration code rather than its documentation. The detailed reading is on its own page here - what browser-use configuration can and cannot change - and the short version is that the levers it exposes are real (a real Chrome binary instead of the bundled build, a user data directory with actual history) while the layer underneath is a Chromium-family browser driven over CDP, which no setting in the repository reaches.
That is not a criticism of the project. It is a scope statement, and it is the same scope statement that applies to every agent framework in this category: the agent is the part they built, the browser is a dependency they inherited. Why an agent gets blocked separates those two layers properly, and it usually turns out the address and the pacing matter more than either.
- You want the repo. The link is in the first paragraph. Nothing below it is a substitute.
- You want to know if it fits your problem. Choosing an AI browser agent is the decision framework across the field.
- You are on it and something is being blocked. browser-use getting blocked, which reads the configuration surface rather than guessing at it.
- You want to know what else exists. browser-use alternatives, which opens by saying browser-use is good and means it.
- You want the whole open-source layer, not just this project. Open-source agentic browsers.
Where is the browser-use GitHub repo? github.com/browser-use/browser-use.
Is browser-use free? The MIT-licensed library in the repository is. Browser Use Cloud is a separate commercial product, and your model tokens are billed by whichever provider you point it at either way.
What Python version does browser-use need? 3.11 or newer, with 3.12 recommended by the README.
Does browser-use need an API key? For a model, yes. The default path expects an OpenAI key; Anthropic and Google are supported. A separate Browser Use API key is only for the cloud product.
Does browser-use solve captchas? Its own README says no browser configuration guarantees that every captcha can be avoided or solved, which is the accurate answer and more honest than most pages written about it. Can an AI agent solve a captcha is the category-level version.
Is browser-use maintained? Actively, on the evidence of roughly two thousand stars added in the week between our two readings, plus its release tempo.
- github.com/browser-use/browser-use, retrieved 2026-09-10: licence, star count, description, topics, requirements, install command, the three shipping paths, and the captcha and profile-sync caveats quoted above.
- This wiki's own reading of the project's configuration code, written up in browser-use getting blocked.
- Our own dated star readings, 2026-09-03 and 2026-09-10, recorded a week apart in this wiki.
Written while maintaining invisible_playwright_mcp, which competes with the project this page is about. That is why the repository's own link is the first thing on the page and why the section on what its configuration cannot reach ends by saying the same limit applies to every framework in the category, ours included.