Agent Framework
browser-use/browser-use
browser-use/browser-use· Python
Python library that lets any LLM drive a real browser via Playwright to complete web tasks.
GitHub stats
- Stars
- 91,196
- Forks
- 10,389
- Watchers
- 429
- Open issues
- 239
meta
- License
- MIT
- Primary language
- Python
- Last commit
- 2026-04-26
- Stats fetched at
- 2026-04-29
browser-use is a Python agent framework that exposes web pages to LLMs as structured, clickable elements, then lets the model plan and execute multi-step tasks (form filling, scraping, booking, research) through Playwright. It works with OpenAI, Anthropic, Gemini, local models via Ollama, and anything LangChain-compatible. Install with `pip install browser-use`, point it at a model and a goal in natural language, and it drives a real Chromium session end-to-end.
Editor's verdict
The default pick if you want an LLM to actually use a browser today — far less brittle than rolling your own Playwright + screenshots loop, and the DOM-extraction approach gives better accuracy than pure vision agents. Trade-off: token costs scale with page complexity, and it still fails on heavy anti-bot sites or unusual UIs. Skip it if your task is reachable via API or a simple `requests` script — that's always cheaper and more reliable than driving a browser.