Browser Use Documentation
by Browser Use
Point an LLM at a real browser and let it click, type and finish the task.
Overview
Browser Use is an MIT-licensed Python library that makes websites usable by AI agents: it identifies the interactive elements on a page, combines that structure with visual understanding, and loops the model through navigate-click-type-extract actions over Chromium via CDP until a natural-language task is complete. The documentation splits into an open-source track and a hosted cloud track. The open-source quickstart covers installation with uv on Python 3.12, installing Chromium with 'uvx browser-use install', and four working provider examples — ChatBrowserUse, a model tuned specifically for browser control, plus Google Gemini, OpenAI and Anthropic Claude — before branching into supported models, browser customization, agent settings, structured output, custom functions and tools, sensitive-data handling and observability. The cloud track documents hosted agents and the v4 API, cloud browsers reachable by CDP URL, stealth mode, proxies, live preview and session recording, persistent authentication profiles with local/cloud cookie sync and 2FA, workspaces and file handling, human-in-the-loop approval, scripts, and interop with Playwright, Puppeteer and Selenium; the docs recommend v4 for long, complex workflows and note that v2 suits cases where low cost and predictable speed matter more than accuracy. The library carries roughly 107.8k GitHub stars, is built by a YC W25 team in Zurich and San Francisco, and claims first place on the Odysseys leaderboard at 87.4% average across 200 long-horizon web tasks.
At a Glance
- Topic
- Agentic
- Level
- Intermediate
- Format
- Documentation
- Cost
- Freemium
- Duration
- ~2-3 hours to get an agent running and customized, then reference
- Provider
- Browser Use
- Hands-on
- Yes — code/exercises
- Certificate
- None
What You’ll Learn
- ✓Install and run a browser agent from a plain natural-language task
- ✓Point the agent at Gemini, OpenAI, Claude or the tuned ChatBrowserUse model
- ✓Get typed, structured output back instead of scraped free text
- ✓Register custom functions so the agent can call into your own code
- ✓Handle logins, cookies and 2FA without putting secrets into the prompt
- ✓Customize browser launch, proxies and stealth settings for real-world sites
- ✓Decide when to self-host the library versus using hosted cloud browsers
Highlights
- •Roughly 107.8k GitHub stars and MIT-licensed — the default open-source answer to 'let an agent use the web'
- •Works against any site rather than only ones with an API: element extraction plus vision, no per-site scraper to maintain
- •Ships a model tuned for browser control (ChatBrowserUse) alongside generic Gemini, OpenAI and Claude support
- •Claims #1 on the Odysseys leaderboard at 87.4% over 200 long-horizon web tasks, with an open benchmark repo you can check
- •The docs are candid about the trade-off — v4 for accuracy on complex workflows, v2 when cost and predictable speed matter more
Who It’s For
Best For
- ✓Engineers automating web workflows on sites that have no API
- ✓QA teams that want agent-driven end-to-end browser testing
- ✓Data teams doing agentic scraping, monitoring or form submission
- ✓Agent builders who need a web-action tool alongside search and RAG
Prerequisites
- •Python 3.11 or newer (the quickstart uses 3.12) and comfort with async code
- •An API key for at least one LLM provider, since the agent spends tokens on every step
- •Basic browser-automation intuition — selectors, page loads, timeouts — for debugging failed runs
FAQ
What is Browser Use Documentation?
Official documentation for Browser Use, the MIT-licensed Python library that lets an LLM drive a real Chromium browser to complete natural-language tasks — clicking, typing, filling forms and extracting data on sites that expose no API. Written for engineers automating web workflows, doing agentic scraping, or building end-to-end QA. After working through it you can run a self-hosted browser agent against your own model, get typed structured output back, register custom tools, and handle logins and 2FA without leaking secrets into prompts.
Is Browser Use Documentation free?
Browser Use Documentation offers free content, with paid options for certificates or premium features.
What level is Browser Use Documentation for?
Browser Use Documentation is aimed at a intermediate audience. Recommended background: Python 3.11 or newer (the quickstart uses 3.12) and comfort with async code, An API key for at least one LLM provider, since the agent spends tokens on every step, Basic browser-automation intuition — selectors, page loads, timeouts — for debugging failed runs.
How long does Browser Use Documentation take?
Expect roughly ~2-3 hours to get an agent running and customized, then reference. Most learners work through it at their own pace.
What will I learn from Browser Use Documentation?
You'll learn: Install and run a browser agent from a plain natural-language task; Point the agent at Gemini, OpenAI, Claude or the tuned ChatBrowserUse model; Get typed, structured output back instead of scraped free text; Register custom functions so the agent can call into your own code; Handle logins, cookies and 2FA without putting secrets into the prompt; Customize browser launch, proxies and stealth settings for real-world sites; Decide when to self-host the library versus using hosted cloud browsers.
Topics
Sources
This page was written from 4 sources, 1 on domains other than docs.browser-use.com.