Browser Use
Python agent library that lets an LLM operate a real Chrome browser to finish web tasks.
About
Give a language model a task in plain English and this Python package turns it into clicks, typing, scrolling, and page reads inside Chrome. On each loop the agent sends the model a trimmed view of the page's interactive elements, gets back the next action, and keeps going until the goal is reached, returning a history object that holds the final answer and every intermediate step. Recent versions drive the browser directly over the DevTools Protocol instead of Playwright, and the package bundles a CLI, an MCP server, a decorator for registering your own tools, and chat wrappers for OpenAI, Anthropic, Google, Groq, AWS Bedrock, and Ollama. Setup needs Python 3.11 or newer, usually through uv, plus a key for whichever model provider you choose; pairing it with Ollama keeps the whole loop on your own machine. The library is MIT licensed and safe for commercial projects, while the company's hosted browsers, its BU2 model, and its agent API are separate paid services you can ignore. A practical fit for developers automating form filling, research chores, or QA flows on sites that offer no API.
Should you use Browser Use?
Pick it when
Use Browser Use when a Python agent has to operate real websites, logging in, filling forms, and extracting data from pages that have no API, and you are comfortable letting the LLM decide each step.
Look elsewhere when
Fully agent-driven browsing spends LLM calls on every step and can behave differently on repeated jobs. For flows you run daily, Stagehand's cached code-plus-AI steps are cheaper; for plain page-to-Markdown extraction, Crawl4AI needs no agent.
Alternatives to Browser Use
- Stagehand
Mixes deterministic Playwright code with AI steps and caches repeated actions; more predictable and cheaper on recurring runs, more code.
- Skyvern
Vision-based agents plus a no-code workflow builder and 2FA logins; stronger for form-heavy RPA, but AGPL-3.0 licensed.
- Crawl4AI
Turns pages into Markdown with no agent in the loop; far cheaper for reading content, but cannot click through interactive flows.
- UI-TARS Desktop
Drives the whole desktop as well as the browser through vision grounding; broader reach, heavier and less scriptable from Python.
Reviews (0)
Leave a Review
No reviews yet. Be the first to review!
Details
- Category
- AI Agents & Orchestration
- Price
- Free
- Platform
- Local/Desktop
- Difficulty
- Easy (2/5)
- License
- MIT
- Added
- Apr 3, 2026