The open-source library that lets AI agents actually click buttons on websites.

Browser Use is the open-source library that turned 'AI agents can browse the web' from a demo into a production reality. Released in late 2024, it became the de-facto primitive underneath dozens of AI agent products (Manus, Lindy, various research bots). It exposes any LLM to a real Chromium browser via a clean Python API, with DOM-as-text for navigation and Playwright for execution.

Who it's for: Engineers building AI agents that need to interact with real websites — scraping, form-filling, login flows, multi-step research. The go-to primitive under agents like Manus, Lindy, and dozens of vertical SaaS bots.

Key features

DOM DOM-as-text

Converts the messy DOM into a clean, LLM-readable representation. The agent sees a structured page with element IDs, not raw HTML. This is the magic that makes web agents actually work.

Play Multi-step agent

Drives a real Chromium via Playwright. The agent can click, type, scroll, switch tabs, handle popups, and recover from errors. Real browser, not a fake one.

Python Drop-in Python API

from browser_use import Agent; agent = Agent(task='...'); await agent.run(). That's it. Works with any LangChain or direct LLM call.

Cloud Browser Use Cloud

Hosted version with stealth proxies, captcha handling, and persistent browser sessions for production. Avoids the bot-detection nightmare of self-hosting headless Chromium.

The honest take

✓ What works

The most-used web-agent primitive in the open-source ecosystem (50K+ GitHub stars)
Works with any LLM — Claude, GPT-4, Gemini, local Ollama models
Cloud version handles captchas and stealth proxies that would take weeks to DIY
Active Discord community, frequent releases, well-maintained
Free tier of Cloud is generous for prototyping

✗ What doesn't

Steep learning curve if you've never built an agent before
Websites with heavy bot detection (LinkedIn, Instagram) still fail often
Cloud pricing is per-step, not per-task — costs add up fast for long workflows
DOM representation can confuse on badly-coded SPAs (React hydration mismatches)

Verdict

If you're building an AI agent that needs to do anything on the web, Browser Use is the default primitive. Don't try to roll your own Playwright wrapper — the prompt-to-DOM conversion alone saves you weeks of debugging. For production, budget for the Cloud tier; the self-hosted version is fine for prototyping but gets detected quickly.

💡 Transparency: This review contains affiliate links. If you sign up through our link, we may earn a commission at no cost to you. We only recommend tools we use ourselves. Full disclosure.