Browser-Use: The Open-Source Repo Teaching AI Agents to Click

1. What Is Browser-Use: Stars, License, and Facts

Browser-Use (github.com/browser-use/browser-use) is an open-source Python framework that makes websites accessible to AI agents. Instead of brittle CSS selectors, the agent observes the page the way assistive tech does, decides the next action with an LLM, and executes it through a real Chromium browser driven by Playwright.

There are two ways to consume it. The open-source agent runs on your own machine: free, model-agnostic (OpenAI, Anthropic, Google, Ollama, or their hosted ChatBrowserUse models), with full control of code and prompts. The hosted cloud adds stealth browsers, proxy rotation, CAPTCHA handling, and 1,000+ integrations for production. The library stays MIT either way.

2. Architecture in 4 Steps

You do not need to read the whole codebase to use it well. The agent loop is just four stages repeating until the task is done or the step budget runs out.

3. Quickstart: Your First Agent in 5 Minutes

Every command below is copied from the current README and the official human quickstart docs, then kept minimal. Python 3.11 or newer required; 3.12 recommended with uv.

Step A — Install the package and the browser

pip install uv
uv venv --python 3.12
source .venv/bin/activate
uv pip install browser-use
uvx browser-use install

No uv? pip install browser-use also works. The last line downloads Chromium. Faster path: uvx browser-use init --template default scaffolds a working agent file.

Step B — Add your LLM key to .env

# .env — pick the provider you will actually use
BROWSER_USE_API_KEY=your-key
# GOOGLE_API_KEY=your-key
# OPENAI_API_KEY=your-key
# ANTHROPIC_API_KEY=your-key

One key is enough. ChatBrowserUse is tuned for browser tasks and reaches most providers with a single key; local models work via Ollama for fully offline runs.

Step C — Run your first agent

from browser_use import Agent, ChatBrowserUse
from dotenv import load_dotenv
import asyncio

load_dotenv()

async def main():
    llm = ChatBrowserUse()
    task = "Find the number 1 post on Show HN"
    agent = Agent(task=task, llm=llm)
    await agent.run()

if __name__ == "__main__":
    asyncio.run(main())

Swap the LLM in one line: ChatGoogle(model=”gemini-flash-latest”), ChatOpenAI(model=”gpt-4.1-mini”), or ChatAnthropic(model=”claude-sonnet-4-0”). Same Agent, same loop.

4. Real Use Cases That Justify the Hype

Browser agents shine where APIs do not exist, change without notice, or sit behind interfaces built for humans. These are the patterns I see surviving contact with production.

5. When NOT to Use Browser-Use

Honest section, because the fastest way to hate agents is pointing them at the wrong job. Browser automation costs latency, tokens, and flakiness — spend that budget where it pays.

Conclusion

Browser-Use earned its 112k stars by solving an unglamorous problem well: letting code use the web as it exists, not as we wish its APIs were. The loop is simple — observe, decide, act, verify — and the quickstart genuinely runs in five minutes.

My verdict: learn it if you automate knowledge work. Start with one boring, repeatable task, keep the scope tight, log the history, and graduate to cloud browsers only when the ROI is obvious.