browser-use

name	browser-use
domain	ai-ml
description	AI agent that autonomously interacts with the web via Chromium/CDP. Makes websites accessible for LLMs.
version	1.1.0
category	automation

browser-use Skill

AI agent that autonomously interacts with the web via Chromium/CDP. Makes websites accessible for LLMs.

⚠️ CRITICAL: Google API Key Required

browser-use's Agent runs an AI agent that makes decisions about web interactions. It requires a Google API key — without it, you get Missing required environment variable: GOOGLE_API_KEY or the agent fails to initialize.

If no Google API key is available, use the built-in Hermes browser tools instead (browser_navigate, browser_click, browser_type, browser_snapshot). These are separate from browser-use and work out of the box.

browser-use vs Built-in Browser Tools

Feature	browser-use Agent	Built-in Hermes browser tools
AI-driven	✅ Yes (autonomous agent)	❌ No (scripted, deterministic)
API key needed	✅ Google API key required	❌ None
Browser engine	Chromium (CDP protocol)	WebKit
Firefox support	❌ Chromium only	✅ WebKit (cross-platform)
Cloudflare behavior	Blocked by Cloudflare (Chromium fingerprint)	Sometimes allowed (different fingerprint)
Best for	Complex multi-step tasks needing AI reasoning	Simple navigation, form fills, scraping
Headless/silent	⚠️ Requires Google key + cloud for best stealth	✅ Fully headless by default

Installation

cd /Users/eric_yiru/Desktop/Github/browser-use
uv venv --python 3.11
source .venv/bin/activate
uv pip install -e .
uv pip install playwright
python -m playwright install chromium

macOS arm64 Chromium Fix (IMPORTANT — playwright installs to wrong path on Apple Silicon)

Playwright on macOS arm64 installs Chromium to chrome-mac-arm64/ but browser-use looks for chrome-mac/. Fix with symlinks:

ls ~/Library/Caches/ms-playwright/
PLAYWRIGHT_CHROMIUM=~/Library/Caches/ms-playwright/chromium-1217  # or whichever version exists

ln -sfn chrome-mac-arm64 "$PLAYWRIGHT_CHROMIUM/chrome-mac"
ln -sfn "Google Chrome for Testing.app" "$PLAYWRIGHT_CHROMIUM/chrome-mac/Chromium.app"

# Verify
ls "$PLAYWRIGHT_CHROMIUM/chrome-mac/"
# Should show: Chromium.app -> Google Chrome for Testing.app, etc.

Without this fix: RuntimeError: Error getting browser path: No local browser path found after: uvx playwright install chromium

Environment Setup

# Required — browser-use Agent needs a Google API key (or one of the alternatives)
GOOGLE_API_KEY=your_google_api_key
# Or: BROWSER_USE_API_KEY, ANTHROPIC_API_KEY, OPENAI_API_KEY

Basic Usage Pattern

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

load_dotenv()

async def main():
    browser = Browser()  # or Browser(use_cloud=True) for stealth cloud browser
    agent = Agent(
        task="Find the number of stars of the browser-use repo",
        llm=ChatBrowserUse(),
        browser=browser,
    )
    await agent.run()

asyncio.run(main())

Key Imports (from browser_use/__init__.py)

• Agent, Browser, BrowserSession, BrowserProfile
• ChatBrowserUse, ChatGoogle, ChatAnthropic, ChatOpenAI, ChatGroq, ChatLiteLLM, ChatMistral, ChatOllama, ChatAzureOpenAI, ChatVercel
• Tools, Controller
• DomService, SystemPrompt
• ActionModel, ActionResult, AgentHistoryList

Cloud/Sandbox Production Pattern

from browser_use import Browser, sandbox, ChatBrowserUse
from browser_use.agent.service import Agent

@sandbox(cloud_profile_id='your-profile-id')
async def production_task(browser: Browser):
    agent = Agent(task="Your authenticated task", browser=browser, llm=ChatBrowserUse())
    await agent.run()

Key Files

• Main package: /Users/eric_yiru/Desktop/Github/browser-use/browser_use/
• Agent service: browser_use/agent/service.py
• Agent views/models: browser_use/agent/views.py
• LLM adapters: browser_use/llm/ (browser_use/, anthropic/, google/, openai/, etc.)
• Examples: examples/models/ directory