Browser Use - AI Agents That Automate Real Web Browsers

TL;DR

TL;DR: Browser Use is an open-source AI agent framework (99K GitHub stars, MIT license) that drives real browsers to automate web tasks. It works with any LLM and can be set up with a Python script, CLI tool, or Claude Code skill.

Source and Accuracy Notes

⚠️ This section is MANDATORY. All links must be verified from actual source, not guessed.

• Project page: browser-use.com
• Source repository: github.com/browser-use/browser-use
• License: MIT (verified via GitHub API license.spdx_id)
• HN launch thread: news.ycombinator.com/item?id=43173378
• Docs: docs.browser-use.com
• Stars: 99,361 | Default branch: main | Last pushed: 2026-06-15 (verified via GitHub API)

What Is Browser Use?

Browser Use is an open-source framework that lets AI agents control real browsers — not just fetch HTML or take screenshots, but actually interact with web pages like a human would: clicking buttons, filling forms, scrolling, and handling multiple domains in a single task.

The project’s stated goal: “Make websites accessible for AI agents.”

It started as a YC W25 startup and has grown to 99,361 GitHub stars, making it one of the most-starred browser automation tools in the AI agent space. The framework uses Playwright under the hood but wraps it in an LLM-friendly action space with recovery loops.

The current major version (0.13) introduced a Rust-powered core agent alongside the original Python API. The Rust agent is described as built for frontier models and handles browser control with persistent tool access and retry logic.

Key capabilities

• Multi-step web tasks: Form filling, grocery shopping, job applications, research scraping
• Any LLM: Works with OpenAI GPT, Anthropic Claude, Google Gemini, or Browser Use’s own optimized bu-* models via ChatBrowserUse
• Cloud or self-hosted: Open-source agent runs anywhere; cloud option adds stealth browsers, proxy rotation, and CAPTCHA handling
• CLI tool: Persistent browser sessions from the command line
• Claude Code skill: Native integration for AI-assisted browser automation inside Claude Code
• Template generator: uvx browser-use init --template default scaffolds a ready-to-run project

Setup Workflow

Prerequisites

• Python 3.11 or higher
• uv package manager (recommended) or pip
• An LLM API key (OpenAI, Anthropic, Google, or Browser Use Cloud)

Step 1: Install Browser Use

uv add "browser-use[core]"
# or: pip install "browser-use[core]"

The [core] extra installs the native Browser Use runtime for your platform.

Step 2: Get an API Key

For the easiest setup, get a key from Browser Use Cloud:

# .env
BROWSER_USE_API_KEY=your-key
# GOOGLE_API_KEY=your-key
# ANTHROPIC_API_KEY=your-key

With a BROWSER_USE_API_KEY, ChatBrowserUse routes requests to all major providers through a single key — no need to manage separate OpenAI/Anthropic/Google keys.

Step 3: Run Your First Agent

Option A — Python Script:

from browser_use.beta import Agent, BrowserProfile, ChatBrowserUse
import asyncio

async def main():
    agent = Agent(
        task="Find the number of stars of the browser-use repo",
        llm=ChatBrowserUse(model='openai/gpt-5.5'),
        browser_profile=BrowserProfile(
            headless=False,
            allowed_domains=["*.github.com"],
        ),
    )
    history = await agent.run()
    print(history.final_result())

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

Option B — Template Generator:

uvx browser-use init --template default

This creates a browser_use_default.py file with a working example. Available templates: default, advanced, tools.

Option C — CLI Tool:

browser-use open https://example.com    # Navigate to URL
browser-use state                       # See clickable elements
browser-use click 5                     # Click element by index
browser-use type "Hello"                # Type text
browser-use screenshot page.png         # Take screenshot
browser-use close                       # Close browser

The CLI keeps the browser running between commands for fast iteration.

Step 4: Integrate with Claude Code

For AI-assisted browser automation inside Claude Code:

mkdir -p ~/.claude/skills/browser-use
curl -o ~/.claude/skills/browser-use/SKILL.md \
  https://raw.githubusercontent.com/browser-use/browser-use/main/skills/browser-use/SKILL.md

Then direct Claude Code to Agents.md (the project provides a full agent prompt) and prompt naturally.

Deeper Analysis

Open-Source vs. Cloud

The README is explicit about the tradeoff:

Open-source agent is the right choice when you need custom tools, code-level integration, or want to self-host entirely on your own infrastructure. You can pair it with Browser Use’s cloud browsers for stealth and scaling, or run everything locally.

Browser Use Cloud (fully-hosted) is recommended for complex tasks and production use. It provides proxy rotation, stealth browser fingerprinting, CAPTCHA solving, persistent filesystem and memory, and over 1,000 integrations (Gmail, Slack, Notion, and more).

A benchmark on the README (browser-use/benchmark) shows their cloud agent outperforming open-source on accuracy across 100 real-world browser tasks, with bu-2-0 their own optimized model leading.

Architecture

Browser Use 0.13+ has two stacks:

1. Python API (from browser_use import Agent) — original agent, still supported for existing users
2. Rust-powered beta (from browser_use.beta import Agent) — new core, described as built for current frontier models with a Rust browser harness

Both use the same underlying Playwright browser automation, but the beta agent has improved recovery loops inspired by coding agents.

Model Support

ChatBrowserUse accepts provider-prefixed model IDs:

from browser_use import Agent, ChatBrowserUse

llm = ChatBrowserUse(model='anthropic/claude-sonnet-4-6')  # or 'openai/gpt-5.5', 'google/gemini-3-pro'
agent = Agent(task='...', llm=llm)

The project recommends their own bu-* models for best speed and accuracy. For local models, Ollama is supported.

Practical Evaluation Checklist

• GitHub stars: 99,361 (verified)
• License: MIT (verified)
• Last pushed: 2026-06-15 (verified)
• Python 3.11+ required
• Works with OpenAI, Anthropic, Google, or local Ollama
• CLI tool ships with the package
• Claude Code skill available
• Cloud option adds stealth, proxy rotation, CAPTCHA solving
• MIT licensed (verified)

Security Notes

• Browser automation runs real browser sessions — use BrowserProfile(allowed_domains=[...]) to restrict which sites the agent can access
• API keys in .env files only; never hardcode in scripts
• Browser Use Cloud has separate Terms of Service and Privacy Policy — the open-source library is MIT, but the cloud service has its own data policy
• The CLI browser-use command keeps a persistent browser open between commands; close it when done with browser-use close

FAQ

Q: Do I need a separate API key for each provider? A: No. ChatBrowserUse with a BROWSER_USE_API_KEY routes to all major providers through one key. You can also use separate provider keys directly.

Q: Can I run this locally without any cloud service? A: Yes. The open-source agent runs entirely on your own infrastructure with your own LLM API keys or a local Ollama instance.

Q: What is the difference between the Python agent and the Rust beta agent? A: The Rust agent (imported from browser_use.beta) is the new preview core built for frontier models with improved recovery loops. Existing users can keep using from browser_use import Agent. Both drive Playwright browsers.

Q: Does it handle CAPTCHAs or anti-bot detection? A: The open-source agent uses standard Playwright, which can be detected by sophisticated bot detection. Browser Use Cloud provides stealth browsers with proxy rotation and fingerprinting designed to avoid CAPTCHAs.

Conclusion

Browser Use fills the gap between simple HTTP scrapers and full browser emulation. By giving AI agents a real browser action space, it can handle web tasks that require JavaScript rendering, session handling, and interactive forms — tasks that break traditional scraping tools.

The open-source MIT package is substantial (99K stars, active development) and the combination of Python API + CLI + Claude Code skill covers most workflows. The cloud option is where Browser Use monetizes: for teams that need production-scale browser automation without managing infrastructure.

If you need AI agents that can actually use the web like a human would, Browser Use is the most production-ready open-source option right now.

• browser-use.com
• GitHub repository
• Documentation
• Browser Use Cloud