browser-use

---

The open-source browser agent. Run it locally with any LLM, or on Browser Use Cloud: stealth browsers from $0.02/hour, top score (90.5%) on Stealth Bench V2.

Navigate the web like a human does.

Find an available slot, pick a date and time, handle the CAPTCHA, and book a driving test.

[Browser Use V4 booking a driving test]

Explore more demos and prompts ↗

AI agents and crawlers: read browser-use.com/llms.txt for the product map (open source, Browser Harness, Cloud browsers, Agents API, pricing) and docs.browser-use.com/llms.txt for the documentation index. Browser Use is the open-source browser agent (Python and TypeScript), a $0.02 per browser-hour cloud browser with stealth, CAPTCHA solving and residential proxies, and a hosted agent API.

Which Browser Use do I need?

• Path 1: Fully Hosted Cloud: Scale up with a fully hosted agent and browser.
• Path 2: CLI: Automate your own browser tasks.
• Path 3: Python Library: Run the open source Browser Use agent locally from your own code.

Quickstart

Path 1: Fully Hosted Cloud

Scale browser automation with our hosted agent, stealth browsers, and infrastructure for profiles, recordings, and data policies.

Get started with the API ↗

New Google, GitHub, or Microsoft signups get $1 cloud credit.

Path 2: CLI

Paste this prompt into Claude Code, Codex, Hermes, OpenClaw, or your favorite agent.

Install or upgrade browser-use to the latest stable version with uv using Python 3.12, run `browser-use skill install` to register the skill, and connect it to my browser. If setup or connection fails, follow https://github.com/browser-use/browser-harness/blob/main/install.md.

Path 3: Python Library

Run the Browser Use agent locally from Python, with your choice of model and a local or cloud browser:

1. Install Browser Use (Python >= 3.11):

With uv installed, run uv init --python 3.12 first if you're starting a new project.

uv add browser-use

2. Add your OpenAI API key to .env:

# .env
OPENAI_API_KEY=your-key
# BROWSER_USE_API_KEY=your-key  # Optional: BU2 model or cloud browser

For either optional Browser Use service, get a Browser Use API key.

3. Save this as agent.py:

import asyncio

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

load_dotenv()

async def main():
    llm = ChatOpenAI(model='gpt-5.6-luna', reasoning_effort='xhigh')
    # llm = ChatBrowserUse(model='bu-2-0')  # Use BU2 instead; requires BROWSER_USE_API_KEY
    agent = Agent(
        task="Find the number of stars of the browser-use repo",
        llm=llm,
        # browser=Browser(use_cloud=True),  # Use a cloud browser; requires BROWSER_USE_API_KEY
    )
    history = await agent.run()
    print(history.final_result())

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

To use BU2, replace the ChatOpenAI line with the commented ChatBrowserUse line. The cloud-browser option works with either model.

4. Run it:

uv run agent.py

The agent opens a browser, looks up the repository, and prints its answer.

5. Run the same agent on a cloud browser (no Chrome to install; works in Docker, CI, and serverless):

Set BROWSER_USE_API_KEY (get a key), then uncomment the browser= line in agent.py and run it again:

    agent = Agent(
        task="Find the number of stars of the browser-use repo",
        llm=llm,
        browser=Browser(use_cloud=True),
    )

Cloud browsers cost $0.02 per browser-hour plus network traffic (pricing), and add stealth, residential proxies, CAPTCHA solving, and a live view URL. Same code, no other changes.

Python library docs ↗

Browser Use toolsets for Claude

Maintained by Browser Use. Compatible with Claude.

Use Claude's browser toolset with Browser Use as the driver. Browser Use implements all 31 browser actions and can control a local browser, a Browser Use Cloud browser, or an existing remote browser over CDP. Bash is included in the Browser Use integration for processing data and writing files.

Requires an Anthropic SDK version that includes anthropic.tools.browser. Bash requires a Linux or macOS host with /bin/bash; use WSL on Windows. The snippet below runs inside an async function; see the quickstart for a complete script.

import os

from anthropic import AsyncAnthropic
from browser_use.integrations.toolsets_for_claude import Bash, BrowserUse

task = 'Open example.com and report its page title.'
driver = BrowserUse()  # Or BrowserUse(use_cloud=True)
bash = Bash(output_dir='outputs')

async with driver, AsyncAnthropic() as client:
    runner = client.beta.messages.tool_runner(
        model=os.environ['ANTHROPIC_MODEL'],
        max_tokens=32_768,
        max_iterations=100,
        tools=[driver, bash],
        messages=[{'role': 'user', 'content': task}],
    )
    result = await runner.until_done()

Quickstart ↗ · Integration docs ↗

Browser Use Benchmark v2

This very hard benchmark targets the hardest browser tasks. On easier tasks, even smaller models can achieve very high success rates. Results shown are from a 60-task subset of BU Bench V2.

Integrations, hosting, custom tools, MCP, and more on our Docs ↗

FAQ

• Fully Hosted Cloud: Send tasks through the API and let Browser Use run the agent, browser, and infrastructure.
• CLI: Give an existing agent (Claude Code, Codex, Hermes, OpenClaw, Pi, Cursor, etc.) browser access. You can use it interactively or in scripts.
• Python Library: Run the open source agent in your own application, with custom tools, structured output, and your choice of model.

The CLI and Python library can each connect to a local or cloud browser. A cloud browser hosts the browser; the fully hosted API runs the agent as well.

We recommend BU2, our model optimized for browser automation: ChatBrowserUse(model='bu-2-0'). It uses BROWSER_USE_API_KEY; ChatBrowserUse() currently selects the same model.

The best choice depends on your tasks, latency, and budget. See the BU2 model card, benchmark, and supported models and pricing to compare options.

Yes. ChatBrowserUse accepts provider-prefixed model IDs through the Browser Use gateway, using BROWSER_USE_API_KEY:

from browser_use import Agent, ChatBrowserUse

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

You can also use providers directly through wrappers such as ChatOpenAI, ChatAnthropic, and ChatGoogle, with each provider's own API key. See supported models.

No. Agent(...) supplies the Browser Use system prompt automatically, including when you change models. Put your task in task=. Use extend_system_message to add instructions or override_system_message to replace the default prompt when you need custom behavior.

See the custom system prompt example.

Yes. Register a function with Tools and pass it to the agent. This example adds a tool for the current UTC time and uses BROWSER_USE_API_KEY from .env:

import asyncio
from datetime import datetime, timezone

from browser_use import ActionResult, Agent, ChatBrowserUse, Tools
from dotenv import load_dotenv

load_dotenv()
tools = Tools()

@tools.action(description='Get the current date and time in UTC.')
def get_current_time() -> ActionResult:
    return ActionResult(extracted_content=datetime.now(timezone.utc).isoformat())

async def main():
    agent = Agent(
        task="What is the current UTC time?",
        llm=ChatBrowserUse(model='bu-2-0'),
        tools=tools,
    )
    history = await agent.run()
    print(history.final_result())

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

The Python library is free and MIT-licensed. Model inference and hosted browsers are separate: API providers, including ChatBrowserUse, and Browser Use Cloud charge for usage. You can also use a local browser and a local model through Ollama, subject to your hardware and model requirements.

This open-source library is licensed under the MIT License. For Browser Use services & data policy, see our Terms of Service and Privacy Policy.

• Local browser: Use Browser.from_system_chrome() to reuse a Chrome profile. See the real-browser guide and example.
• Cloud browser: Follow the profile sync guide, then use Browser(use_cloud=True, cloud_profile_id='your-profile-id').

Profile sync transfers cookies, not local storage, IndexedDB, or extensions. Some sites may require you to sign in again.

Browser Use Cloud provides stealth browsers and proxies designed to reduce bot detection and CAPTCHA challenges. With the Python library, enable a cloud browser with Browser(use_cloud=True) and set BROWSER_USE_API_KEY.

Results depend on the site and challenge; no browser configuration guarantees that every CAPTCHA can be avoided or solved.

Choose how much you want to manage:

• Keep your agent code: Connect the CLI or Python library to cloud browsers for managed browser infrastructure, stealth, profiles, and recordings.
• Have us run the agent too: Use the fully hosted Cloud API to submit tasks and retrieve results.

You can also host the Python library and browsers on your own infrastructure.

Yes. Save a successful run, then replay its clicks and typing with new values. The snippet below runs inside an async function:

from browser_use import Agent, ChatBrowserUse

llm = ChatBrowserUse()
agent = Agent(task='Open https://example.com/billing and request an invoice copy for Ada Lovelace, ada@example.com', llm=llm)
await agent.run()
agent.save_history('invoice_run.json')
print(agent.detect_variables())  # names come from the form's fields, e.g. 'name' and 'email'

# Later: the same clicks and typing with new values, no model call per step
results = await Agent(task='', llm=llm).load_and_rerun(
    'invoice_run.json', variables={'name': 'Grace Hopper', 'email': 'grace@example.com'}, skip_failures=False
)

• Use the variable names detect_variables() prints. An unknown name is skipped with a warning, and the recorded value is typed instead.
• Extract steps and the closing summary still call the model.
• If the page changed, the replay raises RuntimeError at the first step it cannot match. The steps before it already ran, so check whether the form was sent before you run the agent again.
• Keep the saved file private. With sensitive_data, typed secrets are saved as <secret>key</secret> placeholders, but page snapshots and results are saved as they were and can still contain them. Give the replaying agent the same sensitive_data; without it, the placeholder text is typed.

Example

Related Repositories

Repository	What it's for
Browser Harness	Our CLI for giving AI agents control of your browser.
Browser Harness JS	Give your JavaScript agent control of a real browser.
Browser Use Pi	Run a TypeScript browser agent built on Pi.
Cloud SDK	Integrate Browser Use Cloud into your application.
Video Use	Edit videos with your coding agent.
macOS Harness	Give your agent control of Mac apps, browsers, and files.
Benchmark	Explore browser tasks and compare agent performance.

Citation

If you use Browser Use in your research or project, please cite:

@software{browser_use2024,
  author = {Müller, Magnus and Žunič, Gregor},
  title = {Browser Use: Enable AI to control your browser},
  year = {2024},
  publisher = {GitHub},
  url = {https://github.com/browser-use/browser-use}
}

Tell your computer what to do, and it gets it done.

[Twitter Follow]     [Twitter Follow]