Browser Use

0.12.6 · active · verified Thu Apr 09

Browser Use is a Python library designed to empower AI agents with the ability to navigate and interact with web browsers programmatically. It facilitates tasks such as clicking buttons, filling forms, and scraping data, effectively allowing AI to perform web-based actions autonomously. The library is actively maintained with frequent updates.

Warnings

breaking Starting with version `0.12.5`, `litellm` was removed as a core dependency due to a supply chain attack. If you rely on `ChatLiteLLM` or other `litellm` features, you must install `litellm` separately (`pip install litellm`).
Fix: If `litellm` functionality is required, install it explicitly: `pip install litellm`. Ensure your `litellm` version is not `1.82.7` or `1.82.8`.
breaking Version `0.12.3` introduced Browser Use CLI 2.0, which switched from Playwright to direct Chrome DevTools Protocol (CDP). This change provides faster performance but might alter underlying browser interaction behavior or require adjustments if previous Playwright-specific assumptions were made. SDK 3.0 also brought breaking changes to the client.run() API.
Fix: Review your automation logic, especially if it relied on Playwright internals. Consult the latest documentation for `0.12.3` and SDK 3.0 for API changes and new best practices.
gotcha Browser Use primarily leverages Chrome DevTools Protocol (CDP), meaning it only supports Chrome/Chromium-based browsers. Safari and Firefox are not supported.
Fix: Ensure you are targeting Chrome or Chromium for your browser automation tasks.
gotcha The library requires Python 3.11 or higher. Using older Python versions will lead to installation or runtime errors.
Fix: Upgrade your Python environment to version 3.11 or newer.
gotcha API keys (e.g., `BROWSER_USE_API_KEY`, `OPENAI_API_KEY`, `BROWSERLESS_TOKEN`) are essential for most functionalities. These should be set as environment variables, often loaded from a `.env` file using `python-dotenv` and `load_dotenv()`. Failure to set them correctly will result in authentication errors.
Fix: Create a `.env` file in your project root with your API keys and call `load_dotenv()` at the start of your script. Alternatively, set them directly in your environment.

Install

pip install browser-use PyPI
uv pip install browser-use with uv (recommended)
curl -fsSL https://browser-use.com/cli/install.sh | bash CLI Installation (macOS/Linux)

Imports

Agent
```
from browser_use import Agent
```
The core class for defining and running AI browser automation tasks.
Browser
```
from browser_use import Browser
```
For configuring browser settings like headless mode or cloud usage.
ChatBrowserUse
```
from browser_use import ChatBrowserUse
```
One of the recommended LLM wrappers optimized for browser automation tasks.

Quickstart

This quickstart demonstrates how to initialize an `Agent` with a natural language task and an LLM, then run it to automate web browsing. It includes setting up environment variables for API keys and basic browser configuration.

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

load_dotenv()

async def main():
    # Set API keys as environment variables (e.g., in a .env file)
    # BROWSER_USE_API_KEY=your_browser_use_key
    # OPENAI_API_KEY=your_openai_key (or other LLM provider)
    
    # Optionally configure the browser (headless by default)
    browser = Browser(
        # use_cloud=False, # Set to True to use Browser Use Cloud
        # headless=False,  # Set to False to see the browser window
        # window_size={'width': 1920, 'height': 1080}
    )

    agent = Agent(
        task="Go to example.com and extract the main heading text",
        llm=ChatBrowserUse(api_key=os.environ.get('BROWSER_USE_API_KEY', '')), # Or ChatOpenAI, ChatGoogle, etc.
        browser=browser,
    )

    result = await agent.run()
    print("Task completed.")
    print(f"Final result: {result.final_result()}")
    print(f"Visited URLs: {result.urls()}")

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

view raw JSON →