This example demonstrates how to integrate Steel with the browser-use framework to create an AI agent capable of interacting with web browsers. It leverages Steel's session management to provide a secure, cloud-based browser instance for browser-use to connect to.

Features

Seamless integration between Steel and browser-use
Cloud-based browser automation with Steel sessions
AI-powered browser interaction using OpenAI GPT-4
Automatic cleanup of resources
Robust error handling and session management

Prerequisites

Python 3.11 or higher
Python package manager (Recommended uv: pip install uv)
Steel API key (Get 100 free browser hours here.)
OpenAI API key

Installation

Clone the repository:

git clone https://github.com/steel-dev/steel-cookbook.git
cd steel-cookbook/examples/steel-browser-use-starter

Create and activate a virtual environment with venv:

python -m venv .venv
source .venv/bin/activate  # On Windows, use: .venv\Scripts\activate

Install the required dependencies using venv:
```
pip install .
```
Note: We're using venv to manage our environment and pip to install the packages. This ensures all dependencies are resolved consistently using pip's dependency resolution. If you're having issue with speed or dependencies, feel free to switch back to python/pip and try these steps again.
Copy the environment variables file and configure your API keys:
```
cp .env.example .env
```

Edit the .env file and add your API keys:

STEEL_API_KEY=your_steel_api_key_here
OPENAI_API_KEY=your_openai_api_key_here

Usage

Run the example with:

python main.py

The script will:

Create a new Steel session
Connect browser-use to the Steel session using the WebSocket URL
Execute the AI agent's task (e.g., summarizing the Steel Docs changelog)
Automatically clean up resources when done

How It Works

Session Creation: The script creates a Steel session with optional proxy and CAPTCHA-solving capabilities.
Browser Integration: The WebSocket URL from the Steel session is passed to browser-use's Browser configuration.
AI Agent Setup: A browser-use Agent is created with:
- The specified task
- OpenAI GPT-4 language model
- Browser instance connected to Steel
Execution: The agent runs autonomously, performing the specified task.
Cleanup: Resources are properly released, including:
- Closing the browser connection process
- Releasing the Steel session

Error Handling

The example includes comprehensive error handling:

Validation of required environment variables
Try/except blocks for session management
Proper cleanup in case of failures
Informative error messages

Customization

You can modify the example by:

Changing the task in main.py
Adjusting Steel session parameters
Configuring different browser-use settings
Using a different OpenAI model or adjusting model parameters

Contributing

Feel free to submit issues, fork the repository, and create pull requests for any improvements.

License

This example is part of the Steel Cookbook and is licensed under the MIT License. See the LICENSE file for details.