Browser Automation Agent

[Crustdata Build Challenge : AI Agent for Browser Automation]

This project is a browser automation agent that can understand natural language instructions and perform actions in a web browser, using Gemini LLMs and Playwright.

Architecture

The agent consists of three main components:

1. Planner: Breaks down a user's natural language query into high-level steps.
2. Executor: Performs each step by interacting with the browser through Playwright.
3. Verifier: Verifies if each step was completed successfully.

Setup

1. Clone the repository
2. Install the requirements:

   pip install -r requirements.txt
   

3. Install Playwright browsers:

   playwright install chromium
   

4. Create a .env file with your Gemini API key:

   GEMINI_API_KEY=your_api_key_here
   

Usage

Run the main script:

python interact.py

Enter your natural language query when prompted. For example:

login to reddit.com, search ghibli, save the first three posts with metadata like date, author etc.

The agent will:

1. Plan the steps needed to achieve your goal
2. Execute each step in the browser
3. Verify each step was completed successfully
4. Report progress and results

Configuration

Edit config.py to modify the agent's behavior:

• Change model names
• Enable/disable headless mode
• Adjust retry limits
• Enable/disable human-in-the-loop mode

Files

• interact.py: Entry point script
• models.py: Data models
• planner.py: Step planning module
• executor.py: Browser interaction module
• verifier.py: Step verification module
• utils.py: Utility functions
• config.py: Configuration settings