Browser Use CLI

An enhanced interactive CLI tool for browser automation using the official browser-use library. This tool provides a user-friendly interface to control your browser using natural language commands with advanced features and customization options.

🚀 Now aligned with browser-use v0.2.2 - the latest official release!

Features

• 🤖 Multiple LLM Provider Support:

  • OpenAI GPT-4o (default)
  • Anthropic Claude 3.5 Sonnet (20241022)
  • Azure OpenAI Services
  • Google Gemini (via browser-use)
  • DeepSeek (via browser-use)
  • And more through browser-use's LLM integrations

• 🔒 Configurable System Behaviors:

  • Default mode for standard automation
  • Safety First mode with enhanced security
  • Data Collection mode for comprehensive gathering
  • Research mode for systematic exploration
  • Wikipedia First mode for research tasks

• 📸 Advanced Logging and Recording:

  • Automatic conversation logging
  • Session recordings (when configured)
  • Comprehensive task execution logs
  • Structured data storage
  • Debug-level logging support

• 🌐 Modern Browser Integration:

  • Uses browser-use's optimized browser handling
  • Vision support for visual understanding
  • Configurable browser settings
  • Support for headless and headed modes
  • Cloud browser provider compatibility

• 🛠️ Enhanced User Experience:

  • Interactive CLI with clear feedback
  • Structured output formats (JSON, etc.)
  • Error handling and recovery
  • Graceful shutdown handling
  • Cross-platform compatibility

Quick Start

Simple Usage

For basic browser automation, you can use the simple example:

import asyncio
from dotenv import load_dotenv
from langchain_openai import ChatOpenAI
from browser_use import Agent

load_dotenv()

async def main():
    llm = ChatOpenAI(model="gpt-4o")
    agent = Agent(
        task="Compare the price of gpt-4o and DeepSeek-V3",
        llm=llm,
    )
    result = await agent.run()
    print(result)

asyncio.run(main())

Interactive CLI

For an enhanced interactive experience with multiple features:

python main.py

Output Formats

The tool supports structured output formats using Pydantic models. Currently available formats:

Posts Format

class Post:
    post_title: str
    post_url: str
    num_comments: int
    hours_since_post: int

Enable structured output by setting the OUTPUT_FORMAT environment variable:

# Use structured posts format
OUTPUT_FORMAT=posts

Prerequisites

1. Python 3.11 or higher (required by browser-use)
2. API Keys for your chosen LLM provider:
   • OpenAI API Key (for GPT-4o - default)
   • Anthropic API Key (for Claude 3.5 Sonnet)
   • Azure OpenAI credentials (for Azure OpenAI)
   • Google API Key (for Gemini)
   • DeepSeek API Key (for DeepSeek models)

Setup

1. Clone the Repository

git clone https://github.com/PierrunoYT/browser-use-script
cd browser-use-script

2. Install Dependencies

Using pip:

pip install -r requirements.txt

Using uv (recommended):

uv pip install -r requirements.txt

3. Install Playwright Browsers

playwright install chromium --with-deps --no-shell

4. Configure Environment

# Copy the example environment file
cp .env.example .env  # On macOS/Linux
copy .env.example .env  # On Windows

5. Edit .env with your settings:

# Required: Add your API key
OPENAI_API_KEY=your_openai_api_key_here

# Optional: Configure LLM provider
LLM_PROVIDER=openai  # Options: openai, anthropic, azure, google, deepseek

# Optional: Configure system behavior
SYSTEM_PROMPT=default  # Options: default, safety, collection, research, wiki

# Optional: Browser settings
BROWSER_HEADLESS=false
USE_VISION=true

# Optional: Telemetry
ANONYMIZED_TELEMETRY=true

Usage

Interactive CLI

Start the enhanced CLI for interactive browser automation:

python main.py

The tool will display your current configuration:

Welcome to Browser Use CLI!
Using LLM Provider: OPENAI
System Prompt: DEFAULT
Enter your tasks and watch the browser automation in action.
Press Ctrl+C to exit.

Simple Script Usage

For basic automation, use the simple example:

python simple_example.py

Example Tasks

Enter your tasks in natural language. Here are some examples:

• Web Research: "Search for the latest AI news and summarize the top 3 articles"
• Information Gathering: "Go to Wikipedia and find information about quantum computing"
• Comparison Tasks: "Compare the pricing of OpenAI GPT-4 and Anthropic Claude"
• Data Collection: "Visit Hacker News and get the top 5 posts with their titles and URLs"

System Prompt Modes

Default Mode

• Standard browser automation behavior
• Balanced between functionality and safety

Safety First Mode

• Enhanced security and privacy features
• Requires confirmation for form submissions
• Respects robots.txt and terms of service
• Prevents automated logins without permission
• Avoids suspicious or untrusted links

Data Collection Mode

• Focused on comprehensive data gathering
• Automatic search result saving
• Screenshot capture of relevant content
• Organized data storage with timestamps
• Detailed URL documentation

Output and Logs

The tool automatically creates and organizes various outputs:

• logs/conversation_*.json: Detailed conversation history
• logs/results/*.json: Structured search results
• logs/screenshots/*.png: Element screenshots
• logs/recordings/: Browser session recordings
• logs/traces/: Debug trace files

Example Tasks

Here are some example tasks you can try:

• "Go to Reddit, search for 'browser-use' and return the first post's title"
• "Search for flights on kayak.com from New York to London"
• "Go to Google Docs and create a new document titled 'Meeting Notes'"
• "Visit GitHub and star the browser-use repository"

What's New in This Version

🚀 Aligned with Official browser-use v0.2.2

• Modern API: Updated to use the latest browser-use API patterns
• Simplified Architecture: Removed complex custom controller logic in favor of browser-use's built-in capabilities
• Better Performance: Leverages browser-use's optimized browser handling
• Enhanced Compatibility: Full compatibility with the official browser-use ecosystem

🔄 Migration from Previous Versions

If you're upgrading from an older version of this script:

1. Dependencies: The script now uses the official browser-use package instead of custom implementations
2. Configuration: Environment variables remain largely the same for backward compatibility
3. Custom Functions: Complex custom functions have been simplified to align with browser-use patterns
4. API Changes: The core Agent API is now simpler and more reliable

Dependencies

Core Dependencies

• browser-use >= 0.2.2 (official browser automation library)
• langchain-openai >= 0.3.11 (OpenAI LLM integration)
• langchain-anthropic >= 0.3.3 (Anthropic Claude integration)
• langchain-core >= 0.3.49 (LangChain core functionality)
• playwright >= 1.52.0 (browser automation engine)
• python-dotenv >= 1.0.1 (environment variable management)
• pydantic >= 2.10.4 (data validation and serialization)

Optional Dependencies

• rich >= 14.0.0 (enhanced CLI formatting)
• click >= 8.1.8 (CLI framework)
• sentence-transformers >= 4.0.2 (for memory features)

Contributing

Contributions are welcome! Feel free to open issues for bugs or feature requests.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Browser Configuration Options

Standard Browser

The default configuration launches a new browser instance with customizable settings:

# .env configuration
BROWSER_HEADLESS=false
BROWSER_VIEWPORT_WIDTH=1280
BROWSER_VIEWPORT_HEIGHT=1100

Connect to Existing Chrome

Connect to your real Chrome browser with existing profiles and logged-in sessions:

# .env configuration
CHROME_INSTANCE_PATH=C:\Program Files\Google\Chrome\Application\chrome.exe  # Windows
CHROME_INSTANCE_PATH=/Applications/Google Chrome.app/Contents/MacOS/Google Chrome  # macOS
CHROME_INSTANCE_PATH=/usr/bin/google-chrome  # Linux

Cloud Browser Providers

Connect to cloud-based browser services for enhanced reliability:

# .env configuration
# WebSocket connection (wss)
BROWSER_WSS_URL=wss://your-provider.com/browser

# Chrome DevTools Protocol (CDP)
BROWSER_CDP_URL=http://your-cdp-provider.com

Additional Browser Settings

Fine-tune browser behavior with these settings:

# .env configuration
# Page Load Settings
MIN_PAGE_LOAD_TIME=0.5
NETWORK_IDLE_TIME=1.0
MAX_PAGE_LOAD_TIME=5.0

# Security Settings
BROWSER_DISABLE_SECURITY=true
IGNORE_HTTPS_ERRORS=true
JAVASCRIPT_ENABLED=true

# Display Settings
HIGHLIGHT_ELEMENTS=true
VIEWPORT_EXPANSION=500
BROWSER_LOCALE=en-US

# URL Restrictions
ALLOWED_DOMAINS=["example.com","another-domain.com"]

# Debug and Recording
SAVE_RECORDING_PATH=logs/recordings
TRACE_PATH=logs/traces

Common Browser Configurations

Local Development

BROWSER_HEADLESS=false
BROWSER_DISABLE_SECURITY=true
USE_VISION=true

Production Environment

BROWSER_HEADLESS=true
BROWSER_DISABLE_SECURITY=false
USE_VISION=true
ALLOWED_DOMAINS=["trusted-domain.com"]

Using Existing Chrome Profile

CHROME_INSTANCE_PATH=/path/to/chrome
USE_PERSISTENT_CONTEXT=true

Environment Variables

In addition to the basic configuration, you can customize:

Function Control

# Exclude specific functions
EXCLUDED_ACTIONS=[]  # JSON array of action IDs

# Output format
OUTPUT_FORMAT=  # Options: posts, or leave empty for text

Debug Settings

# Enable debug logging for model thoughts
LOG_LEVEL=DEBUG

# Save browser recordings
SAVE_RECORDING_PATH=logs/recordings
TRACE_PATH=logs/traces

Output Directory Structure

The tool organizes outputs in the following structure:

logs/
├── browser_use.log         # Main log file
├── conversation_*.json     # Conversation history
├── results/               # Structured search results
├── screenshots/           # Element screenshots
├── content/              # Extracted page content
├── tables/              # CSV table data
├── downloads/           # Downloaded files
├── recordings/         # Browser session recordings
└── traces/            # Debug trace files