This repository is a fork of browser-use with enhanced logging capabilities and automated bug report generation. It allows AI agents to control web browsers to complete given task while providing detailed logging and analysis of agent behavior.

• YAML-based Configuration: All agent behavior configurable via YAML files
• Enhanced Logging: Detailed logging of agent actions and browser interactions
• Automated Bug Report Generation: Generate comprehensive bug reports
• Screenshot Capture: Visual documentation of agent actions
• Flexible Task Execution: Execute tasks from configurable task files
• Multi-LLM Support: Support for OpenAI, Anthropic, and other LLM providers
• Memory Management: Long-term memory for complex multi-step tasks
• Authentication Management: Persistent login and session management
• Security Controls: Domain restrictions and sensitive data handling

1. Clone this repository

2. Install dependencies:

pip install -r requirements.txt

3. Install Patchright (required for browser automation):

patchright install chromium --with-deps --no-shell

4. Create a .env file with your API keys:

OPENAI_API_KEY=your_openai_api_key_here
ANTHROPIC_API_KEY=your_anthropic_api_key_here
AZURE_OPENAI_ENDPOINT=
AZURE_OPENAI_KEY=
GEMINI_API_KEY=
DEEPSEEK_API_KEY=
GROK_API_KEY=
NOVITA_API_KEY=

For a quick test run:

1. Set your OpenAI API key in .env
2. Run: python main.py
3. The agent will execute the default task with 5 steps
4. Check agent_logs/ and agent_screenshots/ for results

For detailed documentation, see BROWSER_USE_TUTORIAL.md.

Edit configs/config.yaml to configure browser, LLM, and agent settings:

# Browser configuration
browser:
  headless: false
  use_test_profile: true
  test_profile_name: "test_profile"

# Browser context configuration
context:
  disable_security: true
  cookies_file: "auth_data/test_profile/cookies.json"
  allowed_domains: null  # null = allow all domains

# LLM configuration
llm:
  provider: "openai"  # or "anthropic"
  model: "gpt-4o"
  temperature: 0.7

# Agent configuration
agent:
  max_steps: 5            # Maximum steps per task
  max_failures: 3         # Max consecutive failures
  use_vision: true        # Enable screenshot analysis
  enable_memory: true     # Enable long-term memory
  max_actions_per_step: 10

# Authentication management
auth_manager:
  ensure_logged_in: false  # Auto-login to services
  save_auth_state: false   # Save auth between runs

tasks_file: "tasks_test.txt"

Create a task file (e.g., tasks_test.txt) with your tasks:

Academic website: Go to the website "XXX", explore and analyze the content
Google search: Search for "browser automation tools" and extract the top 5 results
Form filling: Fill out the contact form on example.com with test data

Run the setup script to configure authentication:

python setup_test_env.py

This script will:

• Prompt you to input login credentials for automated processing
• Save browser cookies for future launches

Execute the main script:

python main.py

The script will:

• Load configuration from configs/config.yaml
• Either use saved cookies or perform programmatic login
• Read tasks from the configured task file
• Execute each task using the AI agent with your configured settings
• Generate logs in the agent_logs/ directory
• Capture screenshots in the agent_screenshots/ directory

Custom Configuration:

# Use a different config file
python main.py with configs/config_research.yaml

Different Use Cases:

Quick Testing (config_quick.yaml):

agent:
  max_steps: 10
  use_vision: false
  enable_memory: false
llm:
  model: "gpt-4o-mini"

Production (config_prod.yaml):

agent:
  max_steps: 50
browser:
  headless: true
context:
  allowed_domains: ["trusted-site.com"]

Run the bug report generation pipeline:

python generate_bug_report.py

This generates bug reports in the bug_reports/ directory - change the specific task to generate a report on.

Key configuration parameters in configs/config.yaml:

For complete configuration documentation, see BROWSER_USE_TUTORIAL.md.

• main.py - Main entry point for running tasks
• configs/config.yaml - Main configuration file for all agent settings
• setup_test_env.py - Script for setting up login credentials
• generate_bug_report.py - Script for generating bug reports
• analyze_agent_run.py - Script for analyzing agent runs
• agent_logs/ - Directory containing detailed agent logs
• agent_screenshots/ - Directory containing screenshots of agent actions
• bug_reports/ - Directory containing generated bug reports
• tasks_test.txt - Default file containing tasks to be executed
• auth_data/ - Directory containing saved browser profiles and cookies
• BROWSER_USE_TUTORIAL.md - Comprehensive tutorial on agent architecture and usage

This project is licensed under the same license as the original browser-use repository.

This project is based on the original browser-use repository by Gregor Žunič and Magnus Müller.