Multi-Agent System

A sophisticated multi-agent system where a Project Manager (PM) agent coordinates multiple Software Engineering (SWE) agents to build software projects. Inspired by mini-swe-agent's philosophy of simplicity and effectiveness.

Features

• Project Manager Agent: Creates detailed PRDs and coordinates implementation phases
• SWE Agents: Implement atomic tasks with full context awareness
• Git-based Collaboration: Each agent works on separate branches with proper version control
• Real-time Dashboard: Web interface to monitor agent progress
• Comprehensive Logging: All agent activities logged to logs/ directory
• Windows Compatible: Fully tested on Windows with proper command handling

Prerequisites

• Python 3.8+
• Git
• pip

Installation

1. Clone the repository:

git clone <repository-url>
cd multiagent

2. Install dependencies:

pip install flask litellm python-dotenv

3. Configure your LLM:

# Copy the sample environment file
cp .env.sample .env

# Edit .env with your LLM configuration
# See .env.sample for examples of different providers

Configuration

The system supports any OpenAI-compatible API. Configure in .env:

LLM_API_URL=https://api.example.com/v1/
LLM_API_KEY=your-api-key-here
LLM_MODEL=model-name-here

Supported Providers

• OpenAI: GPT-4, GPT-3.5
• Anthropic: Claude 3 Opus, Sonnet, Haiku
• Local Models: Ollama, LM Studio
• Cloud Providers: Cerebras, Together AI, Anyscale
• Any OpenAI-compatible endpoint

Usage

1. Start the server:

python app.py

2. Open your browser to http://localhost:5000

3. Create a new project:

   • Enter project name
   • Provide project description
   • Click "Create Project"

4. Monitor progress:

   • Watch agents create PRD
   • See tasks being assigned
   • Track implementation progress
   • View real-time logs

How It Works

Phase 1: Planning

1. PM Agent creates a detailed Product Requirements Document (PRD)
2. PRD includes file structure, technical requirements, and implementation phases
3. PM Agent creates atomic tasks with clear success criteria

Phase 2: Implementation

1. PM Agent creates SWE agents for each task
2. Each SWE agent:
   • Works on a unique git branch
   • Reads existing files for context
   • Implements only their specific task
   • Commits changes when complete
3. PM Agent monitors progress and verifies completion

Key Design Principles

• Atomic Tasks: Each task does one thing well
• Full Context: Agents see all command outputs and maintain conversation history
• File-based Creation: Uses <file> tags for reliable file creation
• Verification: PM can read files to verify implementation

Project Structure

multiagent/
├── app.py                 # Main Flask application
├── prompts/              # Agent system prompts
│   ├── pm_agent.txt      # Project Manager prompt
│   └── swe_agent.txt     # Software Engineer prompt
├── templates/            # Web interface templates
│   └── dashboard.html    # Main dashboard
├── logs/                 # Agent logs (auto-created)
│   └── <project>/        # Per-project logs
│       └── <timestamp>/  # Per-run logs
└── projects/             # Created projects (auto-created)

Logging

All agent activities are logged to logs/<project_name>/<timestamp>/:

• pm_agent_*.log: Project Manager activities
• swe_agent_*.log: Individual SWE agent activities

Troubleshooting

"Missing required environment variables"

• Ensure you've copied .env.sample to .env
• Fill in all required LLM configuration values

Empty directories not tracked by git

• The system automatically creates .gitkeep files in empty directories
• This ensures git can track the directory structure

Windows command issues

• The system uses Windows-compatible commands
• File creation uses <file> tags, not echo commands

Contributing

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Submit a pull request

License

[Your chosen license]

Acknowledgments

Inspired by mini-swe-agent and its philosophy of simplicity and effectiveness.