# Swarm Agents: A Collaborative Multi-Agent AI System

## Project Overview

Swarm Agents is my implementation of a modular, extensible framework for creating collaborative AI agent systems. I developed this project to explore how multiple specialized AI agents can work together, similar to how swarm intelligence functions in nature. The core concept is that a collection of specialized agents can solve complex problems more effectively than a single general-purpose agent.

## Key Components

### Triage System
I designed a smart routing mechanism that analyzes incoming queries and directs them to the most appropriate specialist agent. This ensures that each query is handled by an agent with the relevant expertise.

### Specialist Agents
The system includes four distinct agent types, each with specialized capabilities:

1. **Research Agent**: Focuses on gathering, synthesizing, and presenting information
2. **Creative Agent**: Handles creative tasks like writing, brainstorming, and design
3. **Code Agent**: Specializes in programming tasks and technical implementations
4. **Math Agent**: Concentrates on mathematical calculations and problem-solving

### Shared Memory Architecture
One of the most important features I implemented is a collective knowledge base that enables agents to:
- Access conversation history
- Build upon each other's work
- Maintain context throughout multi-turn interactions
- Store and retrieve important facts

## Technical Implementation

The project is built with Python and uses the OpenAI API to power the language models behind each agent. Key technical highlights include:

- **Abstraction Layers**: I created a clean inheritance structure with a base Agent class
- **Modular Design**: Each specialist agent is encapsulated with its own capabilities
- **Extensibility**: The system is designed to easily add new specialist types
- **Command-Line Interface**: User-friendly interaction through a CLI built with Typer

## Development Journey

Building Swarm Agents challenged me to think about AI systems differently. Instead of focusing on making a single agent more powerful, I explored how collaboration between simpler, specialized agents could produce emergent intelligence.

The most interesting technical challenge was designing the shared memory system to enable effective collaboration between agents without overwhelming the context window with irrelevant information.

## Future Enhancements

I'm continuing to develop this project with plans to:
- Add more specialist agent types
- Implement a web interface
- Incorporate external knowledge sources
- Build a visualization tool for agent interactions
- Optimize token usage through better memory management

## Links
- [GitHub Repository](https://github.com/mbennett-labs/swarm-agents)
- [Technical Documentation](https://hackmd.io/@xbBf1zpRTXaSATerS04iJg/HkeOGwd6yg)