SwarmDB Testing Platform

A lightweight platform for prototyping and testing swarm-based routing between specialized agents and document / vector stores. Built with Python, LangChain, and the OpenAI Swarm library.

What's This?

This is my sandbox for experimenting with agent routing patterns. It uses a coordinator agent to direct queries to specialized agents (SQL and RAG) based on the query type. Those agents can then call back to the coordinator for more complex queries, or between one another if it's needed. Nothing fancy.

Key Features

• CLI interface with pretty formatting
• Coordinator-based routing between specialized agents
• SQL Agent for database queries (using the SQLite toy DB, Chinook.db)
• RAG Agent for document retrieval and analysis
• TODO: Basic test coverage with pytest

Quick Start Demo

1. Clone the repo

2. Set up your environment:

   python -m venv venv
   source venv/bin/activate  # or `venv\Scripts\activate` on Windows

   2a. Install torch and swarm

   For PyTorch, see: https://pytorch.org

   For Swarm:

   pip install git+https://github.com/openai/swarm.git # swarm

   2b. Install the rest of the dependencies:

   pip install -r requirements.txt

3. Copy .env.example to .env, fill in your OpenAI API key, and optionally update the LangChain API key for tracing:

   OPENAI_API_KEY= <YOUR_KEY_HERE>
   LANGCHAIN_TRACING_V2=true
   LANGCHAIN_API_KEY= <YOUR_KEY_HERE>
   LANGCHAIN_PROJECT=openai-swarm
   TOKENIZERS_PARALLELISM=false
   

4. Download the Chinook.db toy, and move it to data/databases:

   wget https://github.com/lerocha/chinook-database/raw/master/ChinookDatabase/DataSources/Chinook_Sqlite.sqlite
   mv Chinook_Sqlite.sqlite data/databases/Chinook.db
   

5. Run the CLI:

   python main.py

Project Structure

• core/: Agent implementations and core logic

• interfaces/: CLI and other interfaces (TODO)

• tests/: Test suite (TODO)

• config/: Configuration and settings

• utils/: Shared utilities

• data/databases: SQL .db files go here

• data/documents: Documents for RAG Agent go here

• data/vector_stores/chroma_db/: ChromaDB vector store goes here

Notes

I don't really plan on turning this into something production-ready - I'm mainly using it to explore different routing patterns and agent interactions. Feel free to use it as a starting point for your own experiments or Swarm projects.

License

MIT