This repository provides an experimental, educational tool for building multi-agent systems, specifically a natural language interface for querying PostgreSQL databases. It's designed for engineers and researchers interested in understanding and implementing agentic software patterns, leveraging LLMs for reasoning and decision-making with reduced explicit logic.

How It Works

The system utilizes a multi-agent approach powered by GPT-4, OpenAI's Assistants API, AutoGen, and Guidance. Agents, defined as LLM-powered tools with single purposes, collaborate within a "Multi-Agent Team" orchestrated by a central orchestrator. They communicate through defined "Conversation Flows," using "Instruments" (shared state and functions) to interact with the PostgreSQL database and achieve goals, such as answering user queries in natural language.

Quick Start & Requirements

• Install dependencies using poetry install.
• Configure .env with PostgreSQL URL and OpenAI API key.
• Run a prompt: poetry run start --prompt "<your question>".
• Requires Python ^3.10, PostgreSQL, and OpenAI API access.

Highlighted Details

• Demonstrates multi-agent patterns like Orchestrators, Instruments, and Decision Agents.
• Integrates OpenAI Assistants API, AutoGen, and Guidance for structured LLM responses.
• Codebase is structured by video series, with each branch corresponding to a specific video.
• Aims to enable reasoning and decision-making in software, mimicking human-like problem-solving.

Maintenance & Community

This repository is explicitly stated as an experiment and learning tool, not intended for long-term maintenance or updates beyond the lifespan of its associated video series. The codebase will be frozen upon series completion and used solely as a reference.

Licensing & Compatibility

The repository's licensing is not explicitly stated in the provided README text.

Limitations & Caveats

The project is experimental and not a framework or library. Debugging multi-agent systems can be challenging due to LLM non-determinism. Costs associated with running GPT-4 agents can be significant, and managing agent memory within context windows requires intricate code. The project is a snapshot in time and will not be updated.