AG2 Multi-Agent Conversations with Pathway Real-Time RAG

This example demonstrates how to combine AG2 (formerly AutoGen) multi-agent conversations with Pathway's real-time RAG pipeline.

AG2 is a multi-agent conversation framework with 500K+ monthly PyPI downloads, 4,300+ GitHub stars, and 400+ contributors.

Table of Contents

1. Introduction
2. Prerequisites
3. Architecture
4. Setup and Installation
5. Usage
6. How It Works
7. Conclusions

Introduction

This project combines two powerful frameworks:

• Pathway continuously indexes documents in real-time using its streaming engine, serving them through a VectorStoreServer REST API
• AG2 orchestrates multiple AI agents (Researcher + Analyst) that query Pathway's index as a tool during their conversation

The key advantage: Pathway re-indexes documents automatically whenever they change, so AG2 agents always query the latest version of the knowledge base — no manual re-indexing required.

Prerequisites

• Python >= 3.10
• OpenAI API key

Architecture

Documents (live folder) --> Pathway VectorStoreServer (real-time indexing)
                                         |
                                    REST API /v1/retrieve
                                         |
User Query --> AG2 UserProxy --> GroupChat [Researcher + Analyst]
                                         |
                               search_documents tool --> HTTP POST --> Pathway
                                         |
                               Grounded, real-time answers with citations

Setup and Installation

1. Clone the Repository:

   git clone https://github.com/pathwaycom/pathway.git
   cd pathway/examples/projects/ag2-multiagent-rag/
   

2. Install Dependencies:

   pip install -U pathway "ag2[openai]>=0.11.4,<1.0" requests python-dotenv
   

3. Environment Variables: Create a .env file in the project root and add your OpenAI API key:

   OPENAI_API_KEY=your_openai_api_key_here
   

4. Add Documents: Place your documents (TXT, MD, PDF) in the ./data/ directory. A sample document is included for testing.

Usage

1. Run the Pipeline:

   python main.py
   

2. The script will:
   • Start a Pathway VectorStoreServer that indexes documents in ./data/
   • Launch AG2 agents that query the server for information
   • Print the multi-agent conversation with grounded answers
3. Add documents while running — Pathway re-indexes automatically.

How It Works

• Pathway reads documents from ./data/, chunks them with TokenCountSplitter, embeds them with OpenAI embeddings, and serves the index via HTTP
• AG2 Researcher agent queries Pathway's /v1/retrieve endpoint via the search_documents tool to retrieve relevant chunks
• AG2 Analyst agent synthesizes retrieved information into a comprehensive answer
• The agents communicate via AG2's GroupChat, coordinated by a GroupChatManager

Conclusions

This example shows how Pathway's real-time document indexing complements AG2's multi-agent orchestration. The combination is especially useful for scenarios where documents change frequently and agents need access to the latest information — such as live knowledge bases, continuously updated reports, or streaming data pipelines.

You can find more ready-to-run pipelines in our templates section.