Multi-Agent-RAG-Template

A production-ready template for building Multi-Agent RAG (Retrieval-Augmented Generation) systems using the Swarms Framework. This template demonstrates how to create a collaborative team of AI agents that work together to process, analyze, and generate insights from documents.

🌟 Features

Plug-and-Play Agent Architecture

- Easily swap or modify agents without disrupting the system - Add custom agents with specialized capabilities - Define your own agent interaction patterns - Scale from 2 to 100+ agents seamlessly - Any LLM can be used, this template uses GROQ but you can use any other LLM such as OpenAI, Anthropic, Cohere, etc.

Adaptable Document Processing

- Support for any document format through custom extractors - Flexible document storage options (local, cloud, or hybrid) - Customizable chunking and embedding strategies - Dynamic index updates without system restart - Any RAG system can be used, this template uses LlamaIndexDB but you can use any other RAG system.

Configurable Workflows

- Design custom agent communication patterns - Implement parallel or sequential processing - Add conditional logic and branching workflows - Adjust system behavior through environment variables

🚀 Quick Start

Clone the Repository

git clone https://github.com/The-Swarm-Corporation/Multi-Agent-RAG-Template.git
cd Multi-Agent-RAG-Template

Set Up Environment

# Create and activate virtual environment (optional but recommended)
python -m venv venv
source venv/bin/activate  # On Windows: .\venv\Scripts\activate
Install dependencies
pip install -r requirements.txt

Configure Environment Variables

# Create .env file
Edit .env file with your credentials
GROQAPIKEY="your-api-key-here"
WORKSPACEDIR="agentworkspace"
OPENAIAPIKEY="your-openai-api-key-here"

Run the Example

python main.py

🏗️ Project Structure

Multi-Agent-RAG-Template/
├── main.py                    # Main entry point
├── multiagentrag/
│   ├── agents.py             # Agent definitions
│   └── memory.py             # RAG implementation
├── docs/                      # Place your documents here
├── requirements.txt           # Project dependencies
└── .env                      # Environment variables

🔧 Customization

Adding New Agents

Open multiagentrag/agents.py
Create a new agent using the Agent class:

new_agent = Agent(
    agent_name="New-Agent",
    system_prompt="Your system prompt here",
    llm=model,
    max_loops=1,
    # ... additional configuration
)

Modifying the Agent Flow

In main.py, update the flow parameter in the AgentRearrange initialization:

flow=f"{agent1.agentname} -> {agent2.agentname} -> {newagent.agentname}"

Integrating RAG

The memory_system parameter in the AgentRearrange initialization is used to configure the RAG system.
The memory_system parameter is an instance of LlamaIndexDB, which is a database class for storing and retrieving medical documents.
The memory_system class must have a query(query: str) method that returns a string for the agent to use it.

# Import the AgentRearrange class for coordinating multiple agents
from swarms import AgentRearrange
from multiagentrag.agents import (     diagnostic_specialist,     medicaldataextractor,     patientcarecoordinator,     specialist_consultant,     treatment_planner, )
from multiagentrag.memory import LlamaIndexDB
router = AgentRearrange(     name="medical-diagnosis-treatment-swarm",     description="Collaborative medical team for comprehensive patient diagnosis and treatment planning",     max_loops=1,     agents=[         medicaldataextractor,         diagnostic_specialist,         treatment_planner,         specialist_consultant,         patientcarecoordinator,     ],     memory_system=LlamaIndexDB(         data_dir="docs",         filenameasid=True,         recursive=True,         similaritytopk=10,     ),     flow=f"{medicaldataextractor.agentname} -> {diagnosticspecialist.agentname} -> {treatmentplanner.agentname} -> {specialistconsultant.agentname} -> {patientcarecoordinator.agentname}", )
if name == "main":     router.run(         "Analyze this Lucas Brown's medical data to provide a diagnosis and treatment plan"     )

Pinecone Example

Here is an example of how to use Pinecone as the RAG system.

Make sure you have a Pinecone index created and the PINECONEAPIKEY, PINECONEINDEXNAME, and PINECONE_ENVIRONMENT environment variables set.
See the pinecone_swarm.py file for the full example.
The PineconeManager class is used to interface with the Pinecone API.

import os
from swarms import AgentRearrange
from multiagentrag.agents import (
    diagnostic_specialist,
    medicaldataextractor,
    patientcarecoordinator,
    specialist_consultant,
    treatment_planner,
)
from multiagentrag.pinecone_wrapper import PineconeManager
router = AgentRearrange(     name="medical-diagnosis-treatment-swarm",     description="Collaborative medical team for comprehensive patient diagnosis and treatment planning",     max_loops=1,     agents=[         medicaldataextractor,         diagnostic_specialist,         treatment_planner,         specialist_consultant,         patientcarecoordinator,     ],     memory_system=PineconeManager(         apikey=os.getenv("PINECONEAPI_KEY"),         indexname=os.getenv("PINECONEINDEX_NAME"),         environment=os.getenv("PINECONE_ENVIRONMENT"),     ),     flow=f"{medicaldataextractor.agentname} -> {diagnosticspecialist.agentname} -> {treatmentplanner.agentname} -> {specialistconsultant.agentname} -> {patientcarecoordinator.agentname}", )
if name == "main":     router.run(         "Analyze this Lucas Brown's medical data to provide a diagnosis and treatment plan"     )