Multi-Agent RAG Template

A comprehensive template for creating RAG-enabled multi-agent systems using CrewAI and IBM WatsonxAI.

Overview

This template provides both basic multi-agent functionality and advanced RAG (Retrieval-Augmented Generation) capabilities. Choose from two implementations:

Basic Multi-Agent System (agent.py)

• Researcher Agent - Conducts web research using search tools
• Writer Agent - Creates written content based on research findings

RAG-Enabled Multi-Agent System (rag_agent.py)

• RAG-Powered Agents - Access local knowledge bases for document-based research
• Configurable Vector Databases - Support for ChromaDB, FAISS, and Pinecone
• Document Processing - Handle PDF, DOCX, TXT, and Markdown files
• Flexible Templates - Pre-built configurations for legal, technical, and support use cases

Quick Start

1. Install Dependencies

pip install -r requirements.txt

2. Configure API Keys

Set your environment variables:

export API_KEY="your_watsonx_api_key"
export SERPER_API_KEY="your_serper_api_key"  # Optional for web search

3. Choose Your Implementation

Basic Multi-Agent System

python agent.py

RAG-Enabled System

1. Configure your settings in config/rag_config_template.py
2. Add documents to data/documents/
3. Run the RAG system:

python rag_agent.py

RAG Features

Vector Database Support

• ChromaDB - Local vector database (default)
• FAISS - High-performance similarity search
• Pinecone - Cloud-based vector database

Document Processing

• PDF - Extract text from PDF documents
• DOCX - Process Word documents
• TXT/MD - Plain text and Markdown files
• Chunking - Intelligent text splitting with overlap

Embedding Models

• Sentence Transformers - Local embeddings (default)
• OpenAI - Cloud-based embeddings
• Custom - Easy integration of other models

Configuration

RAG Configuration (config/rag_config_template.py)

class RAGConfig:
    VECTOR_DB = "chroma"  # or "faiss", "pinecone"
    EMBEDDING_MODEL = "sentence-transformers"
    CHUNK_SIZE = 500
    RETRIEVAL_K = 3
    # ... more options

Agent Templates

Pre-built templates for common use cases:

• Legal Research - Legal document analysis and research
• Technical Documentation - API docs and technical guides
• Customer Support - FAQ and troubleshooting assistance
• Medical Research - Clinical literature review
• Business Analysis - Market research and intelligence

Examples

Legal Research System

# See examples/legal_research_example.py
python examples/legal_research_example.py

Technical Documentation

# See examples/technical_docs_example.py
python examples/technical_docs_example.py

Customer Support

# See examples/customer_support_example.py
python examples/customer_support_example.py

Project Structure

Multi Agent/
├── agent.py                 # Basic multi-agent system
├── rag_agent.py            # RAG-enabled system
├── config/
│   └── rag_config_template.py
├── rag/
│   ├── vector_stores/       # Vector database implementations
│   ├── document_processors/ # Document processing pipeline
│   ├── embeddings/         # Embedding model abstractions
│   ├── tools/              # RAG retrieval tools
│   └── knowledge_base_manager.py
├── templates/
│   ├── agent_templates.py   # Pre-built agent configurations
│   └── task_templates.py    # Pre-built task templates
├── examples/               # Complete use case examples
├── data/
│   └── documents/          # Your document storage
└── requirements.txt

Customization

Creating Custom Agents

from templates import AgentTemplates

custom_agent = AgentTemplates.create_custom_rag_agent(
    llm=llm,
    rag_tool=rag_tool,
    role="Your Custom Role",
    goal="Your Custom Goal",
    backstory="Your Custom Backstory"
)

Adding Custom Tasks

from templates import TaskTemplates

custom_task = TaskTemplates.create_custom_task(
    agent=agent,
    description="Your task description",
    expected_output="Your expected output format",
    output_file="output.md"
)

Knowledge Base Management

from rag.knowledge_base_manager import KnowledgeBaseManager

kb = KnowledgeBaseManager(config)
result = kb.add_documents_from_path("path/to/documents")
search_results = kb.search_knowledge_base("query", k=5)

Requirements

Core Dependencies

• CrewAI >= 0.28.8
• langchain-ibm >= 0.1.0
• IBM WatsonxAI access

Vector Database Options (choose one)

• chromadb >= 0.4.15 (local)
• faiss-cpu >= 1.7.4 (local)
• pinecone-client >= 2.2.4 (cloud)

Document Processing

• PyPDF2 >= 3.0.1 (PDF support)
• python-docx >= 0.8.11 (DOCX support)
• sentence-transformers >= 2.2.2 (embeddings)

Advanced Features

Multiple Vector Databases

Switch between vector databases by changing configuration:

config = {"vector_db": "chroma"}  # or "faiss", "pinecone"

Custom Embedding Models

Implement custom embeddings:

from rag.embeddings import BaseEmbeddings

class CustomEmbeddings(BaseEmbeddings):
    def embed_documents(self, texts):
        # Your implementation
        pass

Batch Document Processing

Process multiple document types:

from rag.document_processors import DocumentProcessorFactory

chunks = DocumentProcessorFactory.process_documents(file_paths, config)

License

This template is provided as-is for educational and development purposes.