Agentic RAG

A multi-agent Retrieval-Augmented Generation (RAG) system built with FastAPI and LangChain. This demo showcases how multiple AI agents can work together to provide intelligent responses by retrieving relevant information from a knowledge base, in this case about retail and product information.

Features

• Multi-agent architecture: Separate retriever and responder agents for specialized tasks
• Vector-based search: Semantic search using ChromaDB for document retrieval
• FastAPI REST API: Clean HTTP endpoints for querying the system
• Async processing: Background task processing with webhook callbacks and queue.
• Configurable: Environment-based configuration for easy deployment
• Callback integration: RAG responses are sent to a configurable callback URL.

Architecture

The system consists of two main agents:

1. Retriever Agent: Performs semantic search in the vector database to find relevant documents.
2. Responder Agent: Generates responses using retrieved context and handles user interactions. It is responsible for deciding whether to call the retriever agent or not.

Demo

curl -X POST "http://localhost:8000/query" \
     -H "Content-Type: application/json" \
     -d '{
       "query": "What is the price and availability of product 1427b3b0?",
       "user_id": "user123"
     }'

The result of this question comes from 3 files (product description, prices and stock files). The response will be received by the callback endpoint and can be seen in the logs, like this:

INFO:app: * RAG RESPONSE: The product **1427b3b0 (Mystery Book)** is priced at **$111** but is currently **out of stock**.

Prerequisites

• Python 3.12 or higher
• Mistral API key (for the language model)

Setup

The easiest way to manage the project is using the make commands from the Makefile file. Try running this to see all the available commands:

make help

Option 1: Using Docker (Recommended)

1. Clone the repository

   git clone <repository-url>
   cd agentic-rag

2. Set up environment variables

   cp .env.sample .env

   Edit .env and add your Mistral API key:

   MISTRAL_API_KEY=your_mistral_api_key_here
   

3. Build and start the application

   docker compose build
   docker compose up

4. View logs (optional)

   docker compose logs -f

The API will be available at http://localhost:8000

Option 2: Using uv

1. Clone the repository

   git clone <repository-url>
   cd agentic-rag

2. Install uv (if not already installed)

   curl -LsSf https://astral.sh/uv/install.sh | sh

3. Install dependencies

   uv sync

4. Set up environment variables

   cp .env.sample .env

   Edit .env and add your Mistral API key:

   MISTRAL_API_KEY=your_mistral_api_key_here
   

Option 3: Using pip

1. Clone the repository

   git clone <repository-url>
   cd agentic-rag

2. Create and activate virtual environment

   python -m venv .venv
   source .venv/bin/activate  # On Windows: .venv\Scripts\activate

3. Install dependencies

   pip install -r requirements.txt

   If requirements.txt doesn't exist, install from pyproject.toml:

   pip install -e .

4. Set up environment variables

   cp .env.sample .env

   Edit .env and add your Mistral API key:

   MISTRAL_API_KEY=your_mistral_api_key_here
   

Usage

Running the Application

With Docker:

docker compose up

With uv:

uv run uvicorn app:app --port 8000

With pip:

uvicorn app:app --port 8000

The API will be available at http://localhost:8000

API Documentation

Once running, visit http://localhost:8000/docs for interactive API documentation.

API Endpoints

POST /query

Submit a query to the RAG system.

Request body:

{
  "query": "What is the price of Product 1?",
  "user_id": "user123"
}

Response:

{
  "message": "User query enqueued successfully!"
}

POST /webhook

Receives callback responses (for internal use).

Configuration

The application can be configured through environment variables in the .env file:

• MISTRAL_API_KEY: Your Mistral API key (required)
• DOCUMENTS_DIRECTORY: Directory containing documents to index (default: "documents")
• VECTOR_DB_DIRECTORY: Directory for vector database storage (default: ".vector_db")
• CHUNK_SIZE: Text chunk size for document processing (default: 1000)
• CHUNK_OVERLAP_SIZE: Overlap between chunks (default: 100)
• TOP_K: Number of documents to retrieve (default: 5)
• CALLBACK_URL: Webhook URL for responses (optional)

Document Management

The system automatically indexes documents from the documents/ directory on startup. Supported formats:

• .txt files
• .md files

To add new documents:

1. Place files in the documents/ directory
2. Restart the application to re-index

Running Tests

With Docker:

docker compose exec app uv run pytest tests/

With uv:

uv run pytest tests/

With pip:

pytest tests/

Project Structure

agentic-rag/
├── agents/           # AI agents (retriever, responder)
├── app/             # FastAPI application
├── documents/       # Knowledge base documents
├── rag/            # RAG implementation and vector store
├── tests/          # Unit tests
├── config.py       # Configuration management
└── pyproject.toml  # Project dependencies

License

This project is licensed under the terms specified in the LICENSE file.