RAG Flow

A visual, node-based RAG (Retrieval-Augmented Generation) pipeline builder that allows you to create, customize, and execute RAG workflows through an intuitive drag-and-drop interface.

Overview

RAG Flow enables you to build complex document processing and question-answering pipelines by connecting different nodes in a visual graph. Each node performs a specific task—loading documents, splitting text, generating embeddings, storing vectors, retrieving relevant context, and generating AI responses.

Architecture

┌─────────────┐     ┌───────────────┐     ┌─────────────┐
│  PDF Loader │ ──▶ │ Text Splitter │ ──▶ │  Embedder   │
└─────────────┘     └───────────────┘     └─────────────┘
                                                  │
                                                  ▼
┌─────────────┐     ┌───────────────┐     ┌─────────────┐
│ LLM Response │ ◀───│   Retriever   │ ◀───│ Vector Store│
└─────────────┘     └───────────────┘     └─────────────┘

Tech Stack

Component	Technology
Backend	FastAPI, Python
Frontend	React + Vite + ReactFlow
Vector Store	ChromaDB
Embeddings	HuggingFace (sentence-transformers/all-MiniLM-L6-v2)
LLM	Ollama (llama3)
Graph Execution	NetworkX

Getting Started

Prerequisites

• Python 3.9+ (if running locally)
• Node.js 18+ (if running locally)
• Docker & Docker Compose
• Ollama installed and running (for LLM)

Backend Setup

cd backend

# Create virtual environment
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

# Start the backend server
uvicorn app.main:app --reload

Frontend Setup

cd frontend

# Install dependencies
npm install

# Start the development server
npm run dev

Ollama Setup

Make sure Ollama is running with the llama3 model:

ollama pull llama3
ollama serve

Docker Setup

# Build and run all services
docker-compose up --build

# Run in background
docker-compose up -d --build

The services will be available at:

• Frontend: http://localhost:5173
• Backend: http://localhost:8000

Note: The backend is configured to connect to Ollama on the host machine at http://host.docker.internal:11434. If Ollama runs elsewhere, update OLLAMA_BASE_URL in docker-compose.yml.

API Endpoints

Method	Endpoint	Description
POST	/run-flow	Execute a RAG flow
GET	/documents	List all indexed documents
DELETE	/documents/{source}	Delete a document by source

Implemented Features

Nodes

• PDF Loader - Load PDF documents from file paths
• Text Splitter - Split documents into overlapping chunks (default: 500 chars, 50 overlap)
• Embedder - Generate embeddings using HuggingFace sentence transformers
• Vector Store - Persist and query embeddings using ChromaDB
• Retriever - Retrieve relevant context from vector store based on query
• LLM - Generate responses using Ollama with RAG context (streaming support)

Engine

• Graph-based Execution - Flows execute as directed acyclic graphs (DAG) using topological sort
• Node Registry - Dynamic node registration system for easy extensibility
• Streaming Responses - Real-time token streaming from LLM to client

Visual Flow Editor

• Drag-and-drop canvas - Built with ReactFlow
• Sidebar - Add nodes (PDF Loader, Text Splitter, Embedder, Vector Store, Retriever, LLM)
• Node configuration panel - Configure node settings
• Export/Import flows - Save and load flow configurations as JSON files

Document Management

• List all indexed documents with chunk counts
• Delete documents by source
• Persistent vector storage

Future Goals

High Priority

• More Document Loaders - Support for Word, CSV, TXT, HTML, and URLs
• Multiple Embedding Models - Support for OpenAI, Azure OpenAI, and other embedding providers
• Multiple LLM Providers - Support for OpenAI, Anthropic, Google Gemini, and local models

Medium Priority

• Flow Templates - Pre-built templates for common RAG use cases
• Conversation Memory - Maintain chat history across interactions
• Multi-document RAG - Query across multiple document sources
• Hybrid Search - Combine semantic and keyword search
• Re-ranking - Add re-ranking nodes for improved retrieval

Lower Priority

• Flow Versioning - Track and version flow changes
• Analytics Dashboard - Track query performance and metrics
• API Key Management - Secure storage for API keys
• User Authentication - Multi-user support with roles
• Custom Nodes - Plugin system for user-defined nodes

Example Flow JSON

{
  "nodes": [
    {
      "id": "loader",
      "data": {
        "type": "pdf_loader",
        "config": { "path": "./documents/sample.pdf" }
      }
    },
    {
      "id": "splitter",
      "data": { "type": "text_splitter", "config": {} }
    },
    {
      "id": "embedder",
      "data": { "type": "embedder", "config": {} }
    },
    {
      "id": "vectorstore",
      "data": { "type": "vector_store", "config": {} }
    },
    {
      "id": "retriever",
      "data": {
        "type": "retriever",
        "config": { "query": "What is this document about?" }
      }
    },
    {
      "id": "llm",
      "data": { "type": "llm", "config": { "model": "llama3" } }
    }
  ],
  "edges": [
    { "source": "loader", "target": "splitter" },
    { "source": "splitter", "target": "embedder" },
    { "source": "embedder", "target": "vectorstore" },
    { "source": "vectorstore", "target": "retriever" },
    { "source": "retriever", "target": "llm" }
  ]
}

License

MIT