docuRAG.js

💡 Chat with your PDF documents

A JavaScript library for building RAG-powered document question-answering systems. docuRAG.js provides a streamlined solution for implementing Retrieval-Augmented Generation using Qdrant vector database and local LLM integration.

Core Features

• LLM Integration: Flexible local LLM support with streaming responses
• Vector Storage: Qdrant integration for vector similarity search
• Text Processing: RecursiveCharacterTextSplitter from LangChain
• Streaming Responses: Server-Sent Events (SSE) for real-time chat responses
• PDF Processing: Automatic PDF text extraction and chunking
• Session Management: Built-in session handling for document contexts
• Framework Agnostic: Can be used with any Node.js framework

Quick Start

Prerequisites

• Modern JavaScript runtime (Node.js 18+ for server-side)
• Running Qdrant instance
• Local LLM server (e.g., Ollama with Llama2)

  ⚠️ Note: Currently tested and optimized for Llama2. Other models may work but are not officially supported.

Setup

# Start Qdrant
docker run -p 6333:6333 qdrant/qdrant

# Start Llama2
ollama run llama2

# Install docuRAG
npm install docurag

Basic Usage

import { DocuRAG } from 'docurag';

// Initialize DocuRAG
const docuRAG = new DocuRAG({
    qdrantUrl: 'http://localhost:6333',
    llmUrl: 'http://localhost:11434'
});

// Process a PDF buffer
const { sessionId } = await docuRAG.processPDFBuffer(pdfBuffer, fileName);

// Chat with streaming
await docuRAG.chat(sessionId, "What is this document about?", {
    onData: (data) => console.log(data.response),
    onEnd: () => console.log("Done"),
    onError: (error) => console.error(error)
});

// Clean up when done
await docuRAG.cleanup(sessionId);

Configuration Options

{
    // Vector Store Configuration
    qdrantUrl: string,      // Qdrant server URL
    vectorSize: number,     // Default: 3072
    vectorDistance: string, // Default: 'Cosine'

    // LLM Configuration
    llmUrl: string,        // LLM server URL
    llmModel: string,      // Default: 'llama3.2'

    // Text Processing
    chunkSize: number,     // Default: 1000
    chunkOverlap: number,  // Default: 200

    // Search Configuration
    searchLimit: number    // Default: 3
}

Examples

• Express Example - Complete implementation with UI
• NestJS Example - Same features, NestJS implementation

Contributing

Areas for contribution:

• Additional vector store integrations
• Alternative LLM providers
• Enhanced chunking strategies
• Performance optimizations
• Testing infrastructure

License

MIT License - see LICENSE

Resources

• Qdrant Documentation
• LangChain JS
• Ollama
• LLama2

---

Built with ❤️ by Yannis Kolovos