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 DocuRAGconstdocuRAG=newDocuRAG({qdrantUrl: 'http://localhost:6333',llmUrl: 'http://localhost:11434'});// Process a PDF bufferconst{ sessionId }=awaitdocuRAG.processPDFBuffer(pdfBuffer,fileName);// Chat with streamingawaitdocuRAG.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 doneawaitdocuRAG.cleanup(sessionId);

Configuration Options

{// Vector Store ConfigurationqdrantUrl: string,// Qdrant server URLvectorSize: number,// Default: 3072vectorDistance: string,// Default: 'Cosine'// LLM ConfigurationllmUrl: string,// LLM server URLllmModel: string,// Default: 'llama3.2'// Text ProcessingchunkSize: number,// Default: 1000chunkOverlap: number,// Default: 200// Search ConfigurationsearchLimit: 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