MongoDB-RAG

Overview

MongoDB-RAG (Retrieval Augmented Generation) is an NPM module that simplifies vector search using MongoDB Atlas. This library enables developers to efficiently perform similarity search, caching, batch processing, and indexing for fast and accurate retrieval of relevant data.

🚀 Features

• Vector Search: Efficiently retrieves similar documents using MongoDB's Atlas Vector Search.
• Dynamic Database & Collection Selection: Supports flexible selection of multiple databases and collections.
• Batch Processing: Handles bulk processing of documents with retry mechanisms.
• Index Management: Ensures necessary indexes are available and optimized.
• Caching Mechanism: Provides in-memory caching for frequently accessed data.
• Advanced Chunking: Supports sliding window, semantic, and recursive chunking strategies.
• CLI for Scaffolding RAG Apps

🚀 Getting Started

1️⃣ Install the Package

npm install mongodb-rag dotenv

2️⃣ Set Up MongoDB Atlas

1. Create a MongoDB Atlas Cluster (MongoDB Atlas)
2. Enable Vector Search under Indexes:

   {
     "definition": {
       "fields": [
         { "path": "embedding", "type": "vector", "numDimensions": 1536, "similarity": "cosine" }
       ]
     }
   }

3. Get Your Connection String and store it in .env:

   MONGODB_URI=mongodb+srv://<username>:<password>@cluster0.mongodb.net/
   EMBEDDING_PROVIDER=openai  # Options: openai, deepseek
   EMBEDDING_API_KEY=your-embedding-api-key
   EMBEDDING_MODEL=text-embedding-3-small  # Change based on provider
   VECTOR_INDEX=default

3️⃣ Quick Start with CLI

You can generate a fully working RAG-enabled app with MongoDB Atlas Vector Search using:

npx mongodb-rag create-rag-app my-rag-app

This will:

• Scaffold a new CRUD RAG app with Express and MongoDB Atlas.
• Set up environment variables for embedding providers.
• Create API routes for ingestion, search, and deletion.

Then, navigate into your project and run:

cd my-rag-app
npm install
npm run dev

4️⃣ Initialize MongoRAG

import { MongoRAG } from 'mongodb-rag';
import dotenv from 'dotenv';
dotenv.config();

const rag = new MongoRAG({
    mongoUrl: process.env.MONGODB_URI,
    database: 'my_rag_db',  // Default database
    collection: 'documents', // Default collection
    embedding: {
        provider: process.env.EMBEDDING_PROVIDER,
        apiKey: process.env.EMBEDDING_API_KEY,
        model: process.env.EMBEDDING_MODEL,
        dimensions: 1536
    }
});
await rag.connect();

5️⃣ Ingest Documents

const documents = [
    { id: 'doc1', content: 'MongoDB is a NoSQL database.', metadata: { source: 'docs' } },
    { id: 'doc2', content: 'Vector search is useful for semantic search.', metadata: { source: 'ai' } }
];

await rag.ingestBatch(documents, { database: 'dynamic_db', collection: 'dynamic_docs' });
console.log('Documents ingested.');

6️⃣ Perform a Vector Search

const query = 'How does vector search work?';

const results = await rag.search(query, {
    database: 'dynamic_db',
    collection: 'dynamic_docs',
    maxResults: 3
});

console.log('Search Results:', results);

7️⃣ Close Connection

await rag.close();

⚡ Additional Features

🌍 Multi-Database & Collection Support

Store embeddings in multiple databases and collections dynamically.

await rag.ingestBatch(docs, { database: 'finance_db', collection: 'reports' });

🔎 Hybrid Search (Vector + Metadata Filtering)

const results = await rag.search('AI topics', {
    database: 'my_rag_db',
    collection: 'documents',
    maxResults: 5,
    filter: { 'metadata.source': 'ai' }
});

🧪 Testing

Run tests using:

npm test

Run in watch mode:

npm run test:watch

Check test coverage:

npm run test:coverage

🤝 Contributing

Contributions are welcome! Please fork the repository and submit a pull request.

📜 License

This project is licensed under the MIT License.

💡 Examples

• For more examples, check our examples directory.

🔗 Links

• CLI Reference
• Documentation
• GitHub Repository
• Bug Reports
• MongoDB Atlas