Simplifying RAG with MongoDB

Over the past year, I've spent a lot of time talking to developers about Retrieval-Augmented Generation (RAG) and how MongoDB Atlas Vector Search enables efficient vector-based retrieval. While the power of MongoDB as a vector database is undeniable, I noticed a recurring theme: developers wanted a simpler way to implement RAG applications.

That's what inspired me to create MongoDB-RAG, an npm library that abstracts away the complexity of embedding generation, vector search, and document retrieval giving developers a plug-and-play way to build RAG-powered applications with MongoDB.

Today, I want to introduce the library and share some of the exciting new features we've recently added to make RAG with MongoDB even more intuitive and performant.

📌 Why I Built MongoDB-RAG When I first started experimenting with vector search and large language models (LLMs), I found myself repeatedly writing boilerplate code for:

• ✅ Generating embeddings using OpenAI, DeepSeek, or Ollama
• ✅ Storing embeddings in MongoDB efficiently
• ✅ Building vector indexes with optimal settings
• ✅ Running similarity searches using MongoDB Atlas Vector Search
• ✅ Filtering and retrieving documents with hybrid search

Every RAG application required these steps, but the process felt unnecessarily repetitive. What if we could standardize this flow into a reusable library?

MongoDB-RAG does just that—eliminating complexity so that you can go from querying unstructured data to getting intelligent results in just a few lines of code.

🚀 What's New in MongoDB-RAG?

Since launching the first version, we've been working hard to refine and expand the library. Here are some of the latest features that make MongoDB-RAG a must-have for building RAG applications with MongoDB Atlas:

🔹 Dynamic Database & Collection Selection

Developers can now specify custom databases and collections at query time, allowing more flexible data organization. Previously, all data had to be stored in a predefined database, but now you can run searches dynamically across multiple datasets.

const results = await rag.search("What is AI?", {
  database: "research_db",
  collection: "ai_papers",
  maxResults: 5
});

🔹 Intelligent Document Chunking

One of the biggest challenges in RAG is breaking large documents into meaningful chunks. MongoDB-RAG now includes three advanced chunking strategies:

1. Sliding Window - Maintains context with overlapping chunks
2. Semantic Chunking - Uses sentence boundaries to create more meaningful segments
3. Recursive Chunking - Dynamically splits large sections into smaller chunks

const chunkedDocs = chunker.chunkDocument(myDocument, { strategy: "semantic" });

🔹 Multi-Embedding Provider Support

MongoDB-RAG now supports multiple embedding providers—so you're not locked into one ecosystem.

Supported providers include:

• ✅ OpenAI (text-embedding-3-small, text-embedding-3-large)
• ✅ DeepSeek (high-performance embeddings with affordable pricing)
• ✅ Ollama (local LLM-based embeddings for privacy and cost-efficiency)

const rag = new MongoRAG({
  embedding: { provider: "ollama", baseUrl: "http://localhost:11434", model: "llama3" }
});

🔹 Vector Quantization for Faster Queries

We've integrated automatic vector quantization, reducing memory footprint and boosting search performance in MongoDB Atlas. You can now enable scalar or binary quantization effortlessly.

const indexDefinition = {
  fields: [
    { type: "vector", path: "embedding", numDimensions: 1536, similarity: "cosine", quantization: "binary" }
  ]
};

🔹 Hybrid Search: Combining Vector & Metadata Filters

One of the biggest advantages of MongoDB over other vector databases is the ability to perform hybrid searches—combining vector similarity with traditional filters.

MongoDB-RAG makes it seamless:

const results = await rag.search("Latest AI papers", {
  database: "research_db",
  collection: "papers",
  filter: { "metadata.year": { $gte: 2022 } }
});

Now, you can refine vector searches using structured metadata like dates, authors, or categories.

⚡ How to Get Started

Setting up MongoDB-RAG is ridiculously easy. Just install the package and connect it to your MongoDB Atlas cluster:

1️⃣ Install MongoDB-RAG

npm install mongodb-rag dotenv

2️⃣ Set Up MongoDB Atlas Ensure you have an Atlas cluster with Vector Search enabled, then store your connection string in .env:

MONGODB_URI=mongodb+srv://your-user:your-password@your-cluster.mongodb.net/
EMBEDDING_PROVIDER=openai
EMBEDDING_API_KEY=your-openai-api-key
EMBEDDING_MODEL=text-embedding-3-small

3️⃣ 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: "my_db", collection: "docs" });

4️⃣ Perform a Vector Search

const results = await rag.search("How does vector search work?", {
  database: "my_db",
  collection: "docs",
  maxResults: 3
});
console.log("Search Results:", results);

And just like that, your MongoDB-powered RAG application is up and running!

🔮 What's Next?

We're constantly evolving MongoDB-RAG based on developer feedback. Here's what's coming next:

• ✅ RAG Pipelines: End-to-end orchestration for document retrieval & LLM response generation
• ✅ Integration with LangChain: Seamless connection with AI agent frameworks
• ✅ Built-in UI Dashboard: Visualize vector search performance and document embeddings

🎯 Final Thoughts

I built MongoDB-RAG because I wanted an easier, more efficient way to work with vector search in MongoDB. Instead of reinventing the wheel every time I built a RAG system, I wanted a reusable, well-optimized library that handles all the heavy lifting.

Now, with dynamic database selection, hybrid search, intelligent chunking, and multi-provider embeddings, I truly believe MongoDB-RAG is the fastest way to build production-ready RAG applications.

Give it a try, and let me know what you think! 🚀

• 📌 GitHub Repo: github.com/mongodb-developer/mongodb-rag
• 📌 NPM Package: npmjs.com/package/mongodb-rag

Let's simplify RAG development together! 👏