MongoDB Atlas Vector Search
Vector search built into MongoDB Atlas, storing embeddings next to your documents with hybrid search and Voyage AI reranking
Overview
Atlas Vector Search adds vector similarity search to MongoDB Atlas, so embeddings live in the same documents as the rest of your application data. A $vectorSearch stage in the aggregation pipeline runs approximate (HNSW) or exact nearest-neighbor queries with pre-filters on any field, and results can flow straight into the rest of the pipeline. Hybrid search combines vector results with Atlas Search full-text relevance.
MongoDB acquired the embedding company Voyage AI in 2025 and has been building its models into Atlas for embedding generation and reranking. Vector Search became generally available in December 2023 and is available on every Atlas tier, including the free M0 cluster, with dedicated Search Nodes for isolating heavy vector workloads. Also from MongoDB: Voyage AI embeddings.
The Verdict
Who Should Use Atlas Vector Search?
Best For
- Applications already on MongoDB
- RAG over operational data that changes often
- Queries that combine vectors with document filters
- Teams that want one database instead of two
Not Ideal For
- Teams not using MongoDB
- Pure vector workloads at very large scale on a budget
- Self-hosted deployments without Atlas (feature parity varies)
What's Great
- No separate vector database to sync or operate
- Rich pre-filtering on any document field
- Hybrid search with full-text built in
- Voyage AI embeddings and reranking from the same vendor
- Works with LangChain, LlamaIndex, Semantic Kernel and Haystack
Watch Out For
- Costs grow quickly for vector-heavy workloads on dedicated tiers
- Data must live in MongoDB to be searched
- Search Nodes add a separate line item
- Lock-in to the Atlas platform
Pricing
View all features & details
Key Features
- $vectorSearch aggregation stage
- HNSW approximate and exact nearest neighbor
- Pre-filtering on document fields
- Scalar and binary quantization
- Hybrid search with Atlas Search
- Dedicated Search Nodes
Integrations
- LangChain, LlamaIndex, Haystack
- Microsoft Semantic Kernel, Spring AI
- Voyage AI, OpenAI and other embedding models
- AWS, Azure and Google Cloud
How It Compares
| Feature | Atlas Vector Search | pgvector | Elasticsearch | Pinecone |
|---|---|---|---|---|
| Data model | Documents plus vectors | Relational plus vectors | Search index | Vectors only |
| Free tier | Yes (M0) | Open source | Self-managed | Yes |
| Hybrid full-text | Yes | Via Postgres FTS | Best-in-class | Sparse vectors |
| Built-in embeddings | Voyage AI | No | ELSER and inference API | Integrated inference |
| Self-hosted | Limited | Yes | Yes | No |
Is this your tool? Add a badge