MongoDB Atlas
by MongoDB, Inc.
Managed multi-cloud database that stores your operational data and its vectors in one place
MongoDB Atlas is MongoDB's fully managed database service on AWS, Azure and Google Cloud, and since Atlas Vector Search it also serves as the vector store for RAG and agentic applications. Documents, metadata and embeddings live in one system under one query language and one security model, removing the sync pipeline between a primary database and a separate vector database.
MongoDB Atlas is MongoDB, Inc.'s fully managed multi-cloud database platform, and since the arrival of Atlas Vector Search it doubles as the vector store behind retrieval-augmented generation and agentic applications. Operational documents, metadata and embeddings live in one system with one query language and one security model, which removes the synchronisation pipeline teams otherwise run between a primary database and a standalone vector database. Vector Search runs approximate nearest-neighbour queries over HNSW indexes as well as exact nearest-neighbour and k-NN, accepts embeddings up to 4,096 dimensions from any provider, and supports scalar and binary quantisation to cut index memory. A $vectorSearch aggregation stage composes with the ordinary Query API, so metadata filters, graph lookups, geospatial predicates and lexical search combine with semantic search in a single pipeline, and native fusion stages implement hybrid lexical-plus-vector retrieval. Following MongoDB's acquisition of Voyage AI, Atlas auto-generates and re-syncs embeddings as documents change and runs Voyage reranker models inside the query engine, which raises retrieval precision and lowers downstream LLM token spend. Search Nodes isolate the search workload onto separate infrastructure so vector queries scale independently of the transactional cluster. Atlas runs on AWS, Azure and Google Cloud behind a 99.995% uptime SLA, and the same capabilities are available self-managed through Enterprise Advanced and Community Edition. In the first quarter of fiscal 2027, reported on 28 May 2026, MongoDB posted $687.6 million of revenue, of which Atlas was more than $512.5 million and grew over 29% year over year, spread across more than 67,700 customers including 2,895 spending at least $100,000 a year. Named AI references include Novo Nordisk, Okta and Delivery Hero, and the integration list covers LangChain, LlamaIndex, Haystack, Microsoft Semantic Kernel, OpenAI, AWS and Spring.
The platform or data engineering team already running MongoDB that needs a vector store for RAG and does not want to operate a second database and an embedding sync pipeline alongside it.
Semantic search, lexical search and transactional reads execute in one aggregation pipeline against live operational data, so there is nothing to keep in sync and no stale index between the two systems.
At a Glance
- Category
- Data & Analytics
- Pricing
- Freemium, Usage-based, Subscription, Contact for pricing
- Target Market
- CIOs, CTOs, Data Scientists, Enterprise Developers, Platform Engineers
- Deployment
- Cloud-first, Multi-cloud, Self-hosted, Hybrid
- Headquarters
- New York, United States
- Team Size
- 500+
- Customers
- 67,700+ MongoDB customers, of which 66,400+ use Atlas (Q1 FY2027, reported 28 May 2026)
Key Features
- ✓Atlas Vector Search
Native ANN, ENN and k-NN search over HNSW indexes supporting embeddings up to 4,096 dimensions from any provider, inside the same database as your documents.
- ✓Hybrid search with native fusion
Dedicated fusion stages combine lexical and vector results in one query, so relevance tuning does not require an external reranking service.
- ✓Voyage AI auto-embeddings and reranking
Embeddings are generated and re-synced automatically as data changes, and Voyage reranker models run in the query engine to raise precision and cut LLM token cost.
- ✓Search Nodes
Search and vector workloads run on separate, independently scaled infrastructure, giving workload isolation so a heavy retrieval job cannot starve transactional queries.
- ✓Vector quantisation
Scalar and binary quantisation shrink index memory footprint substantially, which is what makes large embedding sets affordable on dedicated clusters.
- ✓Composable Query API
$vectorSearch sits in the normal aggregation pipeline alongside metadata filters, graph lookups, geospatial predicates and lexical search, so one query does the whole job.
- ✓Multi-cloud managed operations
Automated scaling, backup, monitoring and a 99.995% uptime SLA across AWS, Azure and Google Cloud, with FedRAMP Moderate available via Atlas for Government.
Capabilities
Use Cases
- •RAG over operational data
Ground an LLM in documents that are already in the primary database, avoiding a second store and the sync lag it introduces.
- •Filtered semantic product search
Combine vector similarity with hard metadata filters like price, region or stock so results are both relevant and valid.
- •Agent memory and context retrieval
Serve semantic context to autonomous agents from live application state rather than a periodically refreshed embedding snapshot.
- •Recommendation and personalisation
Delivery Hero uses Atlas for e-commerce personalisation and product recommendations built on embeddings of catalogue and behavioural data.
- •Anomaly detection over unstructured data
Detect outliers by comparing new records against embedded historical patterns without exporting data to a separate analytics stack.
Ideal For
Best For
- ✓Retrieval-augmented generation where the source documents already live in MongoDB and embeddings should stay next to them
- ✓Hybrid search that combines keyword relevance with semantic similarity in a single query and fuses the results natively
- ✓Agentic applications needing filtered semantic retrieval — vector similarity combined with metadata predicates and graph lookups in one pipeline
- ✓Multi-cloud or regulated deployments that need region choice across AWS, Azure and Google Cloud with SOC 2, ISO 27001, HIPAA and PCI DSS coverage
- ✓Recommendation and personalisation engines running against live operational data rather than a nightly export
Not Ideal For
- ✗Cost-sensitive teams with stable, predictable workloads — practitioners on Hacker News report cutting bills by roughly 90% moving to self-hosted Postgres, and one described Atlas as 500% more expensive than running in-house
- ✗Workloads dominated by multi-table relational transactions; PeerSpot reviewers specifically call out weak OLTP support compared with traditional relational databases
- ✗Very large single databases — one practitioner reported Atlas could not cope past roughly 4TB per database and migrated away
- ✗Teams that want a pure, best-of-breed vector engine with specialised index types, where a dedicated vector database may still be faster to tune
Integrations
Deployment
Market & Ratings
67,700+ MongoDB customers, of which 66,400+ use Atlas (Q1 FY2027, reported 28 May 2026)
Market Analysis
Pros
- ✓One system for documents, metadata and vectors eliminates the sync pipeline and the staleness window between a primary database and a separate vector store
- ✓Genuine multi-cloud reach across AWS, Azure and Google Cloud with a 99.995% uptime SLA and secure-by-default configuration
- ✓Deep compliance coverage — SOC 2, ISO 27001/27017/27018, HIPAA, HITRUST, PCI DSS, CSA STAR and FedRAMP Moderate for government workloads
- ✓PeerSpot reviewers rate it 4.2 out of 5 across 51 reviews with 96% willing to recommend, praising scaling, the managed operations and the console
- ✓Financially durable and heavily invested in: Atlas passed $512.5 million of quarterly revenue growing over 29% year over year in Q1 FY2027
Cons
- ✗Cost is the dominant complaint. Hacker News practitioners report a 90% bill reduction after moving to self-hosted Postgres, and one described Atlas as roughly 500% more expensive than running in-house
- ✗The add-on billing model compounds it — Search, Vector Search, Stream Processing, egress, backups and Data Federation are all metered separately from the cluster tier
- ✗Support quality is inconsistent at the sharp end: one engineer described a two-day production outage where support could not identify the cause and the customer had to diagnose it
- ✗PeerSpot reviewers cite weak OLTP transaction support relative to relational databases, and ask for better documentation, tutorials and non-developer query tooling
- ✗Scale ceiling reports exist — a practitioner said Atlas could not cope with growth past roughly 4TB per database
- ✗Forced upgrade policies have cost renewals; one commenter described losing a $50K/year Atlas contract over them and moving to PostgreSQL
Pricing
M0 Free
$0
- ✓512 MB storage
- ✓Shared RAM and vCPU
- ✓Free forever, no card required
- ✓No backup
Flex
From $8/mo
- ✓5 GB storage
- ✓$0.011/hour base, capped at $30/month
- ✓Usage-based tiers from 0 to 500 ops/sec
- ✓Burst capacity for dev and test
Dedicated (M10 and up)
From $56.94/mo
- ✓M10 at $0.08/hour with 2 GB RAM and 2 vCPUs
- ✓Scales to M300 at $21.85/hour, 384 GB RAM, 96 vCPUs
- ✓10 GB to 4 TB storage
- ✓Backups, egress and add-ons billed separately
Enterprise Advanced (self-managed)
Contact for pricing
- ✓On-premises or private cloud
- ✓Ops Manager, Cloud Manager, Kubernetes Operator
- ✓Enterprise security features
- ✓Consultative support
Atlas is metered hourly by cluster tier, and the headline cluster price is only part of the bill. The M0 free tier gives 512 MB forever; Flex runs $8 to $30 a month by operations per second; dedicated clusters start at $0.08/hour for M10 (about $56.94/month) and reach $21.85/hour for M300. Search and Vector Search are separately billed add-ons — Search Nodes run $0.12 to $3.27/hour and Vector Search $0.12 to $4.22/hour — with further charges for backups, data egress, Stream Processing ($0.06–$2.49/hour), Data Federation ($5/TB scanned) and Online Archive. Enterprise Advanced for self-managed deployments is quote-only. Cost is the single most common complaint in independent reviews, and the add-on structure is why bills surprise teams that budgeted from the cluster tier alone.
Security & Compliance
Connect
Sources
This page was written from 6 sources, 3 on domains other than mongodb.com.
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
Monte Carlo
Data and AI agent observability for enterprises running production AI
DataBahn
Agentic data control plane that cuts security telemetry 40-70% before it reaches your SIEM
Actian VectorAI DB
A local-first vector database for AI that runs on the edge, on-prem and air-gapped
Nimble
Expert-level web search agents that learn your domain, cutting AI research token spend roughly in half