MongoDB Atlas adds automated embeddings and Azure data federation
MongoDB Atlas introduces Automated Embedding in Public Preview with Voyage AI and brings Data Federation and Online Archive to Azure.
MongoDB Atlas introduced Automated Embedding in Public Preview, integrating Voyage AI embedding models to eliminate client-side embedding pipelines. Additionally, MongoDB Atlas Data Federation and Atlas Online Archive reached General Availability for deployment on Microsoft Azure.
What changed
Automated Embedding integrates Voyage AI embedding models into MongoDB Atlas vector search. The feature uses field-level delta detection to re-embed documents in near real-time only when indexed fields change. It separates inference into two mechanisms:
- Index builds use an Inference Flex Tier that is throughput-optimized to index large datasets without competing with live queries.
- Queries are latency-optimized to complete within agent response time budgets.
Automated Embedding supports search on views, allowing developers to define composite fields declaratively using MongoDB query language. It includes configurable quantization:
- Scalar quantization (default) provides a storage footprint roughly 4x smaller than float32.
- Binary quantization provides a footprint roughly 32x smaller.
It also features built-in dynamic batching that adjusts batch size based on document shape and size, along with error recovery that isolates and retries malformed documents without halting entire indexing jobs. For Azure deployments, Atlas Data Federation queries data across Atlas clusters and Microsoft Azure Blob Storage or Azure Data Lake Storage Gen2. Cross-cloud queries combining Azure storage and AWS S3 in the same federated database instance are not supported. A federated configuration supports at most one configured instance each of AWS S3, Azure Blob Storage, and Google Cloud Storage. For Atlas Online Archive on Azure, any newly created archive on an Azure cluster on or after 02/28 will default to Azure regions, provided there are no pre-existing AWS archives on that specific cluster. Cloud providers or storage regions cannot be edited or modified once configured.
What it means in practice
Eliminating client-side embedding code and manual backfill jobs reduces architectural complexity for agentic applications. Field-level delta detection prevents stale context without maintaining parallel ingestion pipelines. Storage quantization options allow teams to scale to hundreds of millions of vectors while managing resource costs. On the infrastructure side, Azure deployments can now query object storage natively and tier older data without defaulting storage backends to AWS, though multi-cloud queries remain architecturally separated within individual instances.
Context
These releases continue a multi-cloud and AI-focused trajectory for the platform, following the November 2026 releases of Atlas App Connections for secure AI tooling access and initial Azure deployments for Data Federation and Online Archive.
Sources
This article was written from the following, all of which are worth reading in full:
- mongodb
- vector-search
- azure
- ai