Microsoft's DocumentDB Reimagines NoSQL on PostgreSQL
Microsoft open-sourced its vCore Cosmos DB engine as DocumentDB, a PostgreSQL-based NoSQL layer with MongoDB-compatible APIs, BSON support, and ambitions to become a NoSQL standard.
Microsoft has open-sourced the MongoDB-compatible engine behind Cosmos DB's vCore architecture under the MIT license, reviving the original DocumentDB name. The project layers two components on top of PostgreSQL: a BSON extension for handling MongoDB's native data format, and an API layer providing CRUD operations, queries, and index management. This combines PostgreSQL's proven reliability with MongoDB's flexible document model, while also unlocking PostgreSQL's vector indexing for RAG applications and recommendation engines.
Microsoft's broader ambition is to establish a common NoSQL API and engine standard, similar to what exists for SQL. DocumentDB's layered design allows different protocol translation layers—mimicking MongoDB, Aerospike, or CouchDB—to sit atop the same underlying BSON store, positioning it as a vendor-neutral reference implementation.
Notably, the first real-world adopter wasn't Microsoft itself but the open-source FerretDB project. FerretDB 2.0, rebuilt on DocumentDB, claims up to 20x performance gains while retaining its own Apache 2.0 license. Documentation currently lives only in a GitHub wiki, and there's no official SDK yet, meaning developers must hand-build API calls—though the MongoDB-based design should ease most migrations.