Azure Cosmos DB

Azure Cosmos DB is Microsoft's managed NoSQL database on Azure, with NoSQL, MongoDB, Cassandra, Gremlin and Table APIs, for global and AI apps, billed by usage.

Azure Cosmos DB stores the operational data behind applications that must answer quickly for users in several regions at once. Microsoft builds and runs it as a managed service of its Azure cloud, and teams open it from a standard Azure account rather than a separate signup. Documents are indexed automatically, with no schema to define before the first write. AI applications keep their records and their vector embeddings together in it. One service answers through several APIs: the native API for NoSQL, which serves JSON documents, and APIs for MongoDB, Apache Cassandra, Apache Gremlin and Table. Containers are spread across physical partitions by a partition key, and the service splits partitions on its own as data grows. Each account sets a default consistency level between strong and eventual, and multi-region writes settle conflicts with Last Write Wins or a custom policy. The change feed records every change to a container in order, while time to live removes items after a set period. The API for NoSQL adds full-text search with BM25 scoring, vector search on DiskANN indexes and Always Encrypted for chosen properties. Developers building web and mobile backends for a spread-out audience are the core users. AI teams use it as the store that holds application state and embeddings side by side, close to Microsoft Foundry. Teams that already write to MongoDB, Cassandra, Gremlin or table interfaces can keep that code on a managed service. As a hosted Azure service, it fits organisations whose applications already run on Microsoft's cloud. Capacity is either provisioned throughput, fixed or autoscaled, or a serverless mode that needs no capacity planning. Event-driven systems read the change feed through the change feed processor or the pull model and turn each insert and update into work for downstream services. Code running on Azure Functions or Azure Kubernetes Service treats it as the operational store behind global web and mobile apps. For reporting, analytical storage keeps a column-oriented copy of the data that Azure Synapse Link and Apache Spark query. Continuous backup lets an operator restore an account to any moment inside its retention window. Machine learning teams working in Azure Machine Learning keep embeddings next to the records they describe, in the same containers their applications read.