Facilitate mathematical distance calculations between data points: Configure a vector index.

Define the document schema for high-dimensional data: Store data as a numeric array.

Azure Cosmos DB for NoSQL supports native vector search by storing embeddings directly in JSON documents and defining a vector embedding policy plus a vector index on the embedding path. Microsoft documents that vector search compares a query vector with stored vectors by calculating similarity or distance through the VectorDistance() system function. A vector index materially improves this process by reducing search latency, increasing throughput, and lowering RU consumption compared with an unindexed vector scan.

The embedding property itself must be stored as an array of numeric values . Microsoft examples show embedding properties such as " contentVector " : [2, -1, 4, ...] and define the associated vector policy with attributes including path, data type, dimensions, and distance function. This representation is required because embeddings are high-dimensional numerical vectors generated by an embedding model.

A composite index optimizes queries involving multiple scalar properties but does not provide vector-distance indexing. A Base64-encoded string cannot be used directly for native vector similarity calculations because Cosmos DB expects the vector field to contain numeric values matching the configured dimensionality.

Study Guide references: Azure Cosmos DB for NoSQL → vector embedding policies; vector indexes; VectorDistance(); numeric embedding arrays; native vector search.