Cosmos DB vector search: container vector policy (path, data type, dimensions, distance function), vector indexes (flat, quantizedFlat, diskANN) and VectorDistance()

Embedding models turn text or images into vectors: long arrays of numbers where similar meanings sit close together. Vector search finds the items whose vectors are nearest to a query vector. Cosmos DB for NoSQL can store embeddings inside the same JSON items as your operational data and search them, so a RAG (retrieval-augmented generation) app does not need a separate vector database. The feature must be enabled on the account first (the vector search capability for NoSQL).