Skip to content
datastudyguides

AI-200 · 2.1.3 · Develop AI solutions by using Azure data management services

Cosmos vector search

Store and retrieve embeddings and execute vector similarity search for semantic retrieval

Store with items, define a vector policy and index, then rank semantic matches with VectorDistance and TOP N.

Store, index, retrieve

Cosmos DB vector search flow: store embeddings, define a vector policy, create a vector index, and rank nearest matches with VectorDistance and TOP N.

Cosmos DB for NoSQL stores an beside the item it represents. Define the vector path, dimensions and when the container is created, then choose a vector index that fits the workload.

Rank by similarity

SELECT TOP 5 c.title,
       VectorDistance(c.contentVector, @embedding) AS score
FROM c
ORDER BY VectorDistance(c.contentVector, @embedding)

The query vector and stored vectors must come from the same embedding model and have matching dimensions. Use quantizedFlat for smaller sets and diskANN when scale and latency matter more than exact recall.

Exclude the vector path from the regular index so normal writes do not also maintain an unnecessary scalar index.

Key points

  • : fixed-length number list; same model for stored data and query
  • Vector policy: path, data type, dimensions, ; fixed after creation
  • : `flat` exact, `quantizedFlat` small sets, `` approximate at scale
  • Query: `ORDER BY VectorDistance` plus TOP N; WHERE narrows the vectors ranked

Click an underlined term for a one-line explanation, or see them all in the glossary .

Exam trap

Leaving out TOP N, or using a flat index: flat allows fewer dimensions than most models return, so use quantizedFlat or .

Sources · 3 Microsoft Learn pages Show ▾

Checked on October 4, 2026. Report a problem

Go deeper on Microsoft Learn

Checked against Microsoft Learn on October 4, 2026. Report a problem with this card

How well do you know this?