Databricks AI Search
Databricks AI Search is the current Databricks name, previously Mosaic AI Vector Search (since June 2026).
Databricks Vector Search Mosaic AI Vector Search Databricks AI Search
Managed search over Unity Catalog data, covering vector similarity plus hybrid keyword-similarity search over the same index.
- It outgrew pure vector search - on a storage-optimized endpoint you can now build an index with no embedding columns at all and search it with BM25 keyword scoring, a capability still in Beta.
- Its high-QPS mode is enabled by default and yet changes nothing for the endpoints you already run - existing endpoints are pointedly left alone.
Limitations: Row- and column-level permissions aren't supported, indexes can't be cloned into another workspace, the _id column name is reserved, and index capacity is provisioned from the source table size at creation, so starting small risks capacity-exhausted errors. Query results default to 10,000 rows for ANN and 200 for hybrid or full-text search with a 10 MB response limit, all raisable on request, and storage-optimized endpoints add limits such as no continuous sync, embedding dimensions divisible by 16, and no CMK or FedRAMP support.
Open in REbricked →- Category
- AI / ML
- Also known as
- Mosaic AI Vector Search, Databricks Vector Search, Vector Search
- Verified
- 2026-08-20