databricks-vector-search

To set up Databricks Vector Search, create a vector search endpoint and Delta Sync index. You must define index specs like primary key, embedding source, pipeline type, and columns to sync to enable semantic retrieval for RAG.

Delta Sync automatically syncs vector indexes from Delta tables using managed or self-managed embeddings. Direct Access requires manual upserts and deletes, offering different storage and freshness trade-offs for your vector search.

Yes, Databricks Vector Search supports hybrid search. You can apply HYBRID mode combining vector and keyword ranking, and use attribute filters to ensure exact terms like IDs or error codes are matched alongside semantic results.

Yes, Unity Catalog is required. Databricks Vector Search relies on Unity Catalog to manage Delta tables and synchronize them into vector indexes, ensuring governed access for your semantic search endpoints.

You can query Databricks Vector Search using query_text for natural language, query_vector for precomputed embeddings, or hybrid ranking. Query parameters must match embedding dimensions and filter syntax to retrieve top-k results.

Storage-optimized vector search endpoints are designed for specific scale and cost profiles. When choosing between Standard and Storage-Optimized, consider your storage capacity and query latency requirements for your specific vector index lifecycle.