databricks-vector-search

To create a Databricks vector search endpoint for RAG, you configure either a Standard or Storage-Optimized endpoint. You then attach a Delta Sync or Direct Access index to enable large-scale semantic querying across Delta tables.

Yes, Delta Lake vector search supports self-managed embeddings alongside managed embeddings. You can tailor embeddings to your data using Delta Sync indexes, allowing flexible integration of custom embedding models into your semantic search workflow.

Delta Sync indexes automatically sync with Delta tables and manage embeddings for you, while Direct Access indexes require self-managed embeddings. Both support text, vector, and hybrid queries with filters across Delta Lake sources for semantic search.

You query a Databricks vector search index using text, vector, or hybrid queries with applied filters. The index processes these queries against your Delta Lake data to return semantically similar results for your RAG or search applications.

Yes, Databricks vector search works directly with Delta Lake tables to power semantic search. You create indexes on Delta table data, enabling fast vector similarity searches and retrieval-augmented generation workflows.

Use a Standard endpoint for general RAG workloads and a Storage-Optimized endpoint when managing larger vector indexes. Both endpoint types support Delta Sync and Direct Access indexes for querying Delta Lake sources.