databricks-vector-search

To create a vector search index on Databricks, you need to set up an endpoint and choose between Delta Sync or Direct Access index types based on your RAG application's embedding management and data synchronization requirements.

Storage-optimized endpoints should be used when your vector search workloads require cost-effective storage for large datasets, whereas standard endpoints are better suited for standard querying and similarity matching tasks.

Yes, Databricks Vector Search supports semantic, keyword, and hybrid searches. You can execute these queries using advanced filtering options to refine the similarity matching results from your indexes.

Yes, Delta Sync indexes support both managed and self-managed embeddings. This allows you to synchronize data from Delta tables while controlling how embeddings are generated and managed for your pipeline.

To query a Direct Access index for similarity matching, you bypass Delta table synchronization and directly query the index using vector search patterns with advanced filters to retrieve the most relevant documents.

Databricks Vector Search integrates natively with your data platform, enabling Delta Sync for automated embedding updates and providing specialized endpoint options to optimize semantic search and RAG capabilities.