Databricks Vector Search

Databricks Vector Search (now folded into the broader Databricks AI Search product) is a fully managed vector database and retrieval engine built directly on top of the Databricks Data Intelligence Platform. It combines semantic (embedding), keyword (BM25) and hybrid search behind a single API, with built-in reranking, quality evaluation, and serverless autoscaling to billions of records. The headline feature is automatic index sync: point it at a Delta table and Databricks handles embedding generation, incremental updates, and retries without you gluing together a pipeline.

It is aimed squarely at teams already on Databricks who are building RAG apps, agentic systems, product/e-commerce search, or recommendation pipelines and want retrieval to sit inside Unity Catalog's governance boundary rather than in a separate vendor. Access controls, lineage, and fine-grained policies from Unity Catalog carry through to the index, which is a genuine differentiator against standalone vector DBs like Pinecone or Weaviate. Pricing is enterprise / consumption-based via Databricks billing; there is a free trial via the Databricks platform trial.

It integrates natively with Databricks Model Serving, MLflow, Agent Bricks, and Mosaic AI, and exposes a REST API plus Python SDK so it plugs into LangChain, LlamaIndex, and custom retrieval stacks. The obvious caveat: it only makes sense if you are (or plan to be) a Databricks customer — outside that ecosystem the pricing and setup overhead don't compete with dedicated vector stores.