KI-Suche

AI Search is designed for teams building production applications that need real-time search, matching or retrieval over enterprise data, such as product search, content recommendations, e-commerce experiences and AI agents. Lippert uses it to power an agent-assist experience that helps customer support agents find answers from product manuals, YouTube videos and support cases. EQT uses it for search over data already stored in Delta, noting an easy setup completed in a matter of hours. Because it is serverless and governed by Unity Catalog, it fits organizations already running on the Databricks Platform that want search indexes to inherit existing access controls.

AI Search lets you combine semantic vector search with keyword (BM25) search in a single call, then rerank results for production-grade relevance. This hybrid approach is built to power product search, content recommendations, e-commerce experiences and agents on one engine, without operating separate systems for keyword search and vector search. Built-in reranking and quality evaluation are included as part of the same API, so relevance tuning doesn't require a separate service.

AI Search uses automatic sync pipelines that update indexes whenever source tables change, so results stay in step with the underlying data. Databricks handles embedding generation, retries, throughput tuning, batch sizing and autoscaling behind the scenes, removing the need for additional data engineering work. EQT specifically highlighted this integrated managed experience, noting that because their data was already in Delta, the Delta Sync capability let them get up and running in a matter of hours.

AI Search inherits the access controls, lineage and audit logs already defined in Unity Catalog, since each index is managed as a Unity Catalog object. Fine-grained policies set elsewhere in Unity Catalog automatically extend to search indexes, so teams don't need to configure a separate permissions model just for search. This also means lineage tracking and audit logging carry over without extra setup work.

Pricing details are not listed directly on the AI Search product page itself. Databricks publishes current rates on its dedicated AI Search pricing page, where you can review costs before deploying indexes into production. Because AI Search is serverless, there are no clusters, shards or replicas to size separately when estimating usage.

You can start using AI Search by signing up through the Try for free option on the product page, or by scheduling a demo with the Databricks team to see it in action first. For setup details, the AI Search documentation covers how to create indexes, configure sync pipelines and query the API. Because indexes are serverless and scale to billions of records and thousands of queries per second, you can move from initial testing to production scale on the same engine.