Databricks AI Search added Streaming Table, Materialized View and Iceberg managed table sources

The main product change in October was a wider set of sources for AI Search. On October 26, Databricks announced native support for Streaming Tables, Materialized Views and managed tables using Iceberg v3 and later, in addition to Delta tables [3]. The same UI or API workflow applies across these sources, so teams did not need separate synchronization pipelines [3].

SQL-based vector search was the other major theme. Databricks Runtime introduced the NEAREST BY join on October 5. It runs exact or approximate batch vector searches against Lakehouse data, using a fused Photon operator, and it turns liquid-clustered Delta tables into partition-pruned vector indexes with no external vector store to sync [11]. Community Bricktalk sessions on SQL-native vector search and on NEAREST BY followed on October 12 and October 30 [9][1].

Around the platform, Lakebase received attention. Lakebase Search was announced for full-text and vector search in Postgres [10], and a video showed the lakebase vector and lakebase text extensions for hybrid search in agents [4]. A reference architecture combined AI Search for candidate retrieval with Lakebase for online features and Model Serving for inference in an e-commerce recommender [12]. A separate demo covered an agent workflow that begins with vector search indexing and ends with deployment and review apps [7].

SDK releases v0.156.0, v0.170.0 and v0.191.0 added methods to start and stop vector search endpoints, with new starting, stopping and stopped statuses [2][5][6].

Not every report was positive. One community thread described the Databricks-managed Vector Search index as frustrating [8].