Agents rely on search as their primary interface to the world, but stale search results are a correctness problem rather than just a UX annoyance. The core challenge is maintaining computed search documents (denormalized entities assembled from joins, aggregates, pricing logic, and embeddings) as underlying source systems change. Batch reprocessing introduces lag; hand-rolled CDC pipelines are brittle and complex. Materialize addresses this by letting teams define computed entities as SQL views that are continuously maintained as inputs change, then propagating entity-level before-and-after diffs downstream. A companion open-source tool, Perfect Embedding, sits as a Kafka Connect SMT between Materialize and the search index, regenerating vector embeddings only when the fields that drive semantic meaning actually change.
Table of contents
Search maintenance used to be a UX problem. For agents, it is a correctness problem.The agent sees computed entitiesThe hard part is maintaining the computed entityBatch and DIY streaming both fall shortDefine the computed entity once, as SQLSend entity-level changes downstreamBuild search your agents can trust138 Impressions1 Comment