LangChain launches AI Search for querying agent traces in LangSmith

LangChain announces AI Search, a natural-language way to find agent traces in LangSmith by described behavior, such as PII leaks or frustrated users.

The search runs a decision model over trajectories that SmithDB builds by removing repeated messages from each session. The model reads an 8K token budget filled in priority order, starting with user messages and agent replies.

AI Search is in public beta on LangSmith Cloud in the GCP US, EU, and APAC regions at no additional cost, and org admins can enable it in settings.