Search agent traces by behavior with AI Search in LangSmith

LangSmith has introduced AI Search, a natural-language trace-querying feature built on SmithDB to help teams identify agent behaviors, failures, user sentiment, potential PII leaks, product feedback, and other patterns that cannot reliably be found through metadata filters or full-text and vector search alone. Rather than relying only on embeddings, AI Search uses fast decision models to evaluate whether each candidate trace trajectory matches a user’s description after structured filters narrow the search set. SmithDB converts varied, often repetitive trace formats into standardized session-level trajectories that retain messages, tool calls, and results while eliminating duplicated conversation history. To handle very long agent sessions efficiently, searches prioritize user messages and final agent responses within an 8K-token context budget, followed by reasoning, tool activity, and system content. Results are processed in parallel and streamed as matches are found, while future work will explore question-specific context selection, uncertainty handling, improved classification of subtle interactions, and grouping or summarizing similar results. AI Search is available in public beta on LangSmith Cloud in GCP US, EU, and APAC regions at no additional cost.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.