01 / The short version
What happened
LangSmith announced AI Search, a natural-language query feature for agent traces built on its SmithDB database. Users describe behaviors to find matching traces, combining filters with decision-model inference that evaluates trajectories rather than raw traces. SmithDB normalizes inputs from multiple SDKs into trajectories, deduplicates repeated messages, and applies a fixed token budget prioritizing user and agent messages. Matches stream back as found. The publisher notes ongoing work on question-specific context selection and uncertainty signals.
See the exact references02 / Key takeaways
What you need to know
- 01
AI Search pairs natural-language queries with decision models over SmithDB trajectories.
- 02
SmithDB normalizes traces from OpenAI, Anthropic, LangChain, LangGraph, Vercel AI SDK, and OpenTelemetry.
- 03
A fixed token budget prioritizes user and agent messages, trimming or dropping later content.
Keep in perspective
What to watch for
The announcement does not specify pricing, availability, or benchmark results.
Go to the source
Exact references
These are the original pages used for this brief. Publisher claims are not independent evaluations.
01Primary source · LangChainRead the original announcementhttps://www.langchain.com/blog/agent-traces-ai-search-langsmithAI-generated from the linked source. It can miss context; verify consequential details in the original. How the radar works