It aggregates messages from humans, models, and tools across a main agent and its subagents, presenting each message once in order rather than exposing the full nested run structure. The underlying trace remains available for execution details such as inputs, outputs, timing, retries, and metadata.
The company says users can view threads as trajectories, score them with online evaluators, and route them to annotation queues or datasets. It describes uses including locating behavioral deviations in long sessions, enabling subject-matter-expert review, and exporting production examples for supervised fine-tuning workflows. Trajectories are available for traces from LangChain, LangGraph, Deep Agents, several agent SDKs, and coding agents. LangChain says the feature is available now on all LangSmith plans in the US.