Trace

The recorded steps of one run of an AI agent, in order.

By Mackenzie Bowes · mackenzie@mackenziebowes.com · @_mackenziebowes on X

A trace is a collection of runs for a single operation.

From Observability concepts (LangSmith docs, LangChain, undated).

In LangSmith a run is one unit of work, "such as a model call, tool invocation, or information retrieval", and a trace groups the runs that one request caused. If a user request makes an agent call a model, run a tool and call the model again, all three runs belong to one trace. The word comes from distributed-systems tracing, where a trace is the path of a request through many services.

A trace is the raw record that agent observability is built from.