Trace
The complete step-by-step record of a single agent run, from the request to the final answer.
Think of it like
A hiking log with every turn and rest stop marked — the whole route, not just the destination.
Example
The trace shows the agent’s five steps: read the ticket, search docs, draft a reply, check policy, send.
How it actually works
A trace is the top-level container for everything that happened in one run — a tree of steps, each with its inputs, outputs, and timing. It’s the primary unit you inspect when debugging and the raw material for evaluation. The concept comes straight from distributed-systems tracing, adapted to the reality that an agent’s “request” fans out into many model calls and tool invocations.
For product teams
The trace is the story of one run — the first thing to open when something went wrong.
For engineers
A tree of spans capturing the full execution of one agent invocation, end to end.
Related
- Agent Observability — What traces make possible.
- Span — The individual steps inside a trace.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome