Trace
The recorded path of one request through the whole system.
A trace is the recorded path of a single request: the calls it made, the tools it invoked, the documents it retrieved, the tokens it spent, and the order of all of it. In a classical service a trace is a few spans of function calls. In an AI system it is the only way to reconstruct why a particular answer came out the way it did.
Traces matter most when something goes wrong and nobody can tell whether the model was wrong, the retrieval was wrong, or the tool was wrong. A trace separates those three instantly, because it shows what the model was actually looking at when it produced the answer.
The cost is volume, and the fix is sampling: keep every trace for errors and slow requests, and a small fraction of the rest. Storing everything is expensive and usually unnecessary, since the interesting cases are rare and identifiable after the fact.
Related terms
Observability
Being able to see inside a running system well enough to explain what it did.
Monitoring
Watching a running system so you learn about drift before your users complain.
Deployment
Getting a working system in front of real users, reliably, and keeping it there.
Evaluation
Measuring whether the system does the job, on a fixed set of cases, repeatedly.
Tool Call
The model's request naming a tool and its arguments. Structured output, not a side effect.
Want this working inside your team?
It starts with one conversation about your workflow.