Understand performance
Find where time and model costs go, and inspect user feedback.
Investigate a slow run
Separate model latency, tool execution, and time waiting for a person so you can find what made a workflow run slow.
Understand tokens and cost
Read recorded token usage and model costs, distinguish partial totals from complete prices, and compare similar workflow runs.
Collect and inspect feedback
Connect authenticated response ratings to recorded runs and distinguish user feedback from execution failures and evaluation grades.