Inspect recorded agent activity, from the conversation to tool inputs and results. Review token usage, costs, duration, and errors so your team can investigate what happened and decide where to look next.
The detail available depends on what was recorded for the selected run and session.
Conceptual overview. Not a live product screen.
Working with Observability
Give investigation something concrete.
Find the relevant activity
Filter the recorded run list by type, agent, model, or project. Open a run to inspect its details and available session history.
Follow the tool trace
Expand recorded tool calls to examine inputs, results, and errors. Follow sub-agent activity where it is included in the trace.
Understand resource use
Review token and cache usage alongside cost and duration metrics. Use the available detail to investigate resource-intensive work.
Take evidence back to configuration
Identify the instruction, tool choice, or model setting worth investigating next. Observability supports your analysis; it does not replace judgement about the output.