← Home
Studio Guide

Governance

Traces

Traces are a step-by-step history of every run: the inputs, the data that went through each block, tool calls, latency, tokens and cost. Use them to find out why an agent gave a particular answer, or why a run failed. Open them from Governance → Traces in the left menu.

1.List view

The List view shows one row per run, with its workflow, status, trigger, tokens, cost, duration and start time. Open a run to see the data that went into and came out of each block.

  • Status: Filter by Success, Failed, Paused, Running or Cancelled. Each tab shows how many runs have that status.
  • Baseline latency warning: A run that took much longer than usual is flagged, such as 4.4x baseline latency. Open it to see which block slowed it down.
  • Trigger: How the run started: chat, manual or schedule.
  • Search: Search across workflow names, inputs and outputs.
  • Group by session: See all the runs from one conversation together.
  • Compare: Compare runs side by side.

2.Run details

Select a run in the list to open it. The run page shows how the run went as a whole, and each step it took.

  • Header: The workflow name, how the run was triggered (such as manual), when it started, and its status.
  • Summary: The run’s total duration, cost, tokens (split into tokens in and tokens out) and number of steps.
  • Steps: Every block the run went through, in order, with the block’s name, its type (such as approval, agent, json_extractor, merge or output), its status and how long it took. A step that took much longer than the others is usually where to look when a run is slow.

From the run page you can also:

  • Select Tag to label the run so you can find it again.
  • Select Replay to run the workflow again with the same input.
  • Select Promote to Eval to save the run as an evaluation case.
  • Select Back to traces to return to the list.

3.Performance view

The Performance view summarizes one agent or workflow over 7d, 30d or 90d. Choose the agent from the list at the top.

MeasureWhat it shows
RunsHow many times the agent ran
Success rateThe share of runs that finished successfully
Total cost and Total tokensWhat the runs cost, and the tokens they used
LatencyHow long runs took: average, p50, p95 and p99
Cost per runWhat a run costs: p50, p95 and p99

A drop in success rate or a jump in cost is a signal to open the failed or expensive runs in the List view.

Traces — Studio DocumentationAgent Studio