Run Trace
AI · RunTrace
Waterfall of one agent run: every model call, tool call, search and approval, with timing bars and expandable details.
Props: steps ({kind: "model" | "tool" | "search" | "memory" | "approval" | "message" | "error", label, meta, start (ms), ms, detail (string or node), status ("done" | "running" | "error")}), defaultOpen (index), footer.
- The bar shows when a step ran inside the whole run, so slow tools are obvious at a glance.
- Errors are blush; a running step pulses. Put model, tokens and cost in the footer.
- Use it in run history and when an approval asks "why does the agent want this?".
StepTimeline · 5.7 s
1. Read new mail 2. Tag each thread 3. Draft replies for billing questions
Model Claude Sonnet 5.5Tokens 11,960Cost €0.036
Props
RunTrace
| Prop | Type | Notes |
|---|---|---|
steps | TraceStep[] | Required |
defaultOpen? | number | null | |
footer? | ReactNode | |
className? | string | Extra class on the root element. |
style? | CSSProperties | Inline style on the root element. |