AI Traces
The AI Traces page gives you low-level distributed-tracing visibility across the AI Foundry services themselves. It complements the session-level view on the Playbook Sessions page: use AI Traces when you need to diagnose latency and errors at the span and service level, rather than at the conversational-session level.
The page is split into two tabs:
- Overview (span and latency analytics)
- Traces (a searchable log of individual traces).
Filter bar
Both tabs share a filter bar at the top:
| Filter | Description |
|---|---|
| Services | Multi-select one or more services to narrow the analytics and trace list to their activity. |
| Date range | Pick a start and end date. Analytics and traces update to cover only the selected time window. |
Your administrator may scope which services appear in the services selector. In some deployments, only a fixed, pre-configured set of services is tracked and made available for filtering.
Overview tab
The Overview tab shows span and latency analytics for the selected services and date range, including:
- A per-operation latency table, listing the slowest or most frequent operations across the selected services.
- A per-service latency chart, showing how latency trends over time for each selected service.
Use this tab to spot which operations or services are contributing most to overall latency before drilling into individual traces.
Traces tab
The Traces tab lists individual traces for the selected services and date range in a searchable, paginated table.
Columns
| Column | Description |
|---|---|
| Status | Status indicator for the trace. Currently always shown as OK; per-trace error detection is not yet available. |
| Trace ID | Truncated trace identifier; hover to see the full ID. |
| Service | The root service that started the trace. |
| Root Operation | The top-level operation name for the trace. |
| Duration | Total wall-clock duration of the trace. |
| Start Time | When the trace started. |
Opening a trace
Click any row to open the Trace Detail view for that trace, showing the full span waterfall so you can inspect the timing and relationships between all spans that make up the trace.
See also
- Observability: session-level view of agent and playbook runs — start there for conversational debugging.
- Agent: the execution unit whose calls generate spans traced here.