Skip to main content
A workflow run records one execution of an agent. The API request, provider call, and conversation outcome are separate stages: an accepted request can have a run even if the carrier does not connect or the workflow later fails.

​
Follow a run

Keep the workflow_run_id returned by a call or text-session request. Open the run from the agent’s run list, or use the run detail API for that ID. The detail view can include timestamps, duration, user and assistant turn counts, tool calls, visited nodes, transcript or recording links when present, activity, logs, context, tool results, and cost information. Use the activity timeline to locate the step where behavior changed. Compare the inputs and configuration recorded for the run with the intended setup. A missing transcript or recording can mean that the artifact was not created or is unavailable for that run; it does not by itself mean the run failed. For voice calls, Athena Live activity shows time-ordered events such as decisions, key presses, tool calls, pickup checks, phone-menu activity, call/session events, and speech. Filter by event type or search the timeline. Some events include expandable details or a short decision summary when one was recorded.

​
Verify in stages

For a phone integration, check each stage independently:
  1. Athena accepted the request and returned a run ID.
  2. The selected provider accepted the call.
  3. A controlled handset rang and the call connected.
  4. The caller and agent could hear one another.
  5. The run completed with the expected result, and configured exports or webhooks arrived.
The HTTP response only confirms the initiation stage. If a request times out, inspect the run before sending another request, because the first call may already have started. See Trigger a phone call. When a run has an available recording or transcript, the dashboard can preview or download it. The download link is issued through a media token URL that expires after 15 minutes. Treat that URL as temporary access: do not save it in logs, tickets, analytics, or customer-visible messages. A run may have no downloadable artifact, in which case the media request is unavailable. The API checks access to the run before issuing a media URL. Use an authenticated request from your backend when accessing run media; do not share a media token as a substitute for normal access control. When separate user and agent recordings are present, the run page shows Split Tracks with aligned waveforms. You can listen to both tracks together or isolate the user or agent track, seek, and download each recording. This section appears only when those artifacts are available.

​
Reports and exports

Agent reports cover one workflow; campaign reports cover the campaign and its calls. Daily Reports can be filtered by date, time zone, and agent. Depending on the report, the dashboard can show outcome distributions and export a CSV with call details and links to runs. CSV reports can include phone numbers and call outcomes. Limit access to the download, store it according to your organization’s data-handling rules, and remove test exports containing synthetic or personal data when they are no longer needed. Check the response content type before parsing an export as JSON. See Webhooks and call events for delivery paths that send results to your service.