The signed-in /whisper page includes a Whisper workers card that shows the current execution choices available to the queue.
Remote workers use authenticated heartbeat state. A fresh idle remote worker is shown as Available. While that worker owns active remote work it is shown as Running or Transcribing, and the card may show the current safe progress percentage. Once its heartbeat is outside the configured freshness window it is shown as Offline.
The Tools local entry is different. It is the Whisper runtime on the main Tools host and does not pretend to have a remote-worker heartbeat. While any configured remote worker has fresh authenticated presence, Tools local is shown as Standby and does not claim automatic Whisper transcription, even when the remote worker is already busy. The Tools host may still download and stage URL media required for lease-bound remote transfer. When no remote worker is online, enabled Tools local becomes the scheduled fallback. When local execution is disabled, the card shows Disabled.
An administrator can still explicitly target Tools local for a job. That explicit target is the operator override to the normal remote-first automatic scheduling rule.
The local model list reflects the administrator-managed Tools-host model allowlist. Remote workers continue to advertise their own model capabilities independently.
The runner-status summary on a signed-in Whisper job identifies the executor that is actually handling the current step. Local execution is described explicitly as the Local ToolsAPI host and includes the safe local process-runner id. Remote transcription names the exact authenticated remote worker id and, when available, its bounded device and compute-type telemetry.
The same summary shows the scheduling target separately. Automatic selection means the scheduler was free to choose an eligible runner, Tools local means the local host was explicitly selected, and a named remote worker means that exact worker was selected. This distinction is important for URL jobs because ToolsAPI may perform local media staging before a remote worker starts the actual transcription.
Worker credentials, lease identifiers, source media, transcripts, active job ids and filesystem paths are not included in the runner presentation.
The card summarizes actual completed Whisper jobs rather than a synthetic benchmark:
x realtime.The throughput definition is the same one used on Whisper job details: source media duration divided by measured processing time. Jobs without enough duration/timing information are not counted.
Sampling is bounded separately per runner, and same-model local comparison history is also selected separately. A busy runner therefore cannot push another configured runner's valid history out of the comparison window.
Performance and activity presentation reads only the compact timing, runner, status, progress and current-step facts it needs. Transcript text, transcript segments, media content, worker credentials, leases and raw runtime metadata are not exposed by this presentation.