Gateway API
Status unavailable30-day uptime · history begins at launch
Status
Network status
Live aggregates from the public status endpoint. The page shows what is actually reporting; it does not show worker locations, per-worker data, or a map.
Contributor and API access are invite-only. Request access.
Overview
At a glance
Counts cover the rolling 24 hours ending at the update time. A dash means the endpoint does not report that value.
- Workers online
- — of — enrolled
- Models live
- — serving at least one worker
- Jobs 24h
- — user jobs
- Tokens 24h
- — output tokens
- p50 first token
- — median across user jobs, 24h
- Queue
- — not reported by the public API
Components
Scheduler
Status unavailable30-day uptime · history begins at launch
Worker network
Status unavailable30-day uptime · history begins at launch
Models
Model availability
Only models with at least one active worker appear. Registered models without a live worker are omitted rather than shown as serving.
| Model | Status | Active workers | p50 TTFT |
|---|---|---|---|
| Status data has not loaded. | |||
Capacity by model
Active workers by model, shown as a share of the active network. GPU-hours are not reported by the public API, so this page does not invent them.
| Model | Active workers | Share of active workers |
|---|---|---|
| Status data has not loaded. | ||
Incidents
Incident history
Incidents are published here when they affect the public API, scheduler, or worker network.
No incidents recorded.
Method
How to read this page
Definitions used by the status endpoint and this page. Windows are rolling and times are UTC.
- p50 first token
- Median time from the request being accepted to the first streamed token, across user jobs in the last 24 hours.
- Active worker
- A worker whose heartbeat is no more than 60 seconds old and that is not quarantined.
- Models live
- Models with at least one active worker. Registered models without a live worker are hidden from the availability table.
- Jobs and tokens
- User jobs and output tokens in the rolling 24-hour window ending at the update time. Verification jobs are excluded.
- Freshness
- The API caches a snapshot for up to 30 seconds. This page refreshes every 30 seconds while the tab is visible. A failed refresh keeps the last values and marks them Data delayed.
- Queue
- Not reported by the public API. The value stays — until the endpoint exposes it.
- Capacity
- Active workers by model. GPU-hours are not reported by the public API. No worker map or location data is shown.
- Timezone
- All times are UTC. Counts are rolling windows, not calendar-day totals.