Hub
Provider Status
Status observation hub. WebmasterID Models records vendor-reported status as a vendor-reported observation, runs independent HTTP probes against non-inference public endpoints, and keeps the two signals strictly separate. This page does not publish an uptime percentage.
Per provider
Observers (3 total across 2 providers)
Each provider card lists every observer registered against it. Vendor-status observers read the provider's own public feed; independent HTTP probes hit a public, non-inference endpoint from WebmasterID.
- Anthropic
2 observers enabled
- Vendor status API
Vendor status observation
Reads the public Anthropic Statuspage JSON feed (vendor-reported status, NOT an independent uptime probe).
- Independent HTTP probe
Independent HTTP probe
Independent HTTP probe of api.anthropic.com — host-root reachability only, NOT an inference call.
- Vendor status page
- status.anthropic.com
- Run all observers now
- /api/status/anthropic
- Latest stored observation
- /api/status/anthropic/latest
- 24h window
- /api/status/anthropic/window?hours=24
- OpenAI
No observer enabled
- Vendor status page
- status.openai.com
- Google
1 observer enabled
- Vendor status API
Vendor status observation
Reads the Google Cloud public incidents feed and reports the headline state of active incidents touching Gemini / Vertex AI / AI Studio. Vendor-reported, NOT an independent uptime probe.
- Vendor status page
- status.cloud.google.com
- Run all observers now
- /api/status/google
- Latest stored observation
- /api/status/google/latest
- 24h window
- /api/status/google/window?hours=24
- Meta
No observer enabled
- Vendor status page
- Data not yet verified.
- Mistral
No observer enabled
- Vendor status page
- Data not yet verified.
- DeepSeek
No observer enabled
- Vendor status page
- status.deepseek.com
- Groq
No observer enabled
- Vendor status page
- groqstatus.com
- Together AI
No observer enabled
- Vendor status page
- Data not yet verified.
Source: independent probe
Independent HTTP probes (1 active)
Independent probes hit a public, non-inference endpoint and report whether the host responded. No API key is sent, no inference is triggered, no billing is invoked. Probe wall-clock time is never relabelled as the provider's request latency.
Persistence
Durable observation storage
Observations only become useful as a window once they are persisted. The storage layer is optional; when it is not configured, observations are fire-and-forget and no window can be computed.
- Storage state
- Durable storage configured
- Providers with observers
- 2
- Read endpoint shape
- /api/status/[provider]/latest
- Uptime calculation
- Not yet eligible — requires 24 stored observations in the window.
Storage is configured via the KV_REST_API_URL / KV_REST_API_TOKEN env pair on the Vercel project. When unset, the cron still runs but each observation reports skipped_no_store and nothing is persisted.
Honest gaps
Why no uptime percentage is shown
- Uptime is a fraction of observations over time. Until the cron has persisted at least 24 observations in a window, no honest denominator exists.
- Vendor-reported indicators are useful as a colour signal, but they are not an availability measurement. Probe-success rate is a reachability signal, not full API uptime. Any future published number computed from either will be labelled precisely — "vendor-reported operational-sample rate" or "probe-reachability sample rate" — never "uptime" without qualification.
- Once the threshold is met, the per-provider window endpoint will expose a `uptimePercentage` field with the gating decision explained in its `policyNote`. The /status page itself does not display this number.
How observations are scheduled
Every observer is wired to /api/cron/status, which Vercel Cron calls hourly. When CRON_SECRET is set on the deployment, the endpoint requires a bearer token. In production, the cron refuses to run unguarded.