AI monitor
Loading…
Connecting…
Powered by Claude · demo mode
Continuous anomaly detection across worker telemetry
Each run analyses the last 15 min of pings, check-ins, vitals, sensor readings and weather alerts. The model emits one structured tool-call per anomaly via a strict JSON schema.
Critical
0
High
0
Medium
0
Low
0
Loading alerts…
AI Virtual Watcher · welfare-ping engine
Proactively pings workers Claude thinks need a welfare check
Scheduled per-worker scan. Claude looks at the last 60s of telemetry and decides which workers (if any) need an unprompted “are you OK?” ping. Workers ack with one tap or escalate straight to SOS. Conservative by design — zero pings is a valid and frequent output.
0
Pending ack
0
Acknowledged
0
Total today
No welfare pings yet. Hit “Run scan now” to compose a batch.
Production wiring: this scan runs as a Vercel Cron job every 60s (or a Fly.io scheduled task on the dedicated tier). The button above is the manual-fire path for demos and ops on-call.
How it works: Each run sends the workforce roster + frozen rubric as a cached system prompt (so subsequent runs cost ~10× less per input token), plus the recent-events window as the only volatile content. Claude emits one tool call per distinct anomaly. The schema is strict — no free-form text, no hallucinated worker IDs.