Watchby Tensor Dynamics

Sentinel API v1

From your first API call to production

Human workflow

Incidents and controlled actions

Events are machine observations; Incidents are human work. Physical actions use separately enrolled devices, bounded semantic operations, expiry, and human confirmation.

Incident lifecycle

open → claimed → closed
  ↑       ↓       │
  └── released    └── reopen
GET  /v1/incidents
GET  /v1/incidents/{incident_id}
POST /v1/incidents/{incident_id}/claim
POST /v1/incidents/{incident_id}/release
POST /v1/incidents/{incident_id}/close
POST /v1/incidents/{incident_id}/reopen
POST /v1/incidents/{incident_id}/notes
PUT  /v1/incidents/{incident_id}/tags

An analytic cleared event never closes an Incident automatically. Downstream systems should use Incident status as human-workflow truth while retaining Event lifecycle as analytic evidence.

Action safety boundary

  • Models may suggest only enrolled semantic operations; they cannot supply URLs, shell, PLC addresses, or code.
  • Every action requires an Idempotency-Key, bounded expiry/duration, and authorization.
  • Actions configured for human confirmation do not execute before approval.
  • Inference, webhooks, and third-party actions never block the capture/inference loop.
Action APIs are public only when listed in /v1/openapi.json. Treat the current OpenAPI as authoritative and never infer /v1 paths from legacy /api/sdk routes.