Watchby Tensor Dynamics

Sentinel API v1

From your first API call to production

Production checklist

Before connecting production cameras

A successful test event proves the integration path. Production also requires recoverable data, a protected network boundary, capacity evidence, and rehearsed failure handling.

  1. 01Use a 32+ character bootstrap key in a secret manager, then issue least-privilege keys.
  2. 02Publish the API only on localhost or a TLS-protected private network.
  3. 03Back up data and model volumes, and test the restore procedure.
  4. 04Use a substream and unique credential for each RTSP source; confirm responses never expose raw URLs.
  5. 05Smoke-test a real provider and representative images, not only the mock provider.
  6. 06Verify raw-body HMAC and deduplicate webhooks by event_id plus delivery_id.
  7. 07Monitor /readyz, disk capacity, provider state, offline sources, and dead-letter deliveries.
  8. 08Test runtime restart, source loss, provider loss, and receiver outage.
  9. 09Document retention, access roles, evidence export, and incident-response ownership.
  10. 10Back up before upgrades; inspect the OpenAPI diff and version/deprecation notices.

Capacity is not just a camera count

Measure stream resolution and FPS, VLM image size and cadence, GPU/VRAM, inference queue, event density, evidence retention, and webhook latency. Promise camera count only after a soak test with equivalent hardware, model, and monitor settings.

Availability boundaries

process alive              GET /healthz
database + reconciliation  GET /readyz
node and monitor summary   GET /v1/system
CPU / RAM / GPU / disk     GET /v1/system/resources
provider readiness         GET /v1/system/models
delivery backlog           GET /v1/webhook-deliveries
A hosted multi-tenant control plane, SOC 2, ISO 27001, HIPAA, and self-serve SaaS checkout are not currently claimed as generally available. Run a deployment and security review when procurement requires them.