Health

Health is reported by your deployment, as operational telemetry, and by each component’s own endpoints.

In the API

GET /api/v1/health (viewer): for each deployment, the newest health report of each component over the last 7 days, and each connector’s latest pull over the last 30. MCP: health.

What each check means

Each report names a component, its status and a fixed detail code.

Component	What it is
gateway	The model gateway
orchestrator	The runner of your agents
job	A scheduled job
connector	A source’s pull

Status	Means
ok	Working
degraded	Working, but not as it should
down	Not working

Detail code	Means
ok, started, stopped	Normal life
pull-failed	A connector’s pull failed
eval-error	An evaluation could not run
store-unreadable	A store the component needs could not be read
unreachable, timeout	A dependency did not answer
breach	The online monitor found a breach: see Versioning and rollback

Inside your deployment

Endpoint	Answers
Gateway GET /healthz	{"status": "ok"} while the process runs
Gateway GET /readyz	{"status": "ready"} when it can serve; a scale-to-zero gateway waits on it
Relay GET /healthz, GET /readyz	{"ok": true}
Engine services’ actuator	/healthz (alive), /readyz (503 if any check fails), /metrics (Prometheus), /buildinfo

Point your load balancer and your own monitoring at these. The stacks’ load balancers already check /healthz.