Reference · Glossary
Probe endpoint
An HTTP route (or exec check) platforms poll for **liveness or readiness** — kept fast and idempotent.
#When to use
Any containerized AI API behind Kubernetes, ECS, or managed serverless with health routing.
#When not to
Reusing heavy `/chat` for probes — it skews latency metrics and loads the model.
#Example
`/health/ready` pings vector DB with `SELECT 1` cached for 5s; `/health/live` only checks process heartbeat.