Readiness check for an orchestrator
Whether this instance can reach the dependencies it needs. The status code is the verdict, so a plain probe suffices, and each dependency reports one word only.
GET
Use this as an orchestrator’s readiness probe. It takes no credential. The status code is the verdict:
200 when every dependency answered, 503 when at least one did not, so a plain HTTP probe is enough and nothing has to parse the body.
Each dependency in checks is reported as exactly ok or unavailable, never with a reason. The reason is written to the server logs.
Unlike Health check, this route is rate limited, by the caller’s network address, because every request does real work against the database and the rate-limit store. If the rate-limit store itself is unreachable, this route returns 429.
Response
string
required
ready when every entry in checks is ok, otherwise not_ready.object
required
One entry per dependency that decides readiness. Each value is
ok or unavailable.string
required
Whether the background scheduler is running:
ok, stalled (it stopped reporting in) or
unavailable (its state could not be read). Reported for monitoring only. It never affects
status or the status code.string
required
Whether this deployment can answer questions with its default models:
ok; unconfigured when no
model provider key is configured at all; incomplete when at least one key is configured but the
default assistant model or the default embedding model cannot be served by a configured provider;
unavailable when the state could not be assessed. Reported for monitoring only. It never affects
status or the status code: a deployment that cannot answer can still sign people in, serve
administration and accept knowledge sources.Last modified on September 27, 2026