Routing & failover

Routing resolves a requested model profile against provider availability, subscription eligibility, quota, and operator policy.

Selection precedence

1. Invocation --model.
2. JEDEN_MODEL.
3. Interactive or persisted native model selection.
4. Operator default.

Stable logical request

Request, idempotency, and decision identities remain stable across eligible attempts. Each attempt records the route and billing target it actually served.

Failover boundary

Typed quota exhaustion may place a target into a Retry-After-bounded cooldown and select the next eligible target. Failover stops once model output is visible.