Provider failover and a clearer catalogue
Onde Cloud v1.1.0 turns the model catalogue into an ordered routing engine. A public model id can now have more than one candidate behind it, and the handler tries available providers in declaration order while keeping the client-facing model surface unchanged.
Provider failover
Requests now resolve to provider candidates instead of a single backend target. This lets a tier fall through to its next configured candidate when an upstream fails, without exposing the provider name through public ids, response model fields, or /v1/models.
Baseten support
The release adds a Baseten provider adapter and wires it into several routed tiers. Baseten remains hidden behind the same neutral Onde model ids as the rest of the catalogue.
Bounded upstream calls
The shared HTTP client now uses explicit connect and request timeouts, so a wedged provider cannot hold an Onde Cloud request open indefinitely.
Product taxonomy
The docs now record the three product layers: Onde Inference as the platform, Onde Cloud as the router service, and provider candidates inside the router, including first-party Onde Cloud inference.