local-provider-loaded-model-safety

Gateway policy aliases can override local models when alias resolution bleeds into transport-level forwarding. Separating policy alias resolution from model forwarding prevents the gateway from unintentionally switching your manually loaded local provider model.

Configure the provider adapter to forward request.model only if it matches an explicit local-provider model id. This ensures the manually loaded local model remains active and prevents gateway aliases from being forwarded as transport-level model ids.

Record both requestedModel and effectiveModel during provider interactions. This logs the original gateway alias request alongside the actual local model used, preserving traceability and enabling easier debugging of gateway-to-provider routing paths.

The active local model switches because the policy alias is forwarded as a transport-level model id. Requiring the provider adapter to forward explicit local-provider model ids only stops this unintentional model switching behavior.

Yes, this approach supports regression coverage to verify that gateway policy paths do not unintentionally switch the active local model. This validates that alias resolution and transport-level forwarding remain properly separated.