Existing gateway integrations

Existing gateway integrations

APIM

Use Client → APIM → InferQoS → Azure provisioned deployment. APIM retains authentication, subscription, governance, and coarse rate limits. Strip all inbound X-InferQoS-* headers, then set tenant/application from verified APIM identity; preserve traceparent. Use private endpoints and allow only the APIM subnet. InferQoS handles capacity-aware admission, not APIM replacement.

Kong and Envoy

Route OpenAI-compatible paths to InferQoS, preserve streaming/timeouts and tracing, remove untrusted QoS identity headers, and add trusted identity metadata after gateway authentication. Keep retries disabled after response bytes begin. InferQoS forwards to a static configured pool; it is not a semantic router.