Model Gateway — References · Model Management and Inference

Specifications and documentation relevant to operating a proxy on the model path.

HTTP and proxy semantics

RFC 9110 — HTTP Semantics — forwarding, intermediaries, and the status-code semantics a gateway must preserve rather than invent.
RFC 6585 — Additional HTTP Status Codes — defines 429 Too Many Requests, the signal upstream providers use for rate limiting.
RFC 9457 — Problem Details for HTTP APIs — a standard error body, so callers can distinguish a gateway refusal from an upstream failure.
Server-sent events, WHATWG HTML Living Standard — the streaming format the gateway has to forward without buffering.

Envoy documentation — timeouts, retries, connection pooling, and streaming behaviour described precisely enough to configure deliberately.
Kubernetes Gateway API — the route and policy model for gateways running inside a cluster.

RFC 6749 — The OAuth 2.0 Authorization Framework — relevant where caller identity must reach the usage record rather than stopping at the edge.
OpenTelemetry semantic conventions — including the generative-AI conventions for recording model, token counts, and errors on spans in a form other tools can read.