List the catalog, or look up one model
GET /v1/models and GET /v1/models/{owner}/{repo} both answer locally from the gateway’s
catalog, in the same shape as the live router’s own model-list response (each admitted
provider’s pricing, the gateway’s 5.5% fee included, with the pinned default listed
first) — nothing is proxied to Hugging Face for a GET.
Authentication
Your per-client bearer token — the API key the gateway issued to you. Present it in
Authorization: Bearer <token> form. A token that is missing, unknown, or revoked draws a
401. Before the call is sent upstream the token is removed (stripped, or swapped for the
upstream’s own credential); it never makes it past the gateway.
Path parameters
The path to call on Hugging Face’s Inference Providers router, any further / segments
included — all of it after /huggingface/ is passed to https://router.huggingface.co/
when the gateway’s (model, provider) catalog admits the request (see the Hugging Face tag
description above). Only two forms are admitted: v1/chat/completions and
v1/models[/{owner}/{repo}].
Response
Whatever the upstream answered, relayed as-is (status, headers, body). Rendered here as JSON; the true shape is the upstream provider’s to define.

