Skip to main content
Clients that probe client.models.list() against an OpenAI-compatible base_url hit this route. Auth and the consumer slug are the same as chat.
On a Private data plane, add X-AG-Gateway-Slug as in the chat snippets. GET /v1/models returns { "object": "list", "data": [{ "id", "object": "model", "owned_by" }] }. GET /v1/models/{id} is 200 when that id is in the list, otherwise 404. Non-GET methods are 400. Extra path tails (/models/{id}/extra) are 404. This is not a forward of the provider’s /v1/models and not the admin models catalog. The list is the union of native slugs the consumer can actually call.

​
What is listed

For each bound registry (plus fallbacks, role policies, and load-balancing member models):
  1. If the consumer (or role) has an allow-list, those ids are candidates.
  2. If the policy is open, TrustGate expands from the provider catalog.
  3. Only native slugs are kept — no auto, no @provider/…, no pool:….
  4. A slug is kept only when the registry’s provider supports that modality (chat always; embeddings / rerank only when the provider advertises them). Catalog flags win; otherwise the slug is inferred (embed → embeddings, rerank → rerank, else chat).
owned_by is the registry provider (for example openai or anthropic). Files have no model ids and are not listed. Images ids appear only when the catalog (or allow-list) includes them and the consumer has an images-capable registry.