Models

The model id you pass in the request body is matched against the gateway's catalog. The list is dynamic — admin can enable, disable, or remap a model without a client release.

Discovering models

Two surfaces expose the catalog:

• Public catalog — the marketing-facing list, what an anonymous visitor would see if they signed up to the free tier.
• GET /v1/models — authenticated, narrows to the models your API key may call.

Both endpoints reflect live admin state — disable an upstream channel and the model disappears from the catalog within 30 seconds.

Routing

When you call /v1/chat/completions with a model field, the gateway:

1. Finds every enabled channel that serves that model.
2. Filters by your tier's group (free / pro / team / enterprise + the shared default group).
3. Picks the highest-priority healthy channel. Ties break on weight, then id.
4. On a transient upstream failure (5xx, timeout, connection reset), retries with the next channel in the chain. Up to three fallback attempts per request.

Model remapping

A channel can remap the user-facing model id to a vendor-specific deployment id (e.g. gpt-4o → my-azure-deployment-name). The remap is invisible to clients — you still bill against the canonical model id, and the response carries the canonical id too.