Skip to main content
Gatelit ships with a built-in model catalog covering OpenAI, Anthropic, Google, and Mistral — 339 models with pricing, capabilities, and lifecycle data. No runtime network calls.

​
How the catalog works

The catalog lives in @gatelit/model-catalog and is synced from provider APIs. The gateway uses it to:
  • Clamp parameters — strip temperature from reasoning models, rename max_tokens → max_completion_tokens
  • Show capabilities — the dashboard model picker shows context windows, pricing, and supported features
  • Flag deprecations — models with a deprecated_at date get a warning badge; retired models are hidden

​
Enabling and disabling models

Go to Settings → Models. Each model has a toggle. Disabled models:
  • Won’t appear in the playground model picker
  • Won’t appear in prompt editor model selectors
  • Still work through the API — model toggles are a UX convenience, not a security boundary
To enforce which models an org can use, pass a model to the gateway’s token exchange endpoint or use org-scoped policies.

​
Model lifecycle

The catalog tracks three states: The catalog is updated by running pnpm dlx tsx packages/model-catalog/scripts/sync.ts — usually done during deployment.

​
Model format

Always provider/model-id: The provider prefix tells the gateway which upstream API to call. Everything after / is passed to the provider as the model identifier.
Model management