Models

The gateway implements GET /v1/models. The result contains active or experimental models that are available to the authenticated organization and whose providers are active or degraded.

const page = await client.models.list();

for (const model of page.data) {
  console.log(model.id, model.owned_by);
}

Model IDs are provider-qualified:

openai/gpt-5.4-mini
google/gemini-3.5-flash

The current Nirmos SDK does not expose a gateway.models method. Use the OpenAI-compatible client or call GET /v1/models directly.

Nirmos metadata extension

Each model item can include an _meta object with catalog information:

type NirmosModelMetadata = {
  name: string;
  capabilities: unknown;
  context_window: number | null;
  max_output_tokens: number | null;
  pricing: unknown;
};

Treat catalog metadata as descriptive. The gateway still verifies request capabilities while resolving a direct model or route target.