Choose AI models from a live catalog, not a hard-coded list
A model identifier that worked last week may not be enabled for a new workspace today. Treat model discovery as part of the request flow: fetch the catalog, show what it actually returns, and keep a useful fallback when the list changes.
Fetch the catalog with the same key you will use to build
Ngamia exposes enabled models at GET /v1/models. The endpoint returns the public model identifiers available to that API key, plus current pricing and modality details. This is more reliable than copying a provider's global model list into your app.
Load the result when a user opens the model picker, then refresh it periodically or when a request reports that a model is unavailable.
GET https://api.ngamia.cc/v1/models
Authorization: Bearer $NGAMIA_API_KEYFilter by the task, then display the trade-offs
A chat interface needs a model that accepts text and returns text. A document upload needs file or PDF input. A voice product may need audio input, audio output, or both. Use the catalog's modality fields to filter, and show price alongside the model name so the choice has context.
Don't imply that every provider or capability is available to every workspace. The live catalog is the source of truth.
Keep the request path simple
Once a person picks a model, pass its public ID to the endpoint. The gateway resolves it internally, so your client doesn't have to construct a provider-specific name.
If you cache results, give the cache a short lifetime and handle a missing model cleanly. A stale picker should help the person choose again, not strand the entire workflow.