Why ModelRegistry starts with model identity
A model directory becomes useful when its identities are stable, explicit, and independent from provider naming.
An AI model directory can look like a list of provider names and model IDs. That is convenient at first, but it makes the catalog difficult to search and compare as soon as the same model appears under multiple providers or receives a new release.
One identity, many releases
ModelRegistry separates the identity of a model from its delivery options:
- Organization identifies the model owner.
- Model Family groups related releases under one product line.
- Model Version represents a specific release with its own release date and capabilities.
- Provider Offering records how a provider exposes that version, including the provider's external model ID and pricing information.
This gives every model version a platform-owned identity and URL. Provider naming can change without changing the model's place in the catalog.
Structured facts beat labels in prose
Aliases, input and output modalities, capability tags, relations, and resources are stored as structured fields. That makes them useful in search, filters, detail pages, and the comparison workspace instead of being trapped in an unsearchable description.
The model identity layer is the foundation for the public model catalog. Recommendation and query APIs can build on it without redefining what a model is.