AI model availability changes quickly, and our current model catalog is maintained locally through metadata.json.
This works well as a safe fallback, but it also means that newly available models, renamed models, deprecated models, or recommended replacements can only be delivered with the next application update. For AI providers with fast-moving model catalogs, this is too rigid and creates unnecessary friction for merchants who want to use newly released models earlier.
We should add support for a remote AI model catalog.
The remote catalog should provide the same kind of model metadata that is currently stored locally, including model identifiers, display names, capabilities, supported tools, performance level, vision support, preferred/default flags, deprecation status, and optional replacement aliases.
The REST API should expose the catalog in a shape that can be serialized and deserialized as AIMetadata / AIModelEntry instances by the application.
The application should continue to ship with a local catalog as a reliable baseline. At runtime, the remote catalog should be fetched and merged with the local catalog. This allows us to update the available model list centrally without requiring a full application release.
Expected behavior:
- The local
metadata.json remains the fallback and default source.
- A remote catalog can extend, update, deprecate, or replace local model entries.
- Deprecated models can point to a suggested replacement.
- Existing custom/user-defined models remain supported.
- The system remains usable if the remote catalog is unavailable.
- Remote metadata should be cacheable to avoid unnecessary requests and to keep the admin experience fast.
The remote catalog should be maintained through an internal backend UI, so model metadata can be updated by us without code changes or application deployment.
Suggested edit controls for model entries:
| Field |
Edit control |
| AI Provider |
Select |
| Id |
Text input |
| Name |
Text input |
| Type |
Select |
| Description |
Text input |
| Preferred |
Checkbox |
| Vision |
Checkbox |
| Deprecated |
Checkbox |
| Alias |
Text input / model picker |
| Level |
Select |
| Tools |
Checkbox list |
| Stream |
Checkbox |
| Image output capabilities |
Tag input |
| Aspect ratios |
Tag input |
| Resolutions |
Tag input |
| Formats |
Tag input |
This gives merchants faster access to current AI models while keeping the application stable, predictable, and backward-compatible.
AI model availability changes quickly, and our current model catalog is maintained locally through
metadata.json.This works well as a safe fallback, but it also means that newly available models, renamed models, deprecated models, or recommended replacements can only be delivered with the next application update. For AI providers with fast-moving model catalogs, this is too rigid and creates unnecessary friction for merchants who want to use newly released models earlier.
We should add support for a remote AI model catalog.
The remote catalog should provide the same kind of model metadata that is currently stored locally, including model identifiers, display names, capabilities, supported tools, performance level, vision support, preferred/default flags, deprecation status, and optional replacement aliases.
The REST API should expose the catalog in a shape that can be serialized and deserialized as
AIMetadata/AIModelEntryinstances by the application.The application should continue to ship with a local catalog as a reliable baseline. At runtime, the remote catalog should be fetched and merged with the local catalog. This allows us to update the available model list centrally without requiring a full application release.
Expected behavior:
metadata.jsonremains the fallback and default source.The remote catalog should be maintained through an internal backend UI, so model metadata can be updated by us without code changes or application deployment.
Suggested edit controls for model entries:
This gives merchants faster access to current AI models while keeping the application stable, predictable, and backward-compatible.