Reasoning capability not being reported nor being selectable via API

[claudio-silva]

Currently, when we open /settings in Factory CLI (droid), it shows “Reasoning level: Reasoning disabled”, which means the CLI treats the models as if they don’t support reasoning, even though they actually do.

The issue with this is that Factory CLI won’t display reasoning tokens if it can’t determine whether the model supports reasoning.

A more flexible approach would be for VibeProxy to expose the reasoning effort level and allow the client to specify it, instead of relying on separate pseudo-model names.

This way, we could use Factory’s existing reasoning effort settings without having to switch models, and the model’s thoughts would also be displayed.

Would it be possible for you to implement this? It would make this excellent tool even more useful.

[Tutorec]

its a cosmetic problem , it does use extra 32K for ultrathinking , you can check the exposes terminal log : [ThinkingProxy] Transformed model 'claude-sonnet-4-5-20250929-thinking-32000' → 'claude-sonnet-4-5-20250929' with thinking budget 31999

[claudio-silva]

Yes, I know it works, but it's not merely cosmetic.
As I explained above, there would be significant advantages in providing the described features via API.
At the very least, being able to see the model's reasoning would be very welcome.

Note that I'm talking about droid, the factory.ai CLI. But other clients could also benefit, depending on their capabilities.