So perhaps I could be fooling myself, or maybe this is a bad idea…

But why not allow us to use local models as the helper? If we just added a pipeline to a local model as the brain of the helper, and kept all the resources that the old helper uses as a reference… That might alleviate a lot of stress and some cost off the function of the AI itself, which could save some money would it not?

I’m not super savvy on all the technical background required but as a concept I think this might at least somewhat help save on the COST side of things.

Thoughts?