add local models optionally to save tokens

I think in the future you could add optionally some local models to make some part of the work locally, in example audio to text, and save; or even some simple reasoning tasks. :)