@dan Currently the reality is I need to still use some frontier models to get speed and quality on core stuff. But always moving more and more functionality towards local models! The goal is 100% local eventually of course.