We should be supporting and using local models that allow you to run whatever model you want.