We have been thinking a bit about this, and one option would be to have some form of locally hosted runner. You can optimize the task in the cloud and deploy it

We have been thinking a bit about this, and one option would be to have some form of locally hosted runner. You can optimize the task in the cloud and deploy it locally. Something like that. It is possible to plug in custom models so technically feasible.