I suspect it means using ML to identify things in the camera. Like say cat, dog, man with clip board, man carrying box, etc.

I've love to get a hangout/signal/IM text identifying anything approaching my door without having to look at a picture. Bonus if the face recognition is good enough to recognize family.

Quite likely many people don't even want the webserver.

webservers can be local, homes.

Still just another useless attack surface.

Put it on its own subnet.