I didn't mean to imply that all inference is local. A dynamic combination of server and local inference is called "hybrid".