Running a model directly on your own computer or server instead of sending requests to a remote service.