Apple spent a Monday in June teaching apps to load a model the way they load a font. WWDC, 8 June. Core AI is the inference stack that already powers on-device Apple Intelligence, now opened to developers. You point a Swift API at a URL. You get a .aimodel file. No server. No token bill. Ahead-of-time compilation with coreai-build so the first run is not a wait. The framework uses Apple silicon — CPU, GPU, Neural Engine — and ships Python conversion tools beside the runtime.

Henriok / Wikimedia Commons (CC0)