Most developers associate AI with cloud APIs—or consider running it locally as complex, almost “black-magic” work. But modern operating systems already expose native, production-ready APIs for running AI models directly on the device. These OS-level AI capabilities—available on Apple, Android, Windows, and Chrome AI (experimental)—remain largely unknown, sparsely documented outside vendor ecosystems, and rarely used in everyday application development.

In the first part of this talk, we present a state of the union of on-device AI at the operating system level, exploring what AI APIs today’s OSes provide and which types of tasks they are designed to support.

In the second part, we move from theory to practice with a hands-on walkthrough of a simple Flutter app built using flutter_local_ai, showing how these native OS capabilities can be accessed through a unified, cross-platform API to build privacy-first, low-latency, on-device AI features.