Extremely minimal Compose Multiplatform sample tha...

# feed

[John O'Reilly]

Extremely minimal Compose Multiplatform sample that demonstrates use of on-device AI (Foundation Model Framework on iOS and MLKit Prompt API on Android)
https://github.com/joreilly/OnDeviceAI

[Stefan Oltmann]

Nice. How about a Desktop target? 🙂

[Stefan Oltmann]

That's always where it gets difficult on Windows. 😄

[John O'Reilly]

I was mostly aiming at demonstrating the on-device AI capability that Android and iOS have (at least in later devices/OS versions)

[Stefan Oltmann]

Yeah, it's really a bummer Windows doesn't come with something on-board.
With TensorFlow, PyTorch, etc. there are so many choices.

[David Conceição]

This is really cool, thanks for sharing!

[David Conceição]

Also, Windows has an Windows ML to run local models. I haven't tried it yet, still there are multiple APIs popping @Stefan Oltmann.

[Stefan Oltmann]

Ah, interesting....

https://blogs.windows.com/windowsdeveloper/2025/05/19/introducing-windows-ml-the-future-of-machine-learning-development-on-windows/

Would be nice to see also Kotlin-binding for that.

[אליהו הדס]

they have a java lib https://onnxruntime.ai/

[אליהו הדס]

and it's work with android and ios

[אליהו הדס]

work with windows mac, linux and web

[אליהו הדס]

I asked chatgpt for an alternative for Linux and he found this 🙂

[David Conceição]

ONNX is a complete runtime. It's different than the ones embedded in the operating system because the runtime itself must be packaged with the app. Also, it requires downloading or embedding models to be loaded in runtime. It's a valid approach, just different than the examples in the repository that started this thread.