As on-device AI inference becomes central to modern mobile development, CoreML (iOS) and ML Kit (Android) are the primary native frameworks enabling LLM and foundation model execution directly on device without cloud dependency. Currently, Capacitor lacks deep, first-class plugin support for these APIs, forcing developers to write complex native bridges manually.

For: AFM 3 Core — 3B dense, on-device, lightweight/fast

We’re requesting expanded Capacitor plugin support for CoreML on iOS, including model loading, inference pipelines, and Vision/NLP task integration; ML Kit on Android, including text recognition, language ID, smart reply, and custom model inference; and a shared abstraction layer for cross-platform on-device LLM orchestration.

Expo’s bundle/build workflow for React Native already gives developers a clearer path to package native AI capabilities into production apps, while Capacitor still leaves much of this to custom plugin and bridge work. Stronger first-party support would help Capacitor better compete with Expo/React Native for teams that want to ship on-device LLM features with less native overhead.

This request should also accommodate Apple’s newer iOS foundation model direction, including newer patches and API updates related to local LLM support, so developers are not forced to constantly rework integrations as Apple expands its on-device AI stack. Ideally, Capacitor should provide a forward-compatible plugin surface that can evolve with new iOS LLM capabilities while keeping the developer experience stable.

-JTX