Skip to main content

@stewmore/expo-ai-apple-foundation-models

Apple Foundation Models (on-device, iOS 26+) provider for the Expo AI Runtime.

Install & use​

import { ExpoAI } from '@stewmore/expo-ai-core';
import '@stewmore/expo-ai-apple-foundation-models'; // registers the iOS adapter (side-effect import)

const caps = await ExpoAI.getCapabilities();
if (caps.available && caps.provider === 'apple-foundation-models') {
  const result = await ExpoAI.generate({ prompt: 'Write a haiku about Swift.' });
  console.log(result.text); // privacy: on-device
}

Requirements​

  • iOS / iPadOS 26+ on an Apple Intelligence–capable device with Apple Intelligence enabled.
  • A development build (custom native code — not Expo Go): npx expo prebuild && npx expo run:ios.

On unsupported OS/devices the adapter reports available: false with a reason (unsupported_os_version, unsupported_device, apple_intelligence_disabled, model_initializing), so the runtime can fall back per the app's policy.

What it maps​

RuntimeFoundation Models
generateLanguageModelSession.respond(to:options:)
streamLanguageModelSession.streamResponse(...) (cumulative snapshots → token deltas)
createSessionnative stateful LanguageModelSession (generateInSession / resetSession / disposeSession)
capability detectionSystemLanguageModel.availability
errorsLanguageModelSession.GenerationError → ExpoAIError

Structured output (generateObject), summarize/rewrite/proofread are provided by the core runtime over generate. Native guided generation via DynamicGenerationSchema is a planned enhancement.

note

The FoundationModels API surface follows the WWDC25 docs; verify exact symbol names against your installed Xcode SDK if the native build flags a mismatch.