@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
}

Config plugin

The native module autolinks on its own, but add the config plugin so expo prebuild also sets the iOS deployment target the pod needs — no manual expo-build-properties editing:

{
  "expo": {
    "plugins": ["@stewmore/expo-ai-apple-foundation-models"]
  }
}

Override the floor (defaults to 15.1, the pod minimum) if needed:

{
  "expo": {
    "plugins": [["@stewmore/expo-ai-apple-foundation-models", { "iosDeploymentTarget": "16.4" }]]
  }
}

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

| Runtime | Foundation Models | | -------------------- | ------------------------------------------------------------------------------------------------ | | generate | LanguageModelSession.respond(to:options:) | | stream | LanguageModelSession.streamResponse(...) (cumulative snapshots → token deltas) | | createSession | native stateful LanguageModelSession (generateInSession / resetSession / disposeSession) | | capability detection | SystemLanguageModel.availability | | errors | LanguageModelSession.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 (see ios/ source notes).

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