@runanywhere/mlx

Apple MLX backend registration for the RunAnywhere React Native SDK.

The package owns only backend registration and its iOS runtime binaries. Model catalog, download, lifecycle, LLM, VLM, speech, and embedding APIs remain in @runanywhere/core, which is the shared public surface across backends.

Requirements

• @runanywhere/core 0.20.11+
• React Native 0.83.1+
• Xcode 26+ with the Swift 6.2 toolchain
• A physical Apple device running iOS 17.5+

MLX is an Apple-only backend. Android autolinking is intentionally disabled. The arm64 iOS Simulator artifact supports package, compile, link, and startup validation only; MLX.register() and MLX.isAvailable() return false there.

Installation

npm install @runanywhere/core @runanywhere/mlx @runanywhere/proto-ts
cd ios && pod install && cd ..

The published package must contain RABackendMLX.xcframework, RunAnywhereMLXRuntime.xcframework, and RunAnywhereMLXMetal.xcframework. Packaging and CocoaPods fail when any runtime artifact is missing; the TypeScript facade is not published as a standalone substitute for native MLX support.

Usage

import { RunAnywhere } from '@runanywhere/core';
import { MLX } from '@runanywhere/mlx';
import { InferenceFramework } from '@runanywhere/proto-ts/model_types';

const registered = await MLX.register();
if (!registered) {
  throw new Error('MLX is unavailable on this target');
}

await RunAnywhere.initialize();
await RunAnywhere.registerModel({
  id: 'mlx-qwen3-0.6b-4bit',
  name: 'MLX Qwen3 0.6B 4bit',
  url: 'https://huggingface.co/mlx-community/Qwen3-0.6B-4bit',
  framework: InferenceFramework.INFERENCE_FRAMEWORK_MLX,
  memoryRequirement: 650_000_000,
  supportsThinking: true,
});

The native MLX engine uses its fixed router priority. The facade also exposes MLX.unregister(), MLX.isRegistered(), and MLX.isAvailable().