Native Libraries

React Native ExecuTorch ships the core runtime, hardware-accelerated backends (XNNPACK, Core ML, MLX, Vulkan), and native third-party libraries (OpenCV, phonemis) as separate downloadable artifacts.

By default, everything is downloaded and enabled, so no configuration is required to get started. However, because on-device AI backends and vision libraries add substantial binary weight, you can tailor exactly what gets pulled into your app. Declaring only the features or backends you use reduces install times, speeds up builds, and significantly shrinks the final app bundle. At install time, a postinstall script inspects your configuration and fetches only the native artifacts your app needs.

How it works

1. Reads config from the react-native-executorch block in your package.json (if any).
2. Writes rne-build-config.json with boolean flags that the native build reads.
3. Downloads binaries from the GitHub Release tagged with your package version, verifying checksums.
4. Caches them under ~/.cache/react-native-executorch/<version>/.

Configuration

Add a react-native-executorch block to your package.json:

{
  "react-native-executorch": {
    "features": ["classification", "styleTransfer"]
  }
}

Options

Option	Purpose	Accepted values
features	High-level tasks — each automatically expands to the backends and libraries it needs	"classification", "imageEmbeddings", "instanceSegmentation", "keypointDetection", "llm", "multimodalLLM", "objectDetection", "ocr", "privacyFilter", "segmentAnything", "semanticSegmentation", "speechToText", "styleTransfer", "textEmbeddings", "textToImage", "textToSpeech", "tokenizer", "vad", "verticalOCR"
backends	Hardware acceleration backends directly	"xnnpack", "coreml", "mlx", "vulkan"
libs	Extra native C++ libraries	"opencv", "phonemis"

The three lists are merged, so you can pair high-level features with specific backends or libs. Re-run your package manager install after editing.

Monorepos

The block is read from the directory the install was run in, then from every package.json above the installed package. A hoisted workspace resolves to the root either way, so put the block there; an app that keeps its own node_modules (pnpm, nohoist) is found from its own package.json. node_modules/react-native-executorch/rne-build-config.json records what was actually resolved, and the install log names the manifest it read.

pnpm

pnpm 10 and later do not run dependency build scripts unless you allow them, so the download never happens and the native build fails later on a missing file. Run pnpm approve-builds react-native-executorch once. See Troubleshooting.

Backends

Hardware backends provide optimized execution kernels for specific processors and platforms. See the ExecuTorch Backends documentation for details on lowering and delegate compilation:

• XNNPACK — High-efficiency floating-point and quantized neural network inference operators optimized for ARM and x86 CPUs. Supported on both Android and iOS.
• Core ML — Apple's framework for hardware-accelerated machine learning on Apple Silicon, targeting the Apple Neural Engine (ANE) and GPU. Supported on iOS only.
• MLX — An array framework designed for efficient machine learning on Apple silicon via Metal compute shaders, used primarily for accelerated LLM generation. Supported on iOS only (physical device only, no simulator).
• Vulkan — Cross-platform 3D graphics and compute API, leveraging mobile GPUs on Android only for accelerated neural network inference and tensor compute operations.

Third-Party Libraries

• OpenCV — High-performance computer vision library providing image transformations, color space conversions, resizing, and pixel format operations (used by vision pipelines and multimodal LLMs). Provided on iOS via CocoaPods and on Android as static prebuilt libraries.
• phonemis — High-performance C++ library for Grapheme-to-Phoneme (G2P) conversion, delivering universal IPA phonemization as the frontend for Text-to-Speech pipelines. Compiled from source on both Android and iOS when enabled.

Features

Specifying a task under features is shorthand: it automatically expands to the union of backends and native libraries required by the pre-exported models in that domain.

Feature	Expanded Backends	Expanded Extra Libs
llm	xnnpack, mlx, vulkan	—
multimodalLLM	xnnpack, mlx, vulkan	opencv
privacyFilter	xnnpack, mlx	—
speechToText	xnnpack, coreml, mlx, vulkan	—
textToSpeech	xnnpack, coreml, mlx, vulkan	phonemis
vad	xnnpack	—
textEmbeddings	xnnpack, coreml, mlx, vulkan	—
imageEmbeddings	xnnpack, coreml, mlx, vulkan	opencv
classification	xnnpack, coreml	opencv
objectDetection	xnnpack, coreml, vulkan	opencv
keypointDetection	xnnpack, coreml, mlx, vulkan	opencv
semanticSegmentation	xnnpack, coreml	opencv
instanceSegmentation	xnnpack, coreml, vulkan	opencv
ocr	xnnpack, coreml, vulkan	opencv
verticalOCR	xnnpack	opencv
styleTransfer	xnnpack, coreml, vulkan	opencv
textToImage	xnnpack, coreml, vulkan	opencv
segmentAnything	xnnpack, coreml, vulkan	opencv
tokenizer	—	—

Binary size

Approximate size each backend adds to a release arm64 build:

Backend	Android	iOS
xnnpack	+1.7 MB	+1.4 MB
coreml	—	+0.4 MB
mlx	—	+6.0 MB
vulkan	+10.3 MB	—