runanywhere-sdks
RunAnywhere React Native SDK
On-device AI for React Native — run LLMs, speech-to-text, text-to-speech, and voice pipelines locally. Private, offline-capable, production-ready.
Requirements
| Component | Minimum | Recommended |
|---|---|---|
| React Native | 0.83.1+ | 0.85.3 |
| Node.js | 22.12+ | 24 LTS |
| iOS | 17.5+ | 17.5+ |
| Android | API 24 (7.0+) | API 28+ |
| Xcode | 26+ | 26+ |
| RAM | 3 GB | 6 GB+ for larger models |
Apple Silicon devices and Android phones with 6 GB+ RAM are recommended for 3B+ models.
Installation
Install the core package plus the backends you need. Pin to 0.20.11:
npm install @runanywhere/core@0.20.11 @runanywhere/llamacpp@0.20.11
| Package | Purpose |
|---|---|
@runanywhere/core |
Core SDK — required |
@runanywhere/llamacpp |
LLM / VLM (GGUF via llama.cpp) |
@runanywhere/onnx |
STT, TTS, VAD (Whisper, Piper, Silero) |
@runanywhere/mlx |
Apple MLX on physical iOS devices |
@runanywhere/qhexrt |
Optional Qualcomm Hexagon NPU acceleration (Snapdragon Android) |
iOS: run cd ios && pod install && cd .. after installing packages.
Android: no extra setup — native libraries are bundled and downloaded during the Gradle build.
Quick Start
import { RunAnywhere, SDKEnvironment } from '@runanywhere/core';
import { LlamaCPP } from '@runanywhere/llamacpp';
await RunAnywhere.initialize({
environment: SDKEnvironment.SDK_ENVIRONMENT_DEVELOPMENT,
});
await LlamaCPP.register();
await RunAnywhere.downloadModel('smollm2-360m');
await RunAnywhere.loadModel('smollm2-360m');
const result = await RunAnywhere.generate('What is the capital of France?');
console.log(result.text);
For streaming tokens, voice (STT/TTS), vision, RAG, and tool calling, see the documentation.
Production initialization
await RunAnywhere.initialize({
apiKey: '<YOUR_API_KEY>',
baseURL: 'https://api.runanywhere.ai',
environment: SDKEnvironment.SDK_ENVIRONMENT_PRODUCTION,
});
| Environment | Description |
|---|---|
SDK_ENVIRONMENT_DEVELOPMENT |
Keyless OSS mode, verbose logging |
SDK_ENVIRONMENT_STAGING |
Testing with real services |
SDK_ENVIRONMENT_PRODUCTION |
Authenticated control plane, telemetry |
Capabilities
The SDK exposes a unified RunAnywhere namespace. Register backend modules (LlamaCPP, ONNX, MLX) before downloading or loading models.
| Capability | Key APIs | Backend package |
|---|---|---|
| LLM | generate, generateStream |
@runanywhere/llamacpp, @runanywhere/mlx |
| VLM | processImage, processImageStream |
@runanywhere/llamacpp, @runanywhere/mlx |
| STT | transcribe, transcribeStream |
@runanywhere/onnx |
| TTS | synthesize, synthesizeStream |
@runanywhere/onnx |
| VAD | Voice activity detection helpers | @runanywhere/onnx |
| Voice agent | initializeVoiceAgent, streamVoiceAgent, processVoiceTurn |
Core + ONNX + LLM backend |
| Models | registerModel, downloadModel, loadModel, listModels |
@runanywhere/core |
| RAG / tools | ragQuery, generateWithTools |
@runanywhere/core |
Supported model formats
| Format | Use case | Backend |
|---|---|---|
| GGUF | LLM, some VLM | LlamaCPP |
| ONNX / Sherpa archives | STT, TTS, VAD | ONNX |
| MLX bundles | LLM, VLM, speech on Apple silicon | MLX |
| QHexRT bundles | NPU-accelerated models on Snapdragon | QHexRT |
Built-in catalog models (e.g. smollm2-360m) are discovered automatically after initialization — no manual registerModel call required for catalog entries.
Error Handling
Errors throw SDKException with a proto ErrorCode:
import { SDKException, ErrorCode, isSDKException } from '@runanywhere/core';
try {
await RunAnywhere.generate('Hello!');
} catch (error) {
if (isSDKException(error)) {
console.error(error.code, error.message);
}
}
Documentation & Examples
| Resource | Link |
|---|---|
| Docs site | docs.runanywhere.ai/react-native/introduction |
| API reference | Docs/Documentation.md |
| Example app | bindings/react-native/example/ |
| Contributing / building from source | Docs/DEVELOPMENT.md |
FAQ
Do I need an internet connection?
Only for initial model download. Inference runs entirely on-device afterward.
Is user data sent to the cloud?
No. All inference is local. Production mode may collect anonymous telemetry (configurable).
Can I use custom models?
Yes — register any compatible GGUF, ONNX/Sherpa, or MLX bundle via RunAnywhere.registerModel().
Support
- Documentation: docs.runanywhere.ai
- Discord: discord.gg/N359FBbDVd
- GitHub Issues: github.com/RunanywhereAI/runanywhere-sdks/issues
- Email: founders@runanywhere.ai
License
RunAnywhere License. See LICENSE for details.