@zerogpu/expo-android-sdk

Expo / React Native module that turns an Android device into an on-device inference node on the ZeroGPU network. It wraps the native ai.zerogpu:android-sdk AAR and exposes a tiny JS API.

Android-only. iOS builds link a no-op stub so cross-platform apps still compile, but the device only acts as a node on Android. The native library ships arm64-v8a (phones) and x86_64 (emulators) only.

Install

npx expo install @zerogpu/expo-android-sdk

Add the config plugin to app.json / app.config.js:

{
  "expo": {
    "plugins": ["@zerogpu/expo-android-sdk"]
  }
}

Then regenerate the native project:

npx expo prebuild

The plugin wires up the two app-level Android needs automatically:

• a pickFirst '**/libc++_shared.so' packaging rule (the AAR bundles native libs from ONNX Runtime and llama.cpp, both of which ship libc++_shared.so), and
• raises minSdkVersion to 26 if your app is lower (never lowers it).

Permissions (INTERNET, ACCESS_NETWORK_STATE) and largeHeap come from the AAR manifest and merge in automatically — nothing to add.

Requires an Expo app using expo-modules-core (managed or bare). This is not a plain-React-Native library.

Usage

import { start, stop } from '@zerogpu/expo-android-sdk';

// Start the node (typically once, e.g. in your root component / on app launch).
await start('zgpu-sdk-xxxxxxxx'); // your operator key from https://zerogpu.ai

// ...later, on teardown:
await stop();

Keep the operator key out of source control (e.g. via expo-constants / environment config), just like any secret.

start() is idempotent, and the native SDK manages the foreground/background socket itself — you do not need to wire it to AppState. On iOS both calls resolve as no-ops.

Options

await start('zgpu-sdk-…', {
  baseUrl: 'https://devices.zerogpu.ai', // default
  env: 'production',                     // default
  enableTelemetry: true,                 // default
});

Verifying it works

Run the app on an arm64-v8a phone or an x86_64 emulator (Android 8.0+), then watch logcat:

adb logcat -s ZeroGpuSdk:* ZeroGpuWebSocket:*

You want to see status":"idle" and WebSocket ready for tasks. Send an inference request to https://api.zerogpu.ai/v1/chat/completions with your API key and match the response id to the requestId in logcat — see the SDK's INTEGRATION.md for the full end-to-end check.

Versioning

This package is published from the same release as the AAR and always carries the same version as ai.zerogpu:android-sdk. The AAR coordinate is derived from this package's own package.json, so an installed npm version always pulls the matching AAR.

License

Apache-2.0