RunAnywhere Kotlin SDK Starter

January 15, 2026 · View on GitHub

A comprehensive Android starter app demonstrating the RunAnywhere SDK capabilities - privacy-first, on-device AI for Android with Kotlin and Jetpack Compose.

Features

This starter app showcases all major capabilities of the RunAnywhere SDK:

🧠 Chat (LLM Text Generation)

On-device text generation using SmolLM2 360M
Real-time chat interface with message history
Powered by llama.cpp backend

🎤 Speech to Text (STT)

Real-time speech recognition using Whisper Tiny
Microphone permission handling
Voice activity detection
Powered by Sherpa-ONNX backend

🔊 Text to Speech (TTS)

Natural voice synthesis using Piper TTS
Sample texts and custom input
High-quality US English voice (Lessac)
Powered by Sherpa-ONNX backend

🎯 Voice Pipeline (Voice Agent)

Complete voice conversation pipeline
Combines STT → LLM → TTS
Real-time conversation flow
Status indicators for each stage

Getting Started

Prerequisites

Android Studio: Hedgehog (2023.1.1) or later
Minimum SDK: API 26 (Android 8.0)
Target SDK: API 35 (Android 15)
Kotlin: 2.0.21 or later
Java: 17

Installation

Clone the repository

git clone <repository-url>
cd starter_apps/kotlinstarterexample

Open in Android Studio
- Open Android Studio
- Select "Open an Existing Project"
- Navigate to the kotlinstarterexample folder
- Click "OK"
Sync Gradle
- Android Studio will automatically sync Gradle
- If not, click "Sync Now" in the notification bar
Run the app
- Connect an Android device or start an emulator
- Click the "Run" button (▶️) in Android Studio
- Select your device/emulator
- The app will build and install

First Launch

On the first launch:

Home Screen: You'll see 4 feature cards
Load Models: Each feature requires downloading AI models:
- LLM: ~400 MB (SmolLM2 360M)
- STT: ~75 MB (Whisper Tiny)
- TTS: ~20 MB (Piper TTS)
Grant Permissions: STT and Voice Pipeline require microphone permission
Start Using: Once models are loaded, all features are ready!

Architecture

Project Structure

app/src/main/java/com/runanywhere/kotlin_starter_example/
├── MainActivity.kt                    # App entry point
├── services/
│   └── ModelService.kt               # Model management (download, load, unload)
└── ui/
    ├── theme/                        # App theme and colors
    │   ├── Theme.kt
    │   └── Type.kt
    ├── components/                   # Reusable UI components
    │   ├── FeatureCard.kt
    │   └── ModelLoaderWidget.kt
    └── screens/                      # Feature screens
        ├── HomeScreen.kt
        ├── ChatScreen.kt
        ├── SpeechToTextScreen.kt
        ├── TextToSpeechScreen.kt
        └── VoicePipelineScreen.kt

Key Technologies

Jetpack Compose: Modern declarative UI
Material 3: Latest Material Design
Navigation Compose: Screen navigation
Coroutines & Flow: Asynchronous operations
ViewModel: State management
RunAnywhere SDK v0.16.0-test.39: On-device AI

RunAnywhere SDK Integration

Dependencies

The app uses three RunAnywhere packages:

// build.gradle.kts (app module)
dependencies {
    // Core SDK
    implementation("ai.runanywhere:runanywhere-kotlin:0.16.0-test.39")
    
    // Backends
    implementation("ai.runanywhere:runanywhere-llamacpp:0.16.0-test.39")  // LLM
    implementation("ai.runanywhere:runanywhere-onnx:0.16.0-test.39")      // STT/TTS
}

Initialization

// MainActivity.kt
RunAnywhere.initialize(environment = SDKEnvironment.DEVELOPMENT)
ModelService.registerDefaultModels()

Model Registration

Models are registered in ModelService.kt:

// LLM Model
RunAnywhere.registerModel(
    id = "smollm2-360m-instruct-q8_0",
    name = "SmolLM2 360M Instruct Q8_0",
    url = "https://huggingface.co/HuggingFaceTB/SmolLM2-360M-Instruct-GGUF/resolve/main/smollm2-360m-instruct-q8_0.gguf",
    framework = InferenceFramework.LLAMA_CPP,
    memoryRequirement = 400_000_000
)

// STT Model
RunAnywhere.registerModel(
    id = "sherpa-onnx-whisper-tiny.en",
    name = "Sherpa Whisper Tiny (ONNX)",
    url = "https://github.com/RunanywhereAI/sherpa-onnx/releases/download/runanywhere-models-v1/sherpa-onnx-whisper-tiny.en.tar.gz",
    framework = InferenceFramework.ONNX,
    category = ModelCategory.SPEECH_RECOGNITION
)

// TTS Model
RunAnywhere.registerModel(
    id = "vits-piper-en_US-lessac-medium",
    name = "Piper TTS (US English - Medium)",
    url = "https://github.com/RunanywhereAI/sherpa-onnx/releases/download/runanywhere-models-v1/vits-piper-en_US-lessac-medium.tar.gz",
    framework = InferenceFramework.ONNX,
    category = ModelCategory.SPEECH_SYNTHESIS
)

Usage Examples

Chat (LLM)

val response = RunAnywhere.chat("Explain AI in simple terms")

Speech to Text (STT)

val audioData: ByteArray = recordAudio()
val transcription = RunAnywhere.transcribe(audioData)

Text to Speech (TTS)

RunAnywhere.speak("Hello, world!")

Voice Pipeline

RunAnywhere.startVoiceSession().collect { event ->
    when (event) {
        is VoiceSessionEvent.Listening -> updateUI("Listening...")
        is VoiceSessionEvent.Transcribed -> updateUI("You: ${event.text}")
        is VoiceSessionEvent.Thinking -> updateUI("Thinking...")
        is VoiceSessionEvent.Responded -> updateUI("AI: ${event.text}")
        is VoiceSessionEvent.Speaking -> updateUI("Speaking...")
    }
}

Performance

Model Sizes

LLM (SmolLM2 360M): ~400 MB
STT (Whisper Tiny): ~75 MB
TTS (Piper): ~20 MB
Total: ~495 MB

Inference Speed

LLM: 5-15 tokens/sec (device dependent)
STT: Real-time transcription
TTS: Real-time synthesis

Device Requirements

RAM: Minimum 2GB recommended
Storage: 1GB free space for models
CPU: ARMv8 64-bit recommended (supports ARMv7)

Customization

Changing Models

To use different models, update ModelService.kt:

companion object {
    const val LLM_MODEL_ID = "your-model-id"
    const val STT_MODEL_ID = "your-stt-model-id"
    const val TTS_MODEL_ID = "your-tts-model-id"
    
    fun registerDefaultModels() {
        RunAnywhere.registerModel(
            id = LLM_MODEL_ID,
            name = "Your Model Name",
            url = "your-model-url",
            framework = InferenceFramework.LLAMA_CPP
        )
        // ... register other models
    }
}