Android Voice AI SDK in Model-View-ViewModel (ie MVVM)

flowchart LR
    Microphone --> AudioRecord --> VAD --> STT --> ClaudeAI["Claude AI"] --> TTS --> Speaker

The Android Voice AI SDK is a reusable Android library that gives any app a full voice-driven AI conversation pipeline in minutes. It captures audio from the device microphone, transcribes speech to text, sends the transcript to Anthropic Claude for an intelligent response, and speaks the reply back to the user through text-to-speech — all wired together with a single VoiceAISDK.Builder call. The SDK ships ready-to-drop-in Jetpack Compose UI components, swappable STT/TTS engine adapters, on-device emotion detection, and security utilities including PII redaction and encrypted key storage.

Features

Requirements

Quick Start

Step 1 — Add the dependency and manifest permissions

In your app build.gradle.kts:

dependencies {
    implementation("com.sdk:voice-ai-sdk:1.0.0")
}

In app/src/main/AndroidManifest.xml:

<uses-permission android:name="android.permission.RECORD_AUDIO" />
<uses-permission android:name="android.permission.INTERNET" />
<uses-permission android:name="android.permission.ACCESS_NETWORK_STATE" />

Step 2 — Hilt setup

Annotate your Application class with @HiltAndroidApp and your Activity with @AndroidEntryPoint:

@HiltAndroidApp
class MyApp : Application()

@AndroidEntryPoint
class MainActivity : ComponentActivity() { ... }

Step 3 — Add your API key to local.properties

local.properties is git-ignored, so your key never ends up in source control:

ANTHROPIC_API_KEY=sk-ant-...

Then expose it via BuildConfig in app/build.gradle.kts:

defaultConfig {
    buildConfigField(
        "String",
        "ANTHROPIC_API_KEY",
        "\"${project.findProperty("ANTHROPIC_API_KEY") ?: ""}\"",
    )
}

buildFeatures {
    buildConfig = true
}

Step 4 — Build the SDK

Provide the SDK through Hilt by creating an AppModule:

@Module
@InstallIn(SingletonComponent::class)
object AppModule {

    @Provides
    @Singleton
    fun provideVoiceAIConfig(): VoiceAIConfig =
        VoiceAIConfig(anthropicApiKey = BuildConfig.ANTHROPIC_API_KEY)

    @Provides
    @Singleton
    fun provideVoiceAISDK(
        @ApplicationContext context: Context,
        config: VoiceAIConfig,
    ): VoiceAISDK = VoiceAISDK.Builder(context)
        .anthropicApiKey(config.anthropicApiKey)
        .debugLogging(BuildConfig.DEBUG)
        .build()
}

Or construct the SDK directly without Hilt:

val sdk = VoiceAISDK.Builder(context)
    .anthropicApiKey(BuildConfig.ANTHROPIC_API_KEY)
    .debugLogging(true)
    .config { copy(systemPrompt = "You are a concise voice assistant.") }
    .build()

val session: VoiceAISession = sdk.createSession()
session.start()

Step 5 — Add the VoiceScreen composable

Use VoiceSessionPermissionGate to handle the RECORD_AUDIO runtime permission automatically, then place VoiceButton and ConversationView inside:

@Composable
fun VoiceScreen(viewModel: VoiceViewModel = hiltViewModel()) {
    VoiceSessionPermissionGate(
        rationale = "Microphone access is required for voice conversations.",
    ) {
        Column(
            modifier = Modifier
                .fillMaxSize()
                .padding(16.dp),
            verticalArrangement = Arrangement.SpaceBetween,
        ) {
            ConversationView(
                messages = viewModel.messages.collectAsStateWithLifecycle().value,
                modifier = Modifier.weight(1f),
            )
            VoiceButton(
                session = viewModel.session,
                modifier = Modifier.align(Alignment.CenterHorizontally),
            )
        }
    }
}

Architecture

The SDK is organised into six layers, each with a single responsibility:

Available Engines

Configuration Reference

All options are fields on VoiceAIConfig. Pass a config { } block to VoiceAISDK.Builder to override defaults.