A Kotlin Multiplatform (KMP) library providing RAG (Retrieval-Augmented Generation) integration with a ready-to-use chatbot UI. The library integrates with Flowise as the RAG backend and targets Android, iOS, and Desktop (JVM).

Published to Maven Central as io.github.sjaindl:mobile-rag-assistant.

• Ready-to-use Compose Multiplatform chat UI
• SSE streaming responses from Flowise
• Local chat history persistence via SQLDelight
• Configurable app bar title and icon
• Optional welcome message and sample questions
• Source documents and tool call display
• Message reset option
• Customizable user and assistant icons
• Message character limit

Add the dependency to your build.gradle.kts:

implementation("io.github.sjaindl:mobile-rag-assistant:1.0.3")

• Android minSdk 24+
• iOS (iosX64, iosArm64, iosSimulatorArm64)
• JVM Desktop

Initialize Koin at app startup and provide an AssistantConfig:

startKoin {
    modules(assistantModule)
    single {
        AssistantConfig(
            provider = Provider.Flowise(
                baseUrl = "https://your-flowise-domain/flowise/api/v1/prediction/<chatflow-id>",
                apiKey = "your_api_key", // optional
            ),
            appBarTitle = Res.string.appName,
        )
    }
}

The library provides an assistantGraph extension on NavGraphBuilder and a navigateToAssistant() extension on NavController for integration into your existing Compose Navigation setup:

NavHost(navController = navController, startDestination = "home") {
    composable("home") { HomeScreen() }

    assistantGraph(rootNavController = navController)
}

Navigate to the chat screen from anywhere:

navController.navigateToAssistant()

ChatIcon can be either a Compose ImageVector or a DrawableResource:

ChatIcon.Vector(imageVector = Icons.Default.MyIcon)
ChatIcon.Drawable(drawable = Res.drawable.my_icon)

The :assistant module follows clean architecture:

UI Layer        → ChatScreen / ChatViewModel (Compose + MVVM, StateFlow)
Domain Layer    → Use cases (GetAssistantCompletionUseCase, etc.)
Data Layer      → AssistantRepositoryImpl
                  ├─ Remote: KtorAssistantService (Flowise SSE streaming)
                  └─ Local:  ChatMessageDataSource (SQLDelight)
DI              → Koin modules (assistantModule, dataModule, domainModule, platformModule)

Platform-specific code (Ktor client engine, SQLDelight driver, Koin platform module) lives in androidMain, iosMain, and jvmMain via Kotlin expect/actual.

androidApp      → composeApp → assistant (core library)
iosApp (Xcode)  → composeApp (as KMP framework)

Create local.properties in the project root:

FLOWISE_API_KEY=your_api_key_here

# For publishing only:
signing.keyId=...
signing.secretSigningKeyFile=path/to/key.gpg
signing.password=...

The Flowise backend URL is configured in:

• composeApp/src/commonMain/kotlin/com/sjaindl/app/di/KoinInit.kt
• assistant/src/commonMain/kotlin/com/sjaindl/assistant/config/DOMAIN.kt

# Build all modules
./gradlew build

# Run common/JVM tests
./gradlew :assistant:jvmTest

# Run Android host tests
./gradlew :assistant:androidHostTest

# Run a single test class
./gradlew :assistant:jvmTest --tests "com.sjaindl.assistant.TokenizerTest"

# Generate documentation
./gradlew :assistant:dokkaGeneratePublicationHtml

# Publish to Maven Central (requires local.properties signing config)
./gradlew :assistant:publish

# Run the desktop app
./gradlew :composeApp:run

The library is published to Maven Central using the Vanniktech Maven Publish plugin with PGP signing.

• Group: io.github.sjaindl
• Artifact: mobile-rag-assistant
• Version: set in assistant/build.gradle.kts
• License: MIT

MIT License

Stefan Jaindl