Moji 🙂

Website · All Hoverfly models · Maven Central

By Hoverfly. On-device emoji suggestions for Kotlin Multiplatform: Android, iOS, macOS, JVM desktop, JavaScript and WebAssembly. You give it a task or a message and get back the emoji that fit, in 22+ languages.

import io.github.rajumark.hoverfly.moji.Moji

Moji().use { moji ->
    moji.suggestions("Pay my bills")   // 💰 💸 🧾 💵 💳 …
}

• No dependencies. Inference is plain Kotlin in common code. There is no ONNX Runtime, TFLite or native code, and the library adds about 5 MB to an app.
• Private and offline. The model ships inside the library on every platform. There is no network, no permission and no telemetry.
• Fast. About 0.3–0.6 ms per suggestion once warm on JVM, Android, JS and Wasm, and about 2 ms on a mid-range Android phone (Moto G57 Power).
• Identical everywhere. Every platform is tested against the Python reference model on 443 vectors and gets the same top-5 emoji.

// build.gradle.kts: commonMain, or any platform source set
dependencies {
    implementation("io.github.rajumark:moji:2.0.0")
}

It's on Maven Central, so no extra repository is needed. Gradle picks the right artifact for each platform:

The Android-only 1.x releases are io.github.rajumark:moji:1.1.0 and, on JitPack, com.github.rajumark:moji:v1.1.0.

Same model, same device, three languages — suggestions run entirely on-device, no network round trip.

Hindi	Spanish	French
"जिम जाना है"	"Partido de fútbol esta noche"	"Réserver un vol pour Paris"

The KMP sample on each platform:

Android	iOS	Desktop	Web (Wasm)

import io.github.rajumark.hoverfly.moji.Moji
import io.github.rajumark.hoverfly.moji.SkinTone

val moji = Moji()                      // loads the model: tens of ms, do it off the main thread, keep one instance

val top = moji.suggestions("Pay my bills")
top.first().emoji                      // "💰"
top.first().name                       // "money bag"
top.first().confidence                 // 0.36

moji.suggestions("Great job thumbs up", limit = 5, skinTone = SkinTone.MEDIUM)   // 👍🏽 👏🏽 …

moji.close()                           // frees the model's memory

suggestions() is thread-safe and fast enough to call on every keystroke.

With coroutines:

val moji = withContext(Dispatchers.Default) { Moji() }

From Java:

try (Moji moji = new Moji()) {
    List<MojiSuggestion> s = moji.suggestions("Pay my bills");
}

Upgrading from 1.x on Android: Moji(context) still compiles in Kotlin (deprecated). The model no longer needs a Context, so switch to Moji(). Java code must change new Moji(context) to new Moji().

sample/ is a separate Gradle build that uses the published library, never the source. It resolves io.github.rajumark only from Maven Local, or from Maven Central with -PmojiRepo=central. It has a Compose Multiplatform app for Android, desktop and iOS, and a web page built for both Kotlin/JS and Kotlin/Wasm.

./gradlew :moji:publishToMavenLocal
cd sample
./gradlew :androidApp:installRelease
./gradlew :desktopApp:run
./gradlew :webApp:wasmJsBrowserDevelopmentRun     # or :webApp:jsBrowserDevelopmentRun
open iosApp/iosApp.xcodeproj                       # run the iosApp scheme on a simulator

moji/                          the library
  src/commonMain/              public API (Moji, MojiSuggestion, SkinTone) and the model in plain Kotlin
                               (internal/: Featurizer, SentencePiece, Network, UnicodeTables)
  src/{jvm,android,apple,js,wasmJs}Main/   the only platform code: NFKC normalization + model loading
  src/modelData/               moji.bin (int8 weights) · spm_pieces.tsv (tokenizer) · vocab.tsv (1000 emoji)
  src/commonTest/              parity with Python on 443 vectors, API, latency; runs on every target
sample/                        demo apps using the published artifacts
scripts/GenTables.java         generates UnicodeTables.kt (character classes) so every platform agrees
docs/                          website (rajumark.github.io/moji)

On JVM and Android the model ships as Java resources in the jar/AAR. Kotlin/Native and the web have no resources, so the build compiles it into the library (generateEmbeddedModel).

./gradlew :moji:jvmTest
./gradlew :moji:testAndroidHostTest
./gradlew :moji:connectedAndroidDeviceTest              # on a connected device/emulator
./gradlew :moji:iosSimulatorArm64Test
./gradlew :moji:macosArm64Test
./gradlew :moji:jsNodeTest :moji:jsBrowserTest
./gradlew :moji:wasmJsNodeTest :moji:wasmJsBrowserTest

The parity tests require identical featurizer ids and an identical top-5 to the Python reference on all 443 vectors (tasks, 22 languages and edge cases), on every target. Probabilities match to within 1e-4; the current maximum difference is 3e-6.

See PUBLISHING.md.

Moji is one of nine small on-device models by Hoverfly, all with the same install, the same free tier and nothing sent to a server.

See them all on the Hoverfly website.

Free for up to 10,000 monthly active devices. You don't need an API key, an account or a license file: add the dependency and ship. It works in commercial apps too, with no limit on how often each device runs it.

How devices are counted. A monthly active device is a device that runs Moji at least once in a calendar month. The limit applies separately to each product, each platform (Android, iOS, web…) and each Hoverfly model. Once a product passes it, you have 30 days to get a commercial license. The library keeps working and never checks in with a server.

Not allowed under any tier (unless agreed in writing):

• selling or redistributing Moji or its model on its own, or inside another SDK or library
• extracting, modifying, fine-tuning or retraining the model weights
• using the model or its outputs to train or distill another model
• reverse engineering the model or its file format
• offering it as a hosted API for others

Custom models. Hoverfly also designs and trains small, fast on-device models for your needs: moderation, classification, language detection, smart replies and more.

Contact for a commercial license or a custom model: raju348636@gmail.com or +91 63533 21951 (call or WhatsApp).

Full terms: Hoverfly Community License. Versions 1.0.0 and earlier were released under Apache-2.0.