Add AI to your app.
Skip the server.
Tiny Kotlin Multiplatform models that run inside your app, on Android, iOS, desktop and the web. Offline. Private. Milliseconds.
Beaconlanguage detection
⚡ <0.1 ms · offline
¿Vienes a cenar hoy?
ESSpanish
8models, live today
0network calls
2–9MBper model
193×smaller, at best
01 — WHY
No round-trip. No bill.
Cloud AI API THE USUAL WAY
- ✕Needs signal
- ✕User text leaves the phone
- ✕Pay per request, plus a server
Hoverfly ON-DEVICE
- ✓Works in airplane mode
- ✓Text never leaves the phone
- ✓Free up to 10k devices / month
03 — START
Ship it today.
Zero to AI in three lines
- 1Add the dependencyMaven Central · Android, iOS, desktop, web
- 2Create the modelonce, off the main thread
- 3Call itno key, no sign-up, no server
// build.gradle.kts: commonMain implementation("io.github.rajumark:moji:2.0.0") // anywhere in your app Moji().use { moji -> moji.suggestions("Pay my bills") // 💰 💸 🧾 }
04 — SIZE
Tiny by design.
XLM-R1.1 GBF1 0.49
Comma7.8 MBF1 0.83
Drawn to scale · 144× smaller, and more accurate on Hinglish chat
Size vs the alternative we tested
Chalk1.8 vs 348 MB193×
Gatekeeper3.8 vs 711 MB187×
Comma7.8 MB vs 1.1 GB144×
Emotion6.4 vs 499 MB78×
Hideout8 vs 445 MB56×
Beacon8.5 vs 158 MB19×
Comeback4.6 vs 11 MB2.4×
HoverflyAlternativeBars to scale
05 — RESULTS
Head to head.
Hoverfly's own test sets vs the tools you'd otherwise use.
Comeback4.6 MB
ML Kit6.6 MB+
ML Kit also ships native libraries.
Clean98.8
False flag1.2
Even on slang and mixed-script text.
Top 588
Top 168
111 everyday messages in 22 languages.
Need a model that isn't here?
Your task, your data → a tiny offline Kotlin model.
