Translate anywhere — even offline
Tulk runs Meta's NLLB-200 model entirely on your phone, so you can translate across 190+ languages without a connection. One download, then translate forever — on a plane, abroad, or anywhere the network isn't.
- Languages
- 190+
- Model size
- ~1.25 GB
- Network
- Offline
- Platforms
- iOS · Android
A closer look
From the first launch to face-to-face conversations — a quick tour of how Tulk feels in the hand.
Scroll to explore the app
What makes it different
Built for travel, fieldwork, and any place the network isn't.
Offline-first
Download the model once, then translate forever — no internet, no per-language packs, no surprises.
190+ languages
One unified model covers every supported language, from English and Spanish to Bhojpuri, Wolof, and Yue Chinese.
Chat mode
Threaded conversations you can name, save, and return to — your offline phrasebook that grows with you.
Face-to-face mode
Split-screen conversations with the partner's half rotated 180°, so each side reads in their own language.
Voice in, voice out
On-device text-to-speech and Whisper speech-to-text — speak a phrase, hear the translation spoken back.
Private by design
Text and voice are processed locally and never leave your device. No accounts, no tracking, no telemetry.
How it works
The NLLB-200 distilled model — an encoder and decoder in ONNX, ~1.25 GB together — covers every supported language. Its lifecycle has two stages.
One-time download
On first launch, Tulk downloads the model files to local storage. Downloads are queued, resumable, and continue in the background while you switch apps or lock the screen.
Cold-start load
Every launch primes the runtime from disk before translation is available. On iOS the encoder and decoder sessions load lazily and swap so only one is resident at a time; Android keeps a KV-cache decoder warm for faster generation.
Translate forever
Once primed, all 190+ languages are pickable at any time, fully offline — no further downloads, ever.
The journey
From splash to a saved conversation, in a handful of taps.
Engineered for the edge
An Expo + React Native app running an ONNX encoder/decoder runtime on-device, with per-platform session lifecycles, resumable downloads, and local SQLite + MMKV persistence.
Frequently asked questions
The essentials on how Tulk handles offline translation and privacy.
Want early access to Tulk?
Tulk is in private beta. Reach out to join the TestFlight, or to talk about offline, on-device AI for your own mobile product.