Skip to content

Repository files navigation

Edge Translator

Translate offline, without a word leaving this device.

Desktop app for macOS, Windows, and Linux. The model runs on the machine. After install, text, speech, images, and favorites stay local — no outbound request. An Android app is planned.

The interface is available in the same 35 languages as the catalog, including RTL (Arabic, Hebrew, Persian).

Edge Translator translating the project license from English to French on-device, in 4.8 seconds.

Features

  • On-device streaming translation, via Gemma 4 and LiteRT-LM
  • 35 languages (CC3M-35L / PaliGemma catalog), with automatic source-language detection
  • Proofreading: the same model corrects spelling and grammar in the language of the input
  • Dictation: speak up to 30 seconds, transcribed and translated on-device (Gemma 4 FLEURS, 12 languages)
  • Image translation: pick a photo or a screenshot, the model reads its text and translates it on-device
  • Read aloud: Piper voices, 30 languages, downloaded on demand
  • Two model profiles: Fast (E2B) or Precision (E4B)
  • GPU acceleration when available, otherwise CPU
  • Local favorites (search, pinning, filters)
  • Model and voice downloads you control: start, pause, resume, or cancel — Settings never starts a download on its own
  • First launch: resumable download, then SHA-256 verification; optional voice install
  • Desktop auto-update from GitHub Releases
  • About screen with open-source licenses
  • Follows the system light / dark appearance
Proofreading: messy phonetic English rewritten as clean prose, still on-device.

Proofreading. Same model, same machine. No grammar-checker account.

Favorites list with pinned translations in French, English, Japanese, Arabic, German, and more, stored locally.

Favorites. Contracts, tickets, a line of Arabic. Searchable, pinned, on disk.

First-launch screen: choose Fast or Precision, then download the model. After that, no connection is required.

One download. After that, the network cable can come out.

Languages

Capability Count Languages
Translation 35 French, English, Arabic, Bengali, Chinese, Croatian, Czech, Danish, Dutch, Filipino, Finnish, German, Greek, Hebrew, Hindi, Hungarian, Indonesian, Italian, Japanese, Korean, Māori, Norwegian, Persian, Polish, Portuguese, Romanian, Russian, Spanish, Swahili, Swedish, Telugu, Thai, Turkish, Ukrainian, Vietnamese
Dictation 12 French, English, Arabic, Chinese, German, Hindi, Italian, Japanese, Korean, Portuguese, Russian, Spanish
Read aloud 30 every catalog language except Croatian, Filipino, Japanese, Māori, Thai

Models

Weights are not in this repository. They are downloaded from Hugging Face on first launch (the only step that needs Internet).

Profile Model Size Repository
Fast Gemma 4 E2B IT · QAT 2/4/8-bit ~2.6 GB litert-community/gemma-4-E2B-it-litert-lm
Precision Gemma 4 E4B IT · QAT 2/4/8-bit ~3.7 GB litert-community/gemma-4-E4B-it-litert-lm

Recommended disk space for the Fast profile: ~3.2 GB (weights plus install headroom).

Piper voices come from rhasspy/piper-voices (~60–140 MB each) and are optional.

Gemma models are covered by Google’s Gemma license, which is separate from this application’s license.

Requirements

  • JDK 25 or newer (to build or run from source)
  • Internet connection only to download a model, a voice, or an app update
  • Microphone optional (dictation)

Install

Prebuilt packages (DMG, NSIS, Deb) are published on GitHub Releases. Packaged builds update themselves from there.

Running from source

./gradlew :desktopApp:run

Hot reload:

./gradlew :desktopApp:hotRun --auto

Force the install screen: pass --onboarding as a launch argument.

Native packages (GraalVM — DMG / Zip / NSIS / Deb):

./gradlew :desktopApp:packageGraalvmDmg    # macOS
./gradlew :desktopApp:packageGraalvmZip    # macOS updater payload
./gradlew :desktopApp:packageGraalvmNsis   # Windows
./gradlew :desktopApp:packageGraalvmDeb    # Linux

Pushing a v* tag builds these packages for Linux (x64, ARM64), macOS (Intel, Apple Silicon), and Windows (x64), then publishes them as a GitHub Release.

Layout

shared/       logic, Compose UI, translation engine, data
desktopApp/   desktop entry point (Nucleus)

All application code lives in shared. The desktop module only starts the UI.

Android is planned: the shared UI is already Compose Multiplatform, but there is no shipping Android app yet.

License

Copyright (C) 2026 Elie Gambache

This program is free software: you can redistribute it and/or modify it under the terms of the GNU Lesser General Public License version 3 as published by the Free Software Foundation.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more details.

You should have received a copy of the GNU Lesser General Public License (LICENSE) and of the GNU General Public License (COPYING, incorporated by LGPLv3) along with this program. If not, see https://www.gnu.org/licenses/.

About

Offline Ai translator for desktop

Resources

Stars

11 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages