Translategemma - Offline Screen Translator

繁體中文

A powerful, lightweight, and completely privacy-focused screen translator. Select any area on your screen, and get instant Traditional Chinese translations via native notifications. Powered by Ollama (AI) and RapidOCR (Offline OCR).

Features

• 🔒 Privacy First: Everything runs locally. No data is sent to the cloud.
• 🚀 Lightweight: Uses RapidOCR for fast text extraction.
• 🤖 AI Powered: Uses Gemma (via Ollama) for natural translation.
• 🖥️ Cross-Platform: Works on Windows 11 (Native Notifications) and Linux (notify-send).
• 📂 System Tray: Minimized background operation without clutter.
• 📋 Auto-Copy: Translations are automatically copied to your clipboard.
• ⌨️ Global Hotkey: Press Shift + Alt + Z anytime to trigger.
• 📸 Infinite Snipping: Optimized persistent overlay engine.

Installation

Prerequisites

1. Ollama: Download and install from ollama.com.
2. Model: Download translategemma-4b-it-GGUF from HuggingFace.
   • Place the .gguf file in the models/ directory.

Quick Start (Desktop App)

Windows

1. Download the Standalone Distribution (ZIP).
2. Extract the folder.
3. Run setup.bat once to import the model into Ollama.
4. Run Translategemma.exe or run.bat.

Linux

1. Ensure you have python3 and pip installed.
2. Make scripts executable: chmod +x *.sh.
3. Run ./setup.sh to import the model.
4. Run ./build.sh to compile the binary, OR just run ./run.sh to start directly.

Developer Setup (Source Code)

If you want to run from source:

# 1. Clone repo
git clone https://github.com/yourusername/translategemma.git
cd translategemma

# 2. Setup Env (Cross-platform)
python -m venv .venv
# Windows:
.venv\Scripts\activate
# Linux:
source .venv/bin/activate

# 3. Install Deps
pip install -r requirements.txt

# 4. Import Model
# Windows:
setup.bat
# Linux:
./setup.sh

# 5. Run
# Windows:
python src/app.py
# Linux:
python3 src/app.py

Usage

1. Run the application.
   • Windows: Translategemma.exe
   • Linux: ./dist/Translategemma/Translategemma or python3 src/app.py
2. The app will minimize to the System Tray.
3. Press Shift + Alt + Z.
   • Note: On Linux, if the global hotkey doesn't register (due to permissions), right-click the Tray Icon and select "Snip".
4. Draw a rectangle around the English text on your screen.
5. Receive the translation via Notification and Clipboard!

Project Structure

Translategemma/
├── src/
│   ├── app.py           # Main entry point (System Tray & Logic)
│   ├── snipper.py       # Cross-platform Screen Selection Tool
│   ├── utils.py         # Platform specific logic (Notifications, Geometry)
│   └── ...
├── models/              # GGUF Model and Modelfile
├── dist/                # Pre-built executables
├── build.bat / .sh      # Build Scripts
├── setup.bat / .sh      # Setup Scripts
└── run.bat / .sh        # Launcher Scripts

License

Application License

The source code of this application is licensed under the MIT License. See LICENSE for details.

Model License

The Translategemma model is provided under and subject to the Gemma Terms of Use. By using this application with the Gemma model, you agree to comply with these terms. See NOTICE for the mandatory use declaration.

Third Party Licenses

This project uses several open-source libraries (RapidOCR, Ollama, etc.). Please see THIRD_PARTY_NOTICES.md for the full list of third-party licenses and attributions.

Acknowledgements

• Google - Translategemma: For the inspiration and the powerful Gemma family of models.
• mradermacher/translategemma-4b-it-GGUF: For providing the quantized GGUF model optimized for offline use.
• Antigravity: For the advanced AI agent assistance in building this project.