A local AI ebook translation tool built with React and Ollama
Tarjuma translates .txt and .epub files between languages using a locally
running AI model via Ollama. No data leaves your machine.
It automatically chunks files into appropriately sized pieces for the model,
translates them sequentially, and outputs either a .txt file or a structured
.epub with chapter titles and table of contents preserved.
Any model available in Ollama can be used — not just TranslateGemma.
- Ollama installed and running
- A translation model pulled in Ollama (I recommend:
translategemma:12b) - Node.js v20 or higher
- A modern browser (Chrome, Firefox, Edge)
1. Install and start Ollama with CORS enabled
# Windows (PowerShell)
$env:OLLAMA_ORIGINS="*"; ollama serve
# Mac/Linux
OLLAMA_ORIGINS="*" ollama serveYou can also use the desktop app of Ollama. Simply open the app, head to settings, and enable the option 'Expose Ollama on the Network'
2. Pull a translation model
ollama pull translategemma:12b3. Clone and run Tarjuma
git clone https://github.com/AtifSiddiqui20/Tarjuma-AI-Ebook-Translator
cd tarjuma
npm install
npm run dev4. Open your browser to http://localhost:5173
- Configure your Ollama URL, model, source and target languages in the settings panel
- Upload a
.txtor.epubfile - Click Translate and wait for the progress bar to complete
- Click Download to save the translated
.epubor.txt! - (For advanced users) To change the prompt, simply open utils/buildPrompt.js and edit the prompt there.
- Only
.txtand.epubinput formats are supported (PDF on the roadmap) - Images in EPUB files are not carried over to the output
- Hyperlinks in EPUB files are not preserved
- Chapter detection accuracy depends on how well-structured the source EPUB is
- Translation quality depends on the model used and the language pair
| Model | Size | Best for |
|---|---|---|
translategemma:12b |
~8GB VRAM | Best translation quality, purpose-built |
translategemma:4b |
Low VRAM | For lower spec'd machines |
qwen3:8b |
~6GB VRAM | Good multilingual alternative |
- PDF input support
- Image preservation in EPUB output
- Translation memory (avoid re-translating unchanged chunks)
- Batch file processing
MIT © Atif