Your data stays on your device
This app runs the Gemma 4 language model entirely in your browser using WebGPU. Your documents, images, and analysis results are never sent to any server. All processing happens locally on your machine.
What is downloaded
When you first open the app, the following are downloaded once and cached by your browser:
- Model weights from HuggingFace via
cdn.jsdelivr.net
- JavaScript libraries: Transformers.js (model inference), pdf.js (PDF rendering), Mammoth (Word conversion), html2canvas (document-to-image), Marked (Markdown formatting)
After the initial download, the app works offline (as long as the files remain in your browser cache).
What is NOT sent anywhere
- Documents you upload (PDF, Word, images, text files)
- Text you type or paste
- Prompts you create or edit
- Analysis results generated by the model
None of the above ever leave your browser.
Third-party CDN note
The JavaScript libraries and model files are loaded from cdn.jsdelivr.net at runtime. While this is a widely trusted CDN, if you require maximum assurance, you can self-host these dependencies to eliminate any third-party involvement entirely.