Document Reader
Paste text from a historical document — Latin phrases, archaic French, old legal formulas — and get a plain-English explanation. The AI model runs in your browser after loading model code and weights from external providers.
WebGPU not detected. This tool requires a modern browser with WebGPU support: Chrome 113+, Edge 113+, or Firefox with WebGPU enabled. The model load button has been disabled.
AI Model
Model not loaded. Click Load model to begin — first load downloads and caches model files in your browser.
Output will appear here after analysis.
Verify everything
AI models can make mistakes, especially with unusual names, ambiguous abbreviations, or rare languages. Treat output as a helpful first reading to assist your own transcription and interpretation — always check against the original document.
Privacy note
The app code passes document text to the browser-based model, not to an ArchiveIndex server. This page imports WebLLM code and model assets from third-party providers, so avoid using it for sensitive records unless you trust those providers or self-host the assets.
Tips for best results
- Paste clean transcribed text rather than raw OCR output if possible.
- For Latin documents, the 1B model handles standard ecclesiastical Latin well. Use the 3B model for unusual dialects or complex texts.
- Break very long documents into sections — models perform better on focused passages.
- The model is cached in your browser after first download, so subsequent loads are fast.