OCR
Turn scanned pages into editable, searchable text — entirely on your device.
How it works
- Open a scanned PDF (one with no selectable text layer).
- Pick the Edit text tool (
E), then choose OCR this page. - Tesseract runs locally in WebAssembly. Each recognized line becomes a real editable text box with the page background sampled behind it.
- Edit the recognized text like any other line, then export — flat scans become clean vector text.
Languages
Four languages ship with the app:
| Language | Code | One-time download |
|---|---|---|
| English | eng | ~2.8 MB |
| Español | spa | ~1.1 MB |
| Français | fra | ~0.6 MB |
| Deutsch | deu | ~0.8 MB |
Each language pack loads once (on first use) and is cached afterwards — OCR works fully offline from then on.
Privacy
OCR is fully on-device. The scan never leaves your machine; the only network activity is the one-time language-data download, which you trigger yourself.
TIP
After OCR, recognized lines behave like native text: you can edit them in place, and export produces real searchable, selectable vector text.
Automate it
The CLI and MCP server currently focus on text editing, redaction, and page operations. For scanned documents in scripts, pre-OCR in the app first, then automate against the text layer.