Optical character recognition runs entirely on your device via Tesseract — including the language model, which is downloaded once as a static asset. Your document never leaves the browser.
How it works
- Drop a scanned PDF or an image (JPG/PNG/WEBP).
- Start OCR — pages render, preprocess and recognize with live progress.
- Review the text with per-line confidence colors, then export TXT, JSON (word boxes included) or a searchable PDF.
Good to know
- Confidence colors are measurements from the engine: green ≥85%, amber 60–84%, red <60%. They indicate reliability — they are not accuracy guarantees.
- Best results come from clear printed text at decent scan quality. Handwriting and degraded photocopies will score lower.
- English is supported today; other languages require additional trained-data files.
- Searchable-PDF output embeds the recognized text invisibly over each page image — great for archiving scans.
- Scanned invoices can go straight from here to Invoice Extractor.
- Every step runs locally in your browser using Web Workers — your file is never uploaded to any server.
Frequently asked questions
Are my documents uploaded anywhere?
No. Recognition runs locally in your browser through WebAssembly. The English language model (~4–5 MB) is fetched once from this site as a static file; documents themselves never leave your device.
Why is some text red or amber?
Those are low-confidence words — the engine measured them as less certain. Always proof-read colored lines before relying on them.
Can OCR read handwriting?
Reliably no, and we won’t claim otherwise. Tesseract excels at typed text; cursive handwriting produces poor confidence scores.