The Complete Guide to Document Redaction: Privacy-First OCR
May 2026
Passports, bank statements, medical records, contracts with personal data — sometimes the document itself is the sensitive thing, not just the result. Privacy-first OCR means the document never has to leave your device at all.
Three levels of privacy
- Server OCR with deletion: files are processed, then deleted automatically (free tier: 24h, paid: 90 days retention with manual delete anytime).
- Privacy mode: the document is processed but never stored in your history — no file, no result saved.
- On-device OCR (CrazyOCR Small): the image is read entirely in your browser with WASM — nothing is uploaded, ever.
On-device OCR in practice
CrazyOCR's Small engine runs a compact recognition model in your browser (WebGPU when available, WASM otherwise). The image, the model inference, and the extracted text all stay local. Only the text result is sent to your account when you keep it — and even that is skipped in Privacy mode. This is the same client-side sandboxing approach used for redaction workflows: no PII transits your servers, no retention policy to worry about.
Compliance note
We never train on your documents, and we honor deletion requests. See the privacy policy for retention periods and rights. For the most sensitive files, use Privacy mode or the on-device engine — the document never leaves your hands.
