CORPUS Engine
Scrub, de-identify, redact or pseudonymize: whatever your team calls it, CORPUS finds the people and companies in email, chat, PDFs and scans, swaps them for consistent fakes or burns them out, and re-scans every output before it is written, inside a cloud account you control.
Email, Slack and documents keep the disagreements and judgment calls that make them worth training on, with the people and companies CORPUS finds swapped for consistent fakes. Held-back files are listed, so you know what didn't make it.
See identity linking →Scanned pages and photos are read with OCR (handwriting and Devanagari are in development); local detectors look for names, record numbers, dates, ages and contact details, and burn them out. The per-file audit report goes to your privacy officer or expert.
How release works →Each client and counterparty name becomes the same fake across email, CSV exports and spreadsheets; account numbers are replaced at their cell or byte offset, and PDF statements are burned out. Runs inside your own cloud account with no internet access during processing.
How it deploys →PDF values are burned out of the page; no text layer remains. In Word and Excel, choose redaction or a consistent fake per entity type. Every output is re-scanned, comments and hidden sheets included; a hit goes to a reviewer.
How the re-scan works →Every file gets an audit record. An independent verifier must approve every file before release; held files go to your reviewers, with the reason. The job runs inside your own cloud account with no internet access during processing.
Examples on this site are synthetic. Real records never appear here.
Computed in your browser; nothing is sent. This is what manual review costs you today, not a promise of what CORPUS saves. Held-back files still need a reviewer.
Tell us the formats, languages and volume; the founder runs CORPUS live on a synthetic look-alike.
Paid pilots only; we run a few at a time. Join the waitlist
