Redact documents before using ChatGPT, Claude, or any AI
The problem
Your team is already pasting documents into ChatGPT, Claude, Copilot, and Gemini. The productivity is real, and so is the exposure. A patient record, a financial statement, or a legal transcript sent to an external model creates a data-processing relationship you may not have intended: vendor risk reviews, DPAs, impact assessments, and in regulated industries, potential violations.
The risk isn't the AI. It's sending unredacted data to it.
The workflow
- Open zeroegress.ai. No signup, no install.
- Drop in your document, PDF, Word, Excel, PowerPoint, images, or plain text. Scanned documents are OCR'd automatically, in the browser.
- AI detection (a DistilBERT NER model running locally via WebAssembly) finds names, organizations, and locations; 17 pattern categories add email addresses, phone numbers, fax numbers, Social Security Numbers, credit card numbers, IBANs, IP addresses, URLs, dates, ZIP codes, account numbers, medical record numbers (MRN), patient and claim IDs, health plan beneficiary numbers, driver's license numbers, VINs, and national ID numbers. Review and adjust, click any entity to exclude it.
- Download the redacted version and upload that to your AI tool. Alongside the redacted file you can download a signed evidence pack, the signed audit trail, redaction map and compliance certificate, proving what was removed and that nothing left the browser.
Keep the context, lose the identity: reversible tokens
Static redaction ([REDACTED]) destroys the context AI models need to be useful. ZeroEgress can instead replace each entity with a reusable token:
Tokens: Dr. [NAME_1] emailed [EMAIL_1]
The AI sees consistent, distinguishable entities and can reason about them, "what did [NAME_1] agree to?", without ever seeing who they are. When the response comes back, the local Reveal feature restores the original values using a mapping that exists only in your browser. It is never transmitted anywhere.
Why in-browser matters
Every other approach to this problem moves your data somewhere: a redaction SaaS uploads your document to its cloud; an API-based scrubber sends your text to a third-party model. Each hop is a new vendor to assess, a new contract to sign, a new breach surface.
ZeroEgress processes everything client-side. There is no server to trust because there is no server in the data path. The architecture is the privacy policy, enforced by a strict Content-Security-Policy that prevents the page from making any external connection at all, and covered by US Patent 12,665,758.
What it costs
Free during beta. Team and enterprise plans will be announced ahead of general availability.