how it works
STEP 01
Type normally into ChatGPT, Claude, or Gemini — the extension scans every keystroke. Or paste any text into the web app. Wharx works wherever you are.
STEP 02
Real names, accounts, keys, IDs swap to typed placeholders the second they're detected. The protected version is what the AI receives.
STEP 03
When the AI replies, placeholders swap back to real values — inline in the chat with the extension, or in the web app after you paste the reply. The original details never crossed the wire.
who uses wharx
Any time you paste real work into an AI, you also paste a little bit of someone's private life. Wharx is for everyone who'd rather not.
Summarising a deposition without leaking client names, case numbers, or counterparty details.
Drafting patient notes and discharge instructions without exposing PHI to the model.
Preparing client summaries with account numbers, balances, tax IDs, and wire details protected.
Source-protected interview transcripts. Names, locations, and contact details stay private.
Screening résumés or drafting feedback without sending candidate data to a third-party AI.
API keys, connection strings, customer data — all swapped before a single prompt is sent.
what wharx catches
Luhn for cards. MOD-97 for IBAN. Verhoeff for Aadhaar. Issuance rules for SSN. Wharx confirms a match is real before redacting it.
for developers
The redaction engine that powers wharx.com ships as a zero-dependency SDK, a CLI, and an MCP server. Drop it into any agent pipeline. Same patterns, same validation, same privacy guarantee.
the promise
Wharx is a tool, not a service. The redaction engine runs in your browser tab — there is no “wharx server” that sees your prompts, your detections, or the AI's reply. The privacy guarantee is structural: there's no server in the round-trip.
ready when you are
Open the app in this tab. No signup. The whole product runs in your browser — including this page.