Philter Desktop vs Payroll Data Shield

If you have been looking for a way to strip personal data out of something before it touches ChatGPT, you have probably found Philter Desktop. It is a real tool from a real team, and it deserves credit for something a lot of “redaction” software skips: when it redacts a PDF, it rebuilds the page as an image rather than drawing black boxes over live text. There is no copy-pasteable ghost text hiding underneath. That is a stronger guarantee than most products in this category offer.

It also handles a genuinely wide set of formats — PDF, DOCX, XLSX, CSV, TXT, RTF and email, with on-device OCR for scanned PDFs — and runs fully offline, with a policy editor covering categories like Personal, Contact, Location, Financial, Identifiers, Technical and Medical. It is built on Phileas, an Apache‑2.0 open-source engine. If your job is putting a document pipeline behind a compliance policy, that is the right shape of tool, and $100 per user per year is a fair price for it. There is a free tier, but it is licensed for personal and evaluation use, not professional work.

The problem it does not solve

Redaction destroys the thing you were trying to work with.

A bookkeeper pasting a payroll query into ChatGPT does not just want the names gone. They want an answer back that they can actually use. Redact a client's name, NI number and salary, ask ChatGPT to draft the email, and what comes back is a draft full of [REDACTED] that you now have to fill in by hand — which means opening the real data again, next to the AI window, and retyping it. The risky moment you were avoiding just moved.

Payroll Data Shield does the round trip. It strips NI numbers, postcodes, emails, phone numbers and amounts out of text before you paste it, and then puts the real values back into the reply. You get a finished draft with the correct details in it, and nothing identifying ever left your machine. That is the whole product.

Side by side

Both tools are honest about the same thing, which is worth saying plainly: neither sends your data anywhere, and both publish their detection engine so you can read it rather than trust a marketing page. We would rather point you at a competitor that does that than pretend we are the only option.

If you are redacting a queue of scanned PDFs for a compliance team, buy Philter. If your actual problem is the five seconds between copying a payslip and pasting it into ChatGPT, that is what we built.

Try the detection logic before you buy anything.

The Redaction Desk runs the same detection core in your browser. No install, no account, and nothing you type leaves the tab. If it does not catch what you need it to catch, you have lost two minutes rather than £12.99.