NameMasker

Mask before AI. Unmask after. Nothing leaves your browser.

Try a sample:

Preparing the on-device name model…

The map file — placeholders, aliases, and your always-flag list — is the only sensitive artifact NameMasker creates. It is what makes Unmask work. Keep it with your student files; it is never transmitted.

How to use it

How NameMasker works Your device holds paste, mask, and unmask. Only masked text crosses to the AI service. Pasting the AI's version back and unmasking is an optional round trip, used when you want the names back. On your device nothing leaves your browser The AI service only ever sees placeholders 1 Paste or drop a letter, essay, transcript 2 Mask review and approve the flags Student A… 3 Work with any AI calculate a GPA, evaluate an academic path, tighten a draft — placeholders stay placeholders …Student A… 4 Paste it back the AI's version, placeholders and all 5 Unmask real names return, everywhere, instantly Steps 4 and 5 are optional — only when you want the names back
  1. Paste your text above, or drop in a PDF or Word document. Files are read on your device and never uploaded.
  2. Press Mask. Identifying information lights up: names, schools, emails, phone numbers, and passages that might identify the student through context.
  3. Review the highlights — click any one to approve, edit, or dismiss it. Caught something the scan missed? Select the text and press “Mask this,” or keep an always-flag list of names the scan should never miss. You know the student; the tool does not.
  4. Copy the masked text and use it with any AI service you like.
  5. When you get results back, paste them under Unmask. Real names return everywhere the AI kept the placeholders.
  6. Working with the same student across a season? Export the mapping and import it next session — the same student is always “Student A,” across every document.

Three kinds of flags

How you can trust it

First, a promise — from a person, not a company:

I’m Chris Bell, an independent educational consultant. NameMasker is the same masking engine at the heart of the full IEC dashboard I’m building for my own practice — I use it with my own students, and it seemed too useful not to share with the IEC community.

This site does not send your documents anywhere — not to me, not to a server, not to any AI company. Nothing you type or open here leaves your computer. And it’s free, of course.

— Chris

Here’s how that can be true. NameMasker is a website, but it doesn’t work like one. A normal website sends what you type to a server; this page, once loaded, sends nothing anywhere. Every step — finding names, masking them, restoring them — happens in your browser, on your machine. There is no server on the other end. There is nothing to send to.

Most of the work is deliberately old-fashioned: find and replace. “Maya Chen” becomes “Student A” the same way your word processor would do it — no cleverness, perfectly reversible. One part does use AI: finding the names. It’s DistilBERT, a small open-source language model tuned for exactly one skill — spotting names — that your browser downloads once and runs by itself, like a spell-checker. It reads your document on your computer, and nothing is sent out to be read. We name the model on purpose: a named model can be looked up and audited.

A promise is only worth what you can check. Four ways, easiest first:

  1. Turn off your internet. Load the page, disconnect, and use the tool. It works fully offline because nothing it does requires a server.
  2. Watch the network. Open your browser’s developer tools and observe that no request carries your text anywhere.
  3. Read the code. The entire tool is open source under the MIT license — every rule, every model call, every line between your document and your clipboard.
  4. Trust the incentives. This tool is free, has no accounts, and collects nothing. Its only value is its credibility.

One promise we will not make: NameMasker does not “guarantee anonymity.” No tool honestly can. It catches what software can catch, flags what software can only suspect, and puts you in the approval seat for everything. The review step is the product.

The full threat model — what this tool protects against, what it never does, and what can still go wrong — is written in plain language on the security page.