Remove personal details before you ask AI

Docudis finds names, emails, phone numbers, IBANs and addresses in text, PDFs, Word files and photos, and replaces them with labels like [PERSON_1].

You check the result, send it to your AI, and Docudis swaps the labels back in the reply.

Get the app How your data is handled

Processed on your phoneThe app has no server. Your documents stay on the device.

Kept out of cloud backupsHistory and word lists are excluded from backup and device transfer.

Open sourceThe apps are AGPL-3.0 and the engine is Apache-2.0. Anyone can read the code.

No ads, no subscriptionEvery feature is free for everyone.

See what gets replaced

You see every detection before you share anything, and you can add or remove items by hand.

INVOICE_MEYER.DOCX7 items

Hi , following our call on 12 March 2026, please find the invoice for €4,200. You can reach me at or . Payment to .

Best regards,
·

7 items replaced.

In this example the date and the amount stay visible.

Put the details back

The label list stays on your phone. Paste the AI's reply and Docudis restores the originals, as long as the reply keeps the labels intact.

[PERSON_1]Sarah Meyer

[EMAIL_1]t.laurent@example.com

[PHONE_1]+33 6 12 34 56 78

[IBAN_1]FR76 3000 6000 0112…

[PERSON_2]Tom Laurent

[COMPANY_1]Atelier Laurent

[ADDRESS_1]14 Rue de Rivoli, 75001 Paris

ID numbers from 19 countries

National ID, tax and phone formats. For 16 countries the number's check digit is verified too, so random digits are less likely to be flagged.

ATBECHCNDEDKESFIFRGBIEITJPNLNOPLPTSEUS

Your own word lists

Add client or project names that should always be hidden, and public names that should always stay visible.

Always hideProject Falcon

Always keepParis

Text, PDF, Word, photos

Paste text, open a file or take a photo of a paper letter. Share the protected text, or a protected copy of the file.

.txt.md.csv.pdf.docx.jpg.png.heiccamera

Privacy

How your data is handled

Each line below describes what the app does today. If something changes, this page changes with it.

Your documents: Read and processed on your phone. The app has no Docudis server to send them to.
Detection: Docudis rules, a name-recognition model that ships inside the app, and Google ML Kit entity extraction, all running on the device.
Network: The app itself sends nothing. Google ML Kit, which runs through Google Play services, downloads its entity-extraction model on first use and sends Google diagnostic and usage information such as device model, OS and app version, performance metrics, error codes and a per-installation identifier. It does not send your text or images.See Google's ML Kit data disclosure.
History and labels: Your latest 100 documents are kept in the app's private storage; older ones are deleted automatically. Excluded from cloud backups and device transfer. Delete one item, or everything with "Clear data on this device", at any time.
Ads and tracking: No ads, no advertising identifiers and no analytics of our own. No account or sign-in.
Source code: All open source. The Android and desktop apps are licensed under AGPL-3.0: docudis-android, docudis-desktop. The detection engine and all rule packs are licensed under Apache-2.0: docudis-core.You can check every line above against the code.
Who makes it: COMPANY, COUNTRY · Contact: stonetechdigital@gmail.com · Full privacy policy

Limits

What Docudis can't do

It can miss thingsNo automatic detection finds every personal detail. Read the protected copy before you send it.
It hides details, not meaningA diagnosis, a job title or a story can still point to a person. Docudis replaces identifiers such as names and numbers.
Labels must survive the replyRestoring works when the AI keeps labels like [PERSON_1] unchanged. If it rewrites them, those spots stay as labels.

The app

Free on Android and iOS

No account and no subscription. Available in English, Español, Français and 中文.

Google Play App Store

1Paste text, open a file or scan a photo
2Check every item that was replaced
3Send the protected copy to your AI app

Swipe for more →

For developers

Use the detection engine in your own software

Docudis Core is a Rust library with a C interface. A Dart binding is included, and any language that can call C can use it.

A versioned interfacedocudis_v1_* functions and schema_version: 1 JSON keep their behavior. Breaking changes get a new version namespace.
Shared test fixturesEvery binding is tested against the same conformance cases.
Core has no network codeIt only looks at the text you pass in. OCR, PDF parsing and ML models live outside Core.
Builds forAndroid, macOS, Linux, Windows and WebAssembly.

$ docudis --regions fr \
  "Email claire.martin@example.fr
   or call +33 6 12 34 56 78."

Email [EMAIL_1]
or call [PHONE_1].

#include "docudis.h"

const char *req =
  "{\"schema_version\":1,\"text\":\"...\","
  "\"regions\":[\"fr\"]}";

DocudisV1Buffer out = {0};
if (docudis_v1_detect_json(
      (const uint8_t *)req, strlen(req),
      &out) == DOCUDIS_V1_OK) {
  fwrite(out.ptr, 1, out.len, stdout);
}
docudis_v1_buffer_free(&out);

{
  "schema_version": 1,
  "detections": [{
    "type": "EMAIL",
    "value": "claire.martin@example.fr",
    "start": 6, "end": 30,
    "confidence": 0.95,
    "detector": "regex:universal:email",
    "source": "strongRule"
  }, …]
}

Read the Core docs →

Try it on your next document

Paste an email, open a PDF or take a photo of a letter.

Get the app Read the source