> ## Documentation Index
> Fetch the complete documentation index at: https://docs.libratech.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Anonymise a document

> Create an anonymised copy of a document in one click: personal data is replaced with placeholders, the original stays untouched, and you review the copy before you share it.

<video autoPlay muted loop playsInline src="https://mintcdn.com/libra-4206ec93/ANVYcgaHT_Trya7b/assets/videos/features/documents-anonymize.webm?fit=max&auto=format&n=ANVYcgaHT_Trya7b&q=85&s=e3bf8430f096dfa7ed4339e0e7f05c88" />

Libra creates an anonymised PDF copy of a document, with names, addresses, IBANs and other personal data covered by placeholders like `NAME_1`. Your original stays unchanged.

> [!WARNING]
>
> Anonymisation is done by AI. It can miss personal data and cover text that isn't personal. **Always review the copy in full before you share it.**

## Create an anonymised copy

In a project, you start anonymisation from the document viewer toolbar. Short documents take under a minute, long ones several minutes.

**1. Open the document**

Open a document from **Project files** in the document viewer.

**2. Click Anonymize**

![Document viewer toolbar with the Anonymize button and its tooltip, above an open letter](https://mintcdn.com/libra-4206ec93/ANVYcgaHT_Trya7b/assets/images/app/anonymize-button.png?fit=max\&auto=format\&n=ANVYcgaHT_Trya7b\&q=85\&s=60aa84d293fef9e576685038ce30dd0a)

Click **Anonymize** (shield icon) in the viewer toolbar. You can keep working while it runs.

**3. Review the copy**

![Anonymised copy in a new viewer tab: names, addresses and the IBAN covered by black placeholder boxes, Anonymized label in the toolbar](https://mintcdn.com/libra-4206ec93/ANVYcgaHT_Trya7b/assets/images/app/anonymize-copy.png?fit=max\&auto=format\&n=ANVYcgaHT_Trya7b\&q=85\&s=074ba8fbd7c6eb888a3f0230bf9b04a3)

The copy opens in a new viewer tab marked **Anonymized**. [Review it](#review-the-copy) before you share it.

In a database, open the **⋯** menu on the document's row and choose **Anonymize**. The copy opens as *document name (anonymized)*.

If the copy hasn't opened after a few minutes, click **Anonymize** again. A finished copy opens right away and doesn't count toward your [daily limit](#availability-and-limits).

## What gets covered

Each value gets a numbered placeholder, and the same value always gets the same one: every "Max Müller" becomes `NAME_1`, a second person `NAME_2`. An index page at the end lists the placeholders and how often each appears, never the original values.

| Data | Placeholder |
| - | - |
| Personal names | `NAME_1` |
| Street addresses, cities, postal codes, countries | `ADDRESS_1`, `CITY_1`, `POSTAL_CODE_1`, `COUNTRY_1` |
| Email addresses, phone numbers | `EMAIL_1`, `PHONE_1` |
| Bank account numbers, IBANs | `ACCOUNT_NUMBER_1` |
| Tax numbers, VAT IDs | `TAX_NUMBER_1` |
| ID card, passport, driving licence and social security numbers | `ID_NUMBER_1` |
| Vehicle licence plates | `VEHICLE_PLATE_1` |
| Company names, dates, monetary amounts | `COMPANY_1`, `DATE_1`, `CURRENCY_1` |
| Special categories under Art. 9 GDPR (health, religious or political beliefs, union membership, ethnic origin, sexual orientation), plus age, gender and titles | `SENSITIVE_PERSONAL_DATA_1` |

Email addresses, IBANs, phone numbers and dates are also matched by fixed rules, so they're covered even when the AI misses them.

## Review the copy

You're responsible for the copy you share. Work through it in this order:

1. **Read every page**, including headers, footers, tables and footnotes.
2. **Check the index page** for page ranges that *could not be scanned for PII*. Libra didn't check those pages, so go through them yourself.
3. **Take warnings seriously.** If you saw *Anonymization completed with warnings*, review especially carefully.
4. **Redact what's missing in a PDF tool.** You can't edit or approve placeholders in Libra.

Look hardest where automated anonymisation is weakest:

* **Scans and images.** Libra reads only the text layer. Image-only PDFs fail, and text inside images (stamps, signatures, a photographed ID card) isn't covered.
* **Inconsistent spellings.** A name split across lines, hyphenated or capitalised differently can stay visible in some places.
* **Indirect identifiers.** "The plaintiff's brother", a unique job title or a described event usually isn't covered.
* **Rare names and formats.** Nicknames, initials, foreign spellings and unusual number formats are missed more often.
* **Over-coverage.** Case numbers that look like phone numbers, or a court's address that matters to your case, can get covered too.

If Libra finds no personal data, the copy is identical to the original, metadata included. Placeholders also hide names, not context: someone who knows the case may still recognise it.

## About the copy

The anonymised copy behaves differently from your other documents:

* **It's an image-based PDF.** You can't select or search its text, and hidden text and file metadata don't carry over.
* **It can't be reversed.** Libra doesn't keep which value belongs to which placeholder, so keep your original.
* **There's one copy per document.** Clicking **Anonymize** again opens the existing copy, for you and for colleagues. You can't re-run it.
* **The index page language** is the creator's interface language at the time. Languages other than German, English, French, Italian, Spanish, Dutch, Polish, Czech, Slovak, Hungarian and Romanian get a German index page.
* **It's stored with the original**, follows the same access rules and is deleted with it. To find personal data, Libra sends the document's text to an AI model, as described in [GDPR compliance](https://docs.libratech.ai/en/resources/security/gdpr.md).

## Availability and limits

Anonymisation is on every [plan](https://docs.libratech.ai/en/resources/subscription.md), one document at a time. It works on anything that opens in the full PDF viewer ([File types & previews](https://docs.libratech.ai/en/features/documents/file-types.md)) once processing is finished.

* **Projects:** anyone who can open the document.
* **Databases:** only people who can edit the database. See [Roles and limits](https://docs.libratech.ai/en/features/databases/usage.md#roles-and-limits).
* **Not available** for [SharePoint databases](https://docs.libratech.ai/en/features/databases/sharepoint.md), chat citations, email attachments, anonymised or translated copies, or the [Word](https://docs.libratech.ai/en/integrations/word/setup.md) and [Outlook](https://docs.libratech.ai/en/integrations/outlook/setup.md) add-ins.

Each person can start 5 new anonymisations a day on a trial and 10 on Starter, Professional or Team. Failed runs count; reopening an existing copy doesn't. Limits reset at midnight UTC, and paid plans can use **Request higher limit** when they hit it.

## Troubleshooting

Find the message you see in the left column.

| What you see | What to do |
| - | - |
| *Anonymization failed* or *Could not anonymize "document name".* | The document may be scanned, damaged or password-protected. Check that you can select text in the original, then try again. |
| *Failed to start anonymization* | Try again. If it keeps happening, contact support. |
| *An unknown error occurred.* in a database | You've probably reached your daily limit. Try again tomorrow, or contact support. |

## Next steps

> [card] [File types & previews](https://docs.libratech.ai/en/features/documents/file-types.md)
>
> Which formats open in the PDF viewer.

> [card] [GDPR compliance](https://docs.libratech.ai/en/resources/security/gdpr.md)
>
> How Libra processes personal data.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.