Anonymize your documents before using them with an AI.

When you paste information into an AI like ChatGPT, Claude or Gemini, that data might be stored in their systems: to train future models, for human review or simply kept in logs. If your document contains real names, IDs, addresses or medical data, you are sharing people's information who probably haven't given their consent.

Anonymizing beforehand prevents that problem: the AI receives the context it needs but without knowing who it belongs to.

Yes. The General Data Protection Regulation (GDPR) considers personal data to be any information that identifies a person: a name, an ID, an email, a license plate, an address. Sharing this data with a third party (including an external AI) without a legal basis can be an infringement.

Proper anonymization turns the document into something that is no longer personal data and, therefore, falls outside the scope of GDPR. This is what the AEPD calls "irreversible anonymization". With Ocultia, it's reversible only for you, thanks to the key file stored locally.

No. Ocultia is a support tool that makes it easier to anonymize documents before passing them to an external AI, but it is not a legal advisory service and does not guarantee regulatory compliance.

The responsibility for processing personal data always lies with the person or organization managing it. Ocultia cannot guarantee that it detects 100 % of the personal data in every document. AI models make mistakes, and certain formats (standalone images, complex tables, low-quality scans processed by OCR) may evade detection. That's why it is essential to review the result before exporting.

In no event will Ocultia or Lostium be liable for any damage or harm arising from the use of the tool, the processed information, or the decisions made based on its results.

If you manage particularly sensitive data or in regulated contexts, consult a data protection specialist.

No. Once the application is downloaded to your browser, all the magic happens on your computer: neither the document nor the detected data leaves your device.

The only thing downloaded from outside is the AI model the first time you use it (a few tens of MB). Then it is kept in the browser cache to work offline.

Yes, as long as your browser doesn't have extensions that analyze page content with AI. Ocultia doesn't send anything to the internet, but some extensions (AI assistants, correctors, automatic translators, note clippers) can read what you type on any website and send it to their own servers.

If you are going to handle highly sensitive data, review your extensions' permissions or use Ocultia in an incognito window with extensions disabled.

We combine two complementary techniques:

1. Regular expressions for data with a fixed and verifiable format: ID and NIE (with control digit), IBAN, credit card number, email, phone, IP address, postal code, date. These detections are fast and highly accurate.

2. Named Entity Recognition with AI (NER) for data without a fixed pattern: names of people, organizations, places, schools, companies. We use a small Transformer model trained for Spanish that runs directly in your browser with WebAssembly.

Detected data is replaced by tags like [NAME_1], [ID_2], [EMAIL_3]… preserving the original format of the document.

Because Markdown is practically the native format that AIs understand best, like ChatGPT, Claude or Gemini. It maintains the document structure (titles, lists, bold text, links) with very little syntax, without the visual noise of PDF or Word, takes up fewer tokens, and the AI interprets it effortlessly.

Furthermore, it's plain text: you can open it in any editor, copy and paste it wherever you need, or keep editing it manually. When you recover the personal data, the AI response will still come in Markdown; rehydration works just the same.

Automatically: ID, NIE, IBAN, credit card number, email, phone, IP, URL, date, postal address, postal code, person's name, organization, place and school.

In addition, you can define custom types on the fly; useful for sensitive terms in your business context (e.g., PROJECT, CLIENT, MEDICATION, ALIAS). Just select the text in the document and choose "+ new type".

That's why we let you review before exporting. You'll see the highlighted text by color depending on the detected type, and you can:

• Click any highlight to change its type or remove it.
• Select new text to mark it manually with an existing type or your own.
• Consult the complete table with all detections, their occurrences, and origin (regex, AI, or manual).

Nothing is exported until you decide.

Yes. Save the key file that downloads along with the anonymized document and go to the Recover tab. Upload whatever text you have (the original or one rewritten by the AI, as long as it keeps the tags [NAME_1], [ID_2]…) along with the key, and we will reinject the personal data where they belong.

We recommend encrypting the key with a password upon export if the data is sensitive. Without the password, the data cannot be recovered: not by you, nor by anyone else.

Yes, up to 5 files per session. Drop them together in the upload zone or add them one by one; the paste text area also counts as an additional virtual file. Before analyzing you can reorder them by dragging or remove them individually.

The big advantage is the unified mapping: if the same person, ID, IBAN or phone appears in multiple documents, they all receive the same tag [NAME_1]. This way, when you pass the set to the AI, it can relate the blocks without confusing the same name spread across different files.

When exporting, you get a single combined Markdown with clear separators (--- file-1.pdf ---, --- file-2.docx ---) between each block and a single key that covers the entire set. When the AI replies, you rehydrate just the same: a single pass that replaces the data in all blocks.

The ideal format is Markdown (.md) or plain text (.txt): the structure remains intact, no conversion is needed, and the anonymized result can be pasted directly into the AI. If your source is in Word, exporting it to Markdown from the editor itself will give you the best quality.

We also accept .pdf and .docx: we convert them to Markdown in your browser before analyzing them. It works well with digital text and simple formatting, but keep its limits in mind:

• Scanned PDFs or text embedded in images: if we don't detect a text layer, we automatically launch OCR in your browser with Tesseract (~12 MB downloaded the first time). The recognized text goes into the normal anonymization pipeline. You'll see a permanent notice on the review screen reminding you that the OCR might contain errors, so check the detections carefully.
• Complex PDFs (multiple columns, large tables, footers): reading order may get mixed up and titles might not be properly recognized.
• Word (.docx): we extract the text and basic formatting (titles, lists, bold text, links). What is not processed: embedded images, text boxes, shapes, headers/footers, comments, and track changes marks. None of that reaches the anonymized text: if there was personal data there, we won't detect it, but it won't be in the result either.
• Images in general (signatures, ID photos, screenshots, charts with text): they are completely ignored in any format. They are not analyzed and will not be present in the anonymized text you export: the result only contains the text we actually managed to extract.

In any case, before exporting, review the document on the review screen and mark anything missing by hand.

Yes, with Ocultia Enterprise. You upload a .docx or a .pdf, we automatically detect personal data and you review it: add or remove whatever you want manually. Upon download, this data is removed from the file and replaced with a black bar, keeping the original document layout intact.

It is irreversible, meant for delivering the document to a person or publishing it in a transparency portal, not for an AI. Editing while keeping formatting is an Enterprise feature: contact us to activate it in your organization.

Once the model is downloaded (the first time you detect something), Ocultia works completely offline. You can even install it as an app from your browser.

The public version is totally free, requires no registration and has no usage limits other than your own browser's memory. Very large documents (hundreds of pages) might take longer to analyze.

There is also a paid Enterprise version with advanced features (like editing documents while preserving format), in addition to custom developments and integrations for your organization. Ask us for more details.

We don't use tracking cookies or profile you. To understand how Ocultia is used and improve it, we do measure website usage with an anonymous and cookieless analytics (Matomo self-hosted by us, not Google): it counts visits and actions in an aggregated way, without identifying you and without touching the content of your documents, which are processed entirely in your browser.

Since it is anonymous and cookieless, it doesn't need a consent banner. Still, you can disable it in Settings whenever you want, and we respect your browser's Do Not Track signal.