Attorney-client privilege and GDPR stop you from pasting case files into ChatGPT. Occulta redacts every document before it reaches the AI, then restores the real values in the answer. Your clients' data never leaves in clear text.
Affaire RG 23/04512. Mme Sophie Lemaire, demeurant 12 rue des Lilas, 69003 Lyon, s'oppose à M. Julien Lemaire. Pension de 850 € réclamée.
Affaire PIIDO01. PIIPE01, demeurant PIIAD01, PIIAD02, s'oppose à PIIPE02. Pension de PIIMO01 réclamée.
data in clear text sent to the AI
original values stored, the log keeps tokens only
third-party trackers on the page that sees your files
cookie banners, nothing to consent to
Attorney-client privilege and GDPR: pasting a named case file into a consumer AI tool exposes your clients' data and puts you at fault. So you either give up hours of time on summaries, notes and research, or you take the risk. Occulta removes the dilemma.
Paste part of a file, or keep the example. Occulta spots sensitive data and swaps it for opaque tokens, in real time.
Simplified demo, for illustration. The product uses a fuller detection engine (French rules and a language model) and human review before anything is sent. The sample file is in French: the engine is calibrated for French law.
You stay in control throughout. Nothing is sent without your approval.
The text stays on your machine. Occulta automatically detects names, addresses, amounts, case numbers, IBANs, social security numbers and company forms.
A preview shows exactly what will go to the AI. You approve it, or leave a false positive in clear text. You decide, not the machine.
The AI works on the redacted text. Occulta restores the real values in the result: for you, the answer is complete and readable.
Occulta is not one more AI. It is what makes using one defensible.
Every sensitive value is swapped for a token before any external call, then restored on the way back. The token-to-value mapping stays with you and is never sent to the AI.
Nothing is sent automatically. You see and approve what leaves, file by file.
Every operation is logged, types and tokens only, never the values. Enough to evidence your compliance.
Detection calibrated on French formats: social security and company registration numbers, IBANs, case numbers, legal entity forms.
Designed around attorney-client privilege and GDPR: AI without the ethical risk.
Move the sliders to match your actual workload. Every assumption is shown, nothing is hidden.
A conservative estimate, not a guarantee. Formula: files × 12 × time saved. Freed-up time is never fully re-billed, so we only count half of it, then we deduct the subscription matching your volume. Actual savings depend on how you use it.
Every review is posted on a public platform, and the link goes to the original message: you read it in context, with its author.
“Reviewing what gets masked before the LLM call is a really thoughtful touch. Tested it on a client contract and the IBAN swap worked cleanly with the values re-appearing exactly where they should in the response.”
The full loop, redaction and restoring the AI's answer, is
included from the free plan on. The Starter plan opens without a credit card, and you
can change plan whenever you want.
One file = one send to the AI, once you have approved the review. Analysing a text,
seeing the detected data and unmasking a false positive cost nothing.
To try Occulta on your own files.
For the lawyer handling files every day.
For a multi-person firm, on a single invoice.
Features marked “in progress” are not shipped yet. They are coming soon and are included in your subscription as soon as they go live, at no extra cost and with no price change. Everything else in this table works today.
VAT not applicable, Article 293 B of the French General Tax Code. The amounts shown are the amounts charged, no tax is added. Indicative pricing: Occulta assists with document processing, it does not replace professional judgement.
With Occulta, the text sent to the AI is already redacted. But some firms cannot, or will not, let anything out, even redacted. For them, we install everything on your hardware: Occulta and the local AI model. Unplug the internet and it keeps working.
"Local" detection is offered elsewhere, but their AI still calls an external service. Here, the model itself lives with you.
Describe your need in
two lines: we reply within 48 working hours with a priced proposal.
Just a question? Use the contact page
instead, it is quicker: this form is for pricing an installation.
A redaction tool detects the personal data in a document, names, addresses, emails, phone numbers, IBANs, social security numbers, amounts, case numbers, and replaces it before any transmission. Occulta goes further: it replaces each value with a reversible token, sends the redacted text to the AI, then restores the real values in the answer. You get a complete, readable result, without the sensitive data ever leaving in clear text.
From the Latin occultare: to hide, to conceal, to put out of sight. Nothing to do with the occult or esotericism, the kinship is purely etymological. The word says exactly what the tool does, and it carries the distinction that matters: an occultation, in astronomy, is one body concealing another without destroying it. The masked value is not lost, it is behind, and it comes back in the answer. That is also what our mark draws: a ring with a disc covering part of it.
Anonymisation is irreversible: the person cannot be identified again, so the data falls outside GDPR. Pseudonymisation replaces identifiers with tokens, keeping the mapping separately, it stays covered by GDPR (art. 4-5) but is a recognised security measure (art. 32). Occulta performs reversible pseudonymisation: the mapping table never leaves your environment and is never sent to the AI.
Not by pasting a named file as-is. Attorney-client privilege and GDPR both forbid it: sending identifying data to a third-party service is disclosure to an unauthorised party. Working on text that has been pseudonymised beforehand removes the obstacle: the AI tool receives nothing that identifies the client.
You import the PDF (Pro and Firm plans), Occulta extracts its text in memory — nothing is written to disk — detects the personal data and replaces it with tokens. You review what will be masked, leave in clear anything that is not sensitive with one click, then send. The original file never leaves your browser: only the redacted text reaches the AI, and the real values come back in the answer. Same for Word, ODT, Excel and CSV. A scanned PDF is an image and holds no text at all: run it through OCR first, otherwise there is nothing to redact — and nothing to analyse either.
Occulta is designed as a technical protection measure within the meaning of GDPR article 32 (pseudonymisation). The audit log keeps only tokens, data types and counters, never the original values, enough to document your processing. Occulta is a tool: your firm's compliance also depends on your own records and procedures.
This is where the need and the risk peak at the same time: compliance research — GDPR, AML/CFT, sanctions screening, contract clauses — is by definition about material you are not allowed to disclose. Redacted text keeps the facts, the chronology and the shape of the reasoning, so the analysis stays intact; what it no longer keeps is anything identifying. The audit log records every send — tokens, data types, counters, never values — which is what lets you show, on paper, what left and what never did.
Pasted text works from the free plan on. Importing PDF, Word (.docx), ODT, Excel and CSV files is available on the Pro and Firm plans.
Yes. Five request templates are one click away: case summary, timeline of events, letter to the client, weak points of the case, short summary. This is deliberate: asking a professional to "write a prompt" is precisely where an AI tool stops being obvious. The template stays editable, and the answer exports to Word, PDF, plain text or the clipboard, formatting included, ready to be filed.
No. The engine is calibrated on French law, which makes it the most precise option for a law firm, but the data it detects is the same everywhere: notaries, accountants, HR, insurance, banking, public bodies. Any organisation handling named files that wants to use AI without exposing them.
Yes, as long as the document never leaves in clear text. The data that identifies an accounting file is exactly what Occulta detects most reliably, because it can be verified by computation: IBAN (exact length per country), French RIB (mod 97 checksum), payment card (Luhn checksum), company registration numbers, amounts in euros and in foreign currencies. A case reference or a product code is therefore not masked by mistake.
By masking on the way in, not on the way out. In a financial workflow — reconciliation, dunning, expense claims, reading a contract — what leaks is almost never the reasoning, it is the identifier: bank details, account number, beneficiary name, amount. Occulta sits before the model call. And a second pass rescans the redacted text just before sending: if an IBAN, a national ID number, an e-mail or a company number is still there, it is neutralised. The guarantee does not rest on the first pass alone.
One file means one send to the AI, once you have approved the review. Everything before that is free: pasting text, importing a document, seeing the detected data, unmasking a false positive — you can start over as many times as needed without using your quota. Only the send counts. Asking two different questions about the same matter therefore counts as two.
In France. Detection runs on our European servers, and the Sovereign offer allows a 100% on-premise install, with a local AI: in that case no data, not even redacted, leaves your premises.
Create your account and handle your first five files for free, no credit card. You see exactly what gets redacted before anything reaches the AI.