Case file — Prompt confidentiality · Local processing
Neutralyse protects personal, confidential and strategic information on the device, before it is ever sent to an artificial intelligence.
For your organisation: request a demo
Local analysis PDF, Word & images (OCR) Protection before sending Data sovereignty Cloud or local AI
Your text — edit it, everything stays in your browser
🔒 Local vault — the mapping never leaves the device
You see the real text. The AI receives markers in place of detected values. The mapping stays on your device.
Open the chat and test neutralisation right away. In local mode, the AI model runs on your machine: nothing leaves it, not even neutralised.
Open the chat → No installation, no credit card.The desktop app adds key storage in the system keychain and signed updates — the recommended level for professional use.
Per-seat credit pool, spending caps and approved models per member, encrypted shared dictionary, GDPR audit report: Team mode is deployed with you.
Request a demo 20 minutes, on a representative business case.Contracts, client files, invoices, HR documents and legal opinions are used every day to work faster with AI. An employee pastes a document into an assistant to correct, summarise or translate it — and confidential information leaves the organisation through a perfectly ordinary gesture.
Neutralyse steps in at precisely that moment, before the sensitive data leaves the user's environment.
What these documents contain:
Today you have to choose: the power of ChatGPT, or GDPR compliance. Nobody should have to choose — here are the four beats that remove the dilemma.
Text and documents are analysed locally: personal data, contact details, banking data, technical secrets and confidential terms defined by the company.
Detected information is replaced with markers the AI can understand. The mapping between markers and values stays on the device.
The AI works on the markers — Claude, GPT, Gemini, Mistral, or a 100% local model, offline. Detected values are replaced before sending, and the review shows exactly what leaves.
Neutralyse puts the original information back in place — locally, automatically, only on your screen.
What you write, what the AI receives, what it replies — restored on your screen alone. Detected values were neutralized before sending.
No procedure to learn. No document to prepare by hand. No compromise on the quality of the answer.
Drag a file into the conversation: its text is extracted on the device, neutralised like everything else, and only the protected text goes to the AI. The file itself never leaves your machine.
Even a photo of an invoice or a scan works: text recognition (OCR, French and English) runs locally, without sending the image to anyone.
Invoice no. 2026-118 — Martin Bakery, for the attention of Camille Lefevre. Payment to IBAN FR76 3000 6000 0112 3456 7890 189.
Invoice no. 2026-118 — [ORG_1], for the attention of [PERSON_1]. Payment to IBAN [IBAN_1].
The image is sent nowhere: recognition runs on your machine, and the neutralised text goes to the AI — reviewable before sending.
Local extraction, up to 15 MB per file. Scanned PDFs are read by local OCR (first 12 pages) — the image never leaves your device.
The content is extracted on the device, ready to be corrected, summarised or translated — no copy-paste.
Photo, scan, screenshot: local OCR (French and English) extracts the text. The image never leaves the machine.
About forty formats — notes, exports, logs, source code — analysed like any other text.
Drag and drop anywhere in the window, or dictate your request: dictation is 100% local in the desktop app (on the web it uses the browser's voice service). Attachments, OCR and dictation are free.
Create a project and add your reference documents — files or web pages. Then ask your questions: Neutralyse finds the relevant passages and joins them to the conversation, neutralised like everything else.
You can't audit OpenAI. You can audit what you send it — message by message, and on a public test corpus.
Measured on 15 August 2026 on a labelled corpus of 18 samples (39 expected entities), documented and reproducible. A measured figure, not a guarantee: the single missed entity was an organisation name, down to the recognition model; structured types are covered by deterministic rules tested continuously. Replay the 40 examples →
What the measurement doesn't say: a detector can't count what it didn't see, and no automatic detection is perfect. That's exactly why review-before-send, the confidential-terms dictionary and message-by-message inspection exist: verifiable safety nets on top of automation.
Firewalls, antivirus and access control don't stop an authorised employee from pasting confidential data into a prompt. Neutralyse adds a layer of defence at the point of use.
Neutralyse turns several GDPR principles into a verifiable part of your teams' daily work.
The vault that links markers to real values never leaves the device. Detected values and their mapping stay with you, under your law and your control — whatever the AI provider, and it's verifiable by inspecting network traffic.
The right positioning: Neutralyse doesn't replace your cybersecurity and doesn't guarantee GDPR compliance on its own. It contributes to compliance, to PII protection and to reducing the cyber risks tied to using artificial intelligence — and each of these contributions is verifiable in the product.
Adopt AI without losing control of your information assets.
An operational solution rather than a ban that's impossible to enforce.
Protection at the exact point where data risks being sent to an AI.
Minimisation, pseudonymisation and privacy by design, made tangible.
Proofread, summarise, translate — with a simple, immediately productive experience.
Start for free, locally. Move to managed cloud when you need it — protection applies the same way in every case.
Prices exclude VAT, added at checkout where applicable. Local mode stays free and unlimited. Credits are only used for managed cloud AI: as soon as you have credits, every feature is open — the Independent plan is the entry door to managed cloud, and Pro / Enterprise differ only in the number of included credits. For teams: per-seat billing, a shared credit pool, per-member caps and an encrypted shared dictionary — request a demo.
No software alone guarantees an organisation's full compliance. Neutralyse provides a strong technical measure that contributes to minimisation, pseudonymisation and the security of processing — and whose effect is verifiable message by message.
No. Detected values are replaced with markers before sending. The mapping needed to restore the information stays encrypted, locally, on the device.
Yes. The organisation can use a model running entirely on the machine (nothing leaves, even neutralised) or a cloud provider configured in Neutralyse. Neutralisation applies in every case.
Neutralyse is designed fail-closed: when the required level of protection isn't available, sending is blocked rather than risking an unprotected transmission.
Yes. The company defines its own dictionary of clients, projects, products or confidential terms. These terms are always neutralised, on top of automatic detection.
Neutralyse makes your sovereignty independent of the AI provider. Sovereign offers shift the question to the host ("trust us, our servers are in Europe"); Neutralyse settles it on the device: in local mode nothing leaves at all, and with a cloud AI detected values leave replaced by markers — and the review shows exactly what goes out. The two approaches are in fact complementary: you can use a sovereign European provider within Neutralyse.
Text pasted directly, PDFs (text layer), Word documents (DOCX), images with local text recognition (French and English OCR) and about forty text and code formats — all extracted and analysed on the device, free of charge. Fully scanned PDFs are read by local OCR (first 12 pages) — the image never leaves the device. Postal-address detection currently covers the French format; emails, phone numbers, IBANs and cards are detected in international formats.
Yes. Neutralyse is a provider-independent protection layer: managed cloud AI is included in the Independent, Pro and Enterprise plans — Neutralyse IA, plus a catalogue of the market's leading models (Claude, GPT, Gemini, Mistral, Grok, Qwen, DeepSeek), with nothing to configure. Locally, an open-source model runs entirely on your machine. In every case, detected values are replaced by markers before sending, and the review shows exactly what leaves.
A team shares a per-seat credit pool, with spending caps and an approved-model list per member, e-mail invitations and consumption tracking. The shared terms dictionary is encrypted: the server stores only ciphertext and never keeps the key — each member's invitation link installs it on their device.
Local mode is free and unlimited. Managed cloud AI runs on credits: 29 €/month for the Independent plan (3,000 credits included), 99 €/month for the Pro plan (12,000 credits) and 299 €/month for the Enterprise plan (40,000 credits). The GDPR audit report and AI Act attestation are included in every paid plan, and whenever a credit balance is active. Pay-as-you-go credit packs are available at any time. See pricing.
The chat is free and opens in seconds. And when your organisation is ready to deploy, we'll be there to support you.
Open the chat nowNeutralyse: all the power of AI, without needlessly handing it your data.