Clean before AI^™ — PII masking before any AI, on your computer

PII (personally identifiable information) is any data that identifies a person: names, addresses, emails, phone numbers, IDs, IBANs, dates of birth. PII masking replaces it with labels such as [PERSON_1] or [IBAN_1] before the text goes to an AI, so ChatGPT, Claude or Gemini never see the real data. Clean before AI does this on your computer and puts the real data back into the AI’s answer — strictly speaking that is pseudonymisation, because the key stays with you.

Paste the text or open the file (PDF, Word, TXT or a scan) in Clean before AI. Names, national and tax IDs, IBANs, addresses, phone numbers, emails and companies are found on your computer and replaced with labels like [PERSON_1]. Copy the cleaned text into any AI, then paste the answer back to get the real data.

Pasting it into Clean before AI is safe: everything runs in your browser, nothing is uploaded — not even to us — and it works offline. The AI then receives only the cleaned text with labels. Numbers with checksums are found very reliably; check names and other details before sending — medical data is especially sensitive.

Slovenian, English (UK and US), Croatian, Serbian (Latin and Cyrillic, also as used in Montenegro), Bosnian, Macedonian, German, Italian, French, Spanish, Portuguese, Polish, Dutch (also Flemish in Belgium), Romanian, Czech, Slovak and Hungarian — including national IDs with checksums (EMŠO, JMBG, OIB, PIB, Steuer-ID, SVNR, codice fiscale, Partita IVA, NIR, SIREN, DNI, NIE, NIF, CPF, PESEL, NIP, REGON, BSN, KvK, CNP, CUI, rodné číslo, IČO, TAJ, adóazonosító jel, SSN, EIN, Medicare, NI, NHS, UTR) and grammatical cases, so “Janezu Novaku” is the same person as “Janez Novak”.

Yes. It recognises the language of every paragraph and uses that language’s knowledge of ordinary words, streets and squares (so “Piața Unirii”, “Kossuth Lajos tér” or “Najemca” are not taken for people). Names, addresses and ID numbers from all supported languages are still found in every text — a Polish name in a Slovenian email or a Dutch address in an English contract. Labels automatically follow the language of the text, and you can change them.

Any of them: ChatGPT, Claude, Gemini, Microsoft Copilot, Perplexity, DeepSeek, Mistral Vibe (Le Chat), Grok, Llama and others. The Chrome extension works directly inside ChatGPT, Claude, Gemini, Copilot, Perplexity, DeepSeek, Mistral, Grok, Poe, Qwen and HuggingChat.

Yes. It recognises passwords, API keys (OpenAI, GitHub, AWS …), tokens and private keys, internal servers and domain accounts, IPv4/IPv6 and MAC addresses, patient and record numbers (MRN, ZZZS, MBO, KVNR), court case, claim and policy numbers, and participant and sample IDs in research data — also in tables pasted from Excel or CSV, where it uses the column headers.

Yes. PDFs, Word documents and text files are read in your browser; scans and photos are read with OCR. You can also download a redacted PDF or image with black boxes — the hidden text is gone from the file.

It supports data minimisation and pseudonymisation — measures the GDPR requires or encourages (Arts 5, 25 and 32). Pseudonymised data is still personal data while the key exists, so you remain responsible for how you use AI tools; the tool helps you share far less.

Each person, company or number gets one consistent label for the whole text, so the AI understands who is who without knowing the real data. The key that links labels to real data never leaves your browser unless you save it.

Yes, but only our own small AI model that runs on your computer. Together with the rules it finds names, nicknames, addresses and companies that dictionaries don’t know, in 17 languages. In the Chrome extension the model is bundled and works offline; in the web tool it downloads once (~67 MB) when you switch it on and stays in your browser. Your text is never sent to any AI or server to be cleaned.

Yes. Add them to your known-names list — colleagues, clients, partners, project names or internal codes. They are hidden every time, in every grammatical form, in capitals and without accents (“Novak”, “Novaku”, “NOVAK”, “Novakova”). The list stays in your browser; an organisation can distribute it to all employees through a Chrome policy.

Clean before AI hides everything it finds. Words it is less sure about — a lone first name, a nickname, a word the AI model is not certain of — are also shown in a “Check” row. One click keeps a word in the text if it is not personal data; everything else stays hidden.

Yes — the web tool and the Chrome extension are free, with no account.