For your contacts
No identifiable person leaves your browser.
Names, emails, phone numbers, IBANs, social security numbers, addresses, ID documents, medical information.
- Personal data
- Bank details
ONYRI Sanitize finds the private bits in your text — names, emails, card numbers — and hides them before you send anything to an AI. The real answer comes back to you, complete. Everything happens on your computer. Nothing is sent.
100% in-browserno install0 bytes sent
Ticket #4821 — Customer Marie Dupont (marie.dupont@acme.fr) reports a duplicate charge on IBAN FR76 3000 4000 0312 3456 7890 143. Callback available at 06 12 34 56 78 before 6 p.m.
Ticket #4821 — Customer [PERSON_NAME_001] ([EMAIL_001]) reports a duplicate charge on IBAN [IBAN_001]. Callback available at [PHONE_001] before 6 p.m.
38 types across 6 families. Three areas to protect — framed by what you no longer want leaving your machine.
For your contacts
Names, emails, phone numbers, IBANs, social security numbers, addresses, ID documents, medical information.
For your systems
AWS · GCP API keys · OpenAI, Stripe, Anthropic keys · Slack tokens · inline passwords · SSH keys.
For your company
Internal project names, infrastructure URLs, meeting markers, salaries, company IDs, client and partner names.
Pasting a customer export, a support ticket or an HR spreadsheet into an LLM sends it outside your perimeter — with no way back. ONYRI steps in before that send.
Names, IBANs, API keys and internal decisions leave the company in the clear. Once sent, they don't come back — and you have no control over how they're used.
The AI only ever sees anonymous tokens. You keep the mapping — locally — and the full response comes back on your side.
13 detectors plus your own rules spot sensitive data, live, as you paste your text.
Each value becomes a reversible token. The text, now clean, can go to the LLM of your choice.
Paste the model's response: tokens turn back into your original data, exactly in place.
13 detectors plus your own rules spot sensitive data, live, as you paste your text.
Each value becomes a reversible token. The text, now clean, can go to the LLM of your choice.
Paste the model's response: tokens turn back into your original data, exactly in place.
Six families: personal, financial, passwords, API keys, company, strategy. Coverage designed for both French and international contexts.
Add your own patterns (internal references, business identifiers) with a regex and a dedicated color.
Paste text or drop a CSV / Excel file: tokenization applies column by column.
Beyond 1,000 rows, processing moves to a dedicated worker: the interface stays smooth.
Save a combination of detectors and rules, and replay it in one click every session.
Share rules and profiles across a workspace: the whole team applies the same policy.
Pick a plan and start a 7-day Pro trial: €0 today, charged only if you continue after the trial. Cancel in two clicks, any time.
To explore the tool on occasional use cases.
For freelancers and regular use.
€0 today · 7-day trial, no commitment
For teams that share rules and profiles.
€0 today · 7-day trial, no commitment
For organizations with specific requirements.
Practical guides, compliance references and fundamentals for using AI without exposing your sensitive data.
Your ChatGPT conversations aren't read by a machine alone. When a human can step in, what OpenAI keeps, the setting to switch off — and the only fix that covers the content.
Read articleChatGPT keeps “memories” across conversations — including wrong facts. How to inspect, correct or erase them, and why turning memory off isn't enough.
Read articleYour Claude conversations aren't out of reach. When a human can step in, what Anthropic changed in 2025, what you can't opt out of — and the only fix that covers the content.
Read articleCreate your account and reach the app in seconds. Your data stays with you — from the very first paste.