For lawyers, accountants, freelancers and anyone who writes to an AI about other people
Work with AI, without handing it your clients’ details.
filterit sits between you and the AI. The names, tax ids, phone numbers and bank accounts it detects become labels before your message leaves. The AI answers as usual and you read the reply with the real details.
Start for free Try it first, no account needed
- No card
- Built for Greek documents: ΑΦΜ, ΑΜΚΑ, IBAN
- Servers in the EU
One minute: what you write and what reaches the AI provider
- Web chat
- Agent
- Browser extension
A scripted demonstration with fictional people, no sound.
What happens today when you write to an AI chat
Whatever you paste reaches the company that runs the AI in full: the client’s name, the tax id, the bank account. With filterit the same text arrives, with labels in their place.
You write
My client Thanasis Mavridis (tax id 123456783) owes 2,400 € in rent. Write him a polite reminder.
The AI provider receives
My client [NAME_1] (tax id [AFM_1]) owes 2,400 € in rent. Write him a polite reminder.
You read
Dear Mr Mavridis, this is a friendly reminder that 2,400 € in rent is outstanding.
You write as always
No new way of working and no deleting names by hand.
The AI understands just the same
The same label stays with the same person through the conversation, so the answer makes sense.
You see what left
Every message shows what was masked. If something slipped through, you add it before sending.
Try it now with your own text
Paste an email or a paragraph and see what would reach the AI. No account. Only you see the result and the text is not stored.
You try the demo on your own responsibility: check the result, because anything not detected stays as written. Prefer synthetic details or your own, not those of third parties.
Three ways. Pick the one that suits you.
The same protection, whether you talk to the AI inside filterit or stay in the chat you already know.
Web chat
The simplest way
You open filterit in the browser and talk to the AI there. Nothing to install.
- Conversation, documents and email in one place
- Every message shows what was masked
- Works on a computer and on a phone
You need: a free account.
filterit Agent
For those who stay in ChatGPT or Claude
A program on your computer. You keep using the AI chat you already use and the agent masks the details behind it, before they leave the device.
- The replacement happens on your own computer
- It runs quietly and tells you what it masked
- Also for files you upload to the chat
For Windows. Included in the Team and Enterprise plans and in pilot access.
Browser extension
The light option
A small add-on in your browser. It masks the details inside the page of ChatGPT, Claude, Gemini and Copilot, the moment you press Send.
- You see the labels before the message leaves
- The reply shows with the real details
- One key to mask what you have typed
For Chrome, Edge and Firefox, with a filterit account. Availability per browser is shown on the download page.
In your own work
Four examples from an ordinary day. The details are fictional.
Lawyer
You ask: Summarise the formal notice and propose a reply.
The AI reads: [NAME_1] owes [COMPANY_1] 8,300 € under the lease of [DATE_1].
Names of the other side, case numbers, addresses and tax ids that are detected become labels. The legal question stays whole.
Accountant
You ask: Check the payroll calculation.
The AI reads: [NAME_1] (ΑΜΚΑ [AMKA_1]) earns 1,450 € gross and is paid to [IBAN_1].
Greek tax ids, ΑΜΚΑ and IBANs are recognised with their check digits, the way Greek documents write them. The amounts stay, so the calculation comes out right.
Freelancer
You ask: Write a quote for the client.
The AI reads: Quote for [NAME_1], phone [PHONE_1], for a renovation at [ADDRESS_1].
Engineer, consultant, estate agent, insurance broker: the provider does not need your clients’ contact details for a good text to be written.
Individual
You ask: Write a request to my bank.
The AI reads: I am [NAME_1], phone [PHONE_1], and I ask to change my address to [ADDRESS_1].
Requests, appeals, CVs, messages about the family: your own details and those of your people go back into the text after it is written.
Common questions, in plain words
What exactly does the AI company see?
Your text with labels, for example [NAME_1] or [AFM_1], in place of the personal details that were detected. Who [NAME_1] is, only your side knows.
Will the AI understand my request without the names?
Yes. The same label always stands for the same person within a conversation, so the AI can tell who is who. In the reply you read, the labels have become names again.
Does it catch everything?
No, and no automatic tool does. That is why we publish measured accuracy per kind of detail and show you on every message what was masked, so you can add what is missing. Whatever is not detected leaves as written.
Is this the same as anonymisation?
No. It is pseudonymisation: the mapping between a label and the real detail exists, encrypted, precisely so the details can return in the reply. The data remains personal data and the responsibility for using it lawfully stays with you.
Where are the mappings kept?
In the web chat, encrypted on servers in the European Union. With the agent and the extension, on your own computer.
Do I need technical skills?
No. The web chat works like any AI chat: you write and press Send. The agent and the extension are installed once.
What does it cost?
You start free, without a card, with a monthly usage allowance. If it proves useful, there are plans for professionals and teams.
Your next message to the AI can leave without the names.
Free to start, no card. It takes a minute.