You are talking to an AI.
The EU AI Act, Article 50, says you have to be told. This page says exactly what the machine does.
Last updated: 2026-10-14
You are told before you type
- Both chats on this site, the assistant and “Ask idem”, show one line above everything else: I am an AI system by idem UNIVERSAL. You are not talking to a person.
- On Telegram, that same line is the first thing the assistant says to you, at first contact.
- It is not in small print, and it is not behind a click.
Every file it makes is marked
- A table or a document ends with one line: Generated by an AI system (idem UNIVERSAL).
- Next to that line sits a machine-readable tag, [ai-generated: idem UNIVERSAL], so a program can see it too.
- A calendar file carries the same thing as a field of its own, X-AI-GENERATED:idem UNIVERSAL.
- If it had to assume something, the file also carries a “To check:” list of what to look at.
- The demo film on the home page is made from real pages captured in a browser, not generated by AI. The pictures on its demo page are AI-made and labelled, and the film’s first frame says so.
It prepares; you send
- One job, one answer from the model (a voice note is turned into text first). There is no chain of agents running off on its own.
- It gives you back an answer, the files it made, and a button for anything that has to go out.
- Nothing leaves without your tap. The engine has no function that sends to somebody else by itself.
- It never asks you for a password.
Which model answers
The text of your job, and any file you send, goes to Anthropic — the Claude Opus 5.5 model — which produces the answer. If that model’s safety check declines a job, Anthropic can retry it on another Claude model it picks for that case. A voice note is turned into text by OpenAI first. Your text is not used to train the model. A second provider can replace Anthropic behind one switch. “Ask idem”, the question box behind the round red button, uses Anthropic’s Claude Haiku 5.5 model and is told to answer only from the text of this site’s pages; idem keeps none of your words. When you tap “Talk to it” on the home page, the spoken conversation is handled by ElevenLabs and answered by Google’s Gemini model, which ElevenLabs connects, not by Claude. Recording is off, and no words of the call are kept, except what you ask it to pass to Leonard for a plan or a call.
Things it does not do
- It does not read emotions, and it does not recognise a face or a voice.
- It makes no decision about you: no score, no ranking, no automatic yes or no.
- It builds no advertising profile, and nothing it sees is sold.
- It is not legal, tax, medical or financial advice.
What it remembers, and how to empty it
The assistant keeps one plain letter about you — what you do, how you like things done. Say “show letter” and it shows you the whole of it. Say “forget” and it is deleted, together with your record, the files it made and the buttons it prepared.
How to reach one
Write to the address below, or book a call. A person answers, not the machine.