Data and AI

What the built-in assistant does

The assistant explains and summarises what is already in your record, with a citation to the source. It does not diagnose, does not interpret results and does not recommend treatment.

Every substantial statement it makes about your record carries a reference to the specific entry or document it came from, with a label and a date, so you can open the source and check it. When the record does not contain enough to answer, it says so instead of filling the gap.

What the built-in assistant cannot do

• It cannot diagnose, prescribe, change or stop treatment.
• It cannot triage, and it is not a route to emergency care.

If you connect ChatGPT or another external AI service

You choose the access before the connection is created. Read-only access can retrieve only the sections you grant. Read/write access can read, create and update records in those sections and can request deletion.

A delete request from an external AI service does not delete anything by itself. Kanzen creates a short-lived confirmation and executes the deletion only after you approve that specific action in Kanzen.

You can revoke the connection at any time from Sharing. Revocation stops future access, but it cannot pull back information that has already been sent to the external service.

The external service is a separate recipient chosen by you. Its own terms and privacy settings govern what it does with data after disclosure; Kanzen cannot control that service after the data has left Kanzen. For ChatGPT that recipient is OpenAI, and it is established outside the European Union.

What may be retrieved depends on the sections you grant: conditions, allergies, medications, immunisations, procedures, implants, encounters, measurements, laboratory results, document metadata, the journal, nutrition and the health profile. A section you do not grant is never queried, so nothing about it reaches the connected service — not a value, not a count, not an empty list.

Records written by a connected service are stored as agent-entered and patient-reported. They are what you told the assistant, transcribed; they are not a clinical record made by anyone qualified, and Kanzen adds no interpretation to them.

Where your record is stored

Your record is stored in the European Union. The application runs in the Frankfurt region and its database is in the eu-central-1 region.

Where AI processing happens

When you use the AI assistant, the relevant part of your record is sent to our AI provider for processing. We do not claim a processing region for that provider, because we have not established one we can prove. We would rather leave the question open than answer it with something we cannot verify.

The provider does not store the conversation: every request is sent with storage switched off.

We also do not claim that your content is excluded from provider training. That would require a written commitment we do not yet have, and we will not describe it as settled before it is.

General information from the web

The assistant can look up general educational information. Before it searches, the query passes through a layer that removes identifying detail, and anything it finds is shown separately from the facts in your record, with its source, so the two are never blended.

Your control

You can delete a conversation, and you can opt out of AI processing entirely. With AI switched off, the record itself remains fully usable — the assistant is an addition, not the product.

If something looks wrong

If the assistant states something you believe is incorrect, the citation is the place to start: open the source and see what it actually says. Report anything that looks wrong to zen@kanzenkai.org. An answer that cites nothing should be treated as an answer that proves nothing.