What the Assistant reads

The Assistant reads only what a request needs, never above the classification ceiling, and never a Restricted diagram. Secret-like values are removed before anything is sent, and you can read the exact text of your last question.

What goes with a request

A request sends a part of the diagram you have open, not your whole account. The app states the rule: a question reads the selection and one step around it, or the whole diagram when nothing is selected, in summary when it is large, keeping the elements you name.

The panel's ⋯ menu has a page, What the Assistant reads, that shows it for the diagram in front of you, right now:

• The diagram: about how much of it goes with a question; large parts are summarised.
• The open findings of the checks, so that answers can cite them.
• The notes you keep for this diagram. See Conversations and retention.
• Your library: for Research only, the diagrams you may read.
• The web: never without asking you first, question by question.
• Your organisation's standards: designs and reviews read the technologies to adopt or on hold, and the standard patterns. See Standards.

What is never sent

• A Restricted diagram. It is never sent to the AI service, on any plan.
• A diagram above the ceiling (next section), nor a view whose title block is classified above it.
• Secret-like values: passwords, keys and tokens in names, descriptions and properties are replaced before sending. The page counts them, for example 2 secret-like values are removed before sending.
• Your files after the request. Images and PDFs are used for the request and never kept. See Bring images and PDFs.

The classification ceiling

Every diagram has a classification: Public, Internal, Confidential or Restricted. The ceiling is the highest classification the Assistant may read.

Above the ceiling, the box says why, for example This diagram is classified Restricted: the assistant does not read it. Your settings let it read up to Confidential. In an organisation it adds Ceiling set by your organisation. Instant tools still work on that diagram, since they send nothing.

In an organisation, the owners set the ceiling: lowering it takes effect at once; raising it needs a recent sign-in, and on some plans a second owner or security admin. See Classification and AI ceiling.

See what was sent

Open the panel's ⋯ menu, then What was sent:

• after a question, it shows The exact text of your last question, with its size;
• before any question, it shows the exact text your next question would send.

This is what your question sends from the diagram. The iyedy server adds its own instructions before the AI service reads it. A longer run (a change, a review, research) lists what it read and did under How I got this, on its card.

The web

Web research is off by default. When it is on, a research request may ask the web: you see every question first, under Search the web?, and approve, edit or remove each one. Nothing from your diagrams goes with them. Web searches are counted apart from credits.

Who processes it

In iyedy, the Assistant uses our AI model provider, which processes data worldwide and deletes it within 30 days. iyedy does not use your diagrams to train AI models. An Enterprise organisation may use its own AI service instead, and an organisation can turn the Assistant off. See AI and your diagrams.

What this means for you

• Alone on your account: the Assistant is on, up to Confidential. Classify a diagram Restricted to keep it away from the AI completely, or turn the Assistant off in your AI settings: off stops every Assistant request from your diagrams at once, and local checks keep working.
• In an organisation: the Assistant stays off until an owner accepts the AI addendum, then reads up to the ceiling the owners set. See Turn on the Assistant for your organisation.
• Before a sensitive question, open What was sent to read what your next question would send.