Features explained
Interactive visuals the assistant builds for you
Ask for a calculator, a slider or a small simulation and the assistant can build a working one inside its reply. It runs in a sealed frame with no access to your account and no web requests allowed, and the reply keeps a text version.
Charts, comparisons and choice cards cover most answers that are better seen than read. Some questions need something you can move: a slider that changes a forecast, a calculator with your own numbers, a small simulation of how a process plays out.
For those, the assistant can build a custom interactive visual and show it inside its reply. You use it right there in the chat.
What it is
A custom visual is a small working page that the assistant writes for one answer: controls you can drag or type into, and a result that updates as you do. It can draw charts, and it follows your light or dark theme.
It appears in a card in the reply with its title and an "AI-generated" tag, the same as the built-in visuals. The assistant wrote it from your conversation, so check its logic and figures before you rely on them.
How to get one
Ask for something interactive in plain words: "build me a calculator for this", "make that interactive", "give me a slider for the growth rate", "simulate this". Custom visuals appear when you ask for one. The assistant does not add them on its own.
They are available on models that support them. On other models, or when a built-in chart or comparison fits the request better, you get one of those instead, or a text answer.
On a turn where a visual may be shown, the reply arrives all at once instead of word by word.
What a custom visual can and cannot do
A custom visual runs in a sealed frame inside the reply.
- It cannot see your account. It has no access to your sign-in, your other conversations, or anything else on the page around it.
- It is allowed no web requests. It cannot load anything from another site or call another site's address. Everything it needs is inside it. These rules do not close every route a browser offers, so do not type passwords or other secrets into a visual.
- It cannot act for you. It cannot send a message, run a tool, or change anything in GPTree.
There are two things it can ask GPTree to do, and both leave the decision with you:
- Suggest a follow-up. A visual can place a question in your message box. Nothing is sent until you send it.
- Offer a link. If a visual wants to open a web page, GPTree shows an "Open link?" prompt with the full address first. The page opens only if you agree.
The text version is always there
Every custom visual is also kept in the reply as text: a short summary of what it shows, and the figures behind it as a table when it has them. That text is what the assistant reads on later turns. It is also what you see on a shared link, in Compare, on the Merge Board and in exports, and anywhere the live visual cannot be shown.
If a visual fails to load, the reply shows the text version in its place. You do not lose the answer.
Limits
A reply can carry up to three visuals of any kind. A custom visual is small by design: it is one self-contained piece for one answer, not a full application.
Availability
Custom visuals are part of visuals in chat, on plans that include the feature.