Generative UI: in-app answers that are screens, and tasks that get done
Last updated 13 October 2026
Generative UI is an interface put together for each request: a person asks in their own words, and the answer is a screen of real components, filled with their own data, that can also carry out the task. Rendel is a Generative UI SDK for Flutter and web apps: it adds this to the app you already have, using your components, your API and your rules.
What it is
In a generative interface the model decides what to show and the app decides how it looks. Asked “where did my money go this month?”, the app answers with a spending chart and the largest payments, not a paragraph about them.
The same answer can lead to an action: set a budget, freeze a card, return an item. Reads run at once; anything that changes data waits for the person to confirm.
Chatbot, voicebot, or an assistant inside the app
They solve different problems, and many companies run more than one.
| Chatbot widget | Voicebot / IVR | Generative UI in the app | |
|---|---|---|---|
| Where it lives | A bubble on a page | On the phone line | Inside your app, as its own screens |
| What an answer is | Text and links | Speech | Components with the person's own data |
| Can it do the task? | Usually hands off to a person or a page | Within scripted flows | Through your app's actions, after a confirmation |
| Who it knows | Often anonymous | The caller, after checks | The signed-in user, through your own token |
What it needs in a real app
Actions: the things the assistant may do are your existing API calls, described once. Parameters are validated against each action's schema before anything runs.
Confirmation: every change shows a card that Rendel's server builds from your action, never from the model's wording, and runs only after the tap. Every action is on the record in the console.
Identity and limits: actions run with the signed-in user's own token, so your API's permissions still decide. A disabled action is refused from the next turn, without a release.
Data: Rendel's data is processed in Frankfurt (eu-central-1). Emails and card numbers that users type are masked before the model sees them, and data is never used to train models.
Native components, on Flutter and the web
Every component is drawn natively, by Flutter or by your web page; nothing is a web view. Each build tells Rendel which components it can draw, so an older version of your app is never sent one it cannot show. Swift, Kotlin and React Native are planned and not shipped.
How to add it
Your own coding agent sets Rendel up: it adds the SDK to your app and fills in your Rendel console. Every setting can also be entered by hand in the console.
Conversations on a test key are not billed, so you can try it against your own API before anyone else sees it.
Questions
Is generative UI only for chat screens?
No. The answer is a page of components, and it can sit wherever the app opens the assistant. People can type, speak, or attach a photo or a file.
Can the AI move money or change an order on its own?
No. A change runs only after the person confirms it on a card built by Rendel's server, and it runs as that person, under your API's rules.
Does Rendel hold a SOC 2 report or an ISO certificate?
No. We do not claim certifications we do not have. What we can show is how the system works, and that data is processed in the EU, in Frankfurt.