Generative UI

An assistant message is normally markdown text. Generative UI (GenUI) lets the agent render real components instead: tables, forms, charts, and cards, inline in the message and streaming as the reply arrives.

Turn it on with one prop. On OpenUI Cloud, the components and model instructions are managed for you.

import { AgentInterface } from "@openuidev/react-ui";
import { openuiLibrary } from "@openuidev/react-ui/genui-lib";

<AgentInterface llm={llm} componentLibrary={openuiLibrary} />;

openuiLibrary covers layout, content, tables, charts, forms, and buttons, enough for most agents. To define your own components, see the OpenUI Lang overview.

Interactivity

Generated components are interactive, and the runtime wires up the behavior. When a user fills a form or clicks a button, one of two things happens:

• It continues the conversation. The interaction goes back to the agent as the next turn, carrying the current field values.
• It updates in place. The component refreshes data, changes a value, or opens a link, with no round trip to the model.

Form state is tracked automatically and saved with the thread, so a half-filled form survives a reload. See Interactivity for the details.

GenUI vs artifacts

Use GenUI when the UI is the reply: a form to fill, a table to scan. Use artifacts when the output is a durable thing the user returns to on its own surface. An agent can do both in one thread.