assistant-ui Generative UI

Always consult assistant-ui.com/llms.txt for the latest API.

Generative UI inverts the usual tool-rendering relationship. Instead of one hand-written component per tool call, you ship a vocabulary of components and let something assemble a tree from it: either the model, through the present tool, or your backend, through a message part or a LangGraph event. @assistant-ui/generative-ui ships a default vocabulary of 29 components (cards, facts, tables, charts, forms, controls) plus converters that turn the same tree into Slack Block Kit or a Microsoft Teams Adaptive Card, and one that turns an A2UI surface into it, so a composition built once can render in the browser and outside it.

References

• ./references/vocabulary.md -- the 29 default components and how to add your own with defineGenerativeComponents
• ./references/actions.md -- _action dispatch, createActionRegistry, and the $input shapes interactive components send back
• ./references/renderers.md -- MessagePrimitive.GenerativeUI and its allowlist, the Slack and Teams converters, and why the two spec shapes do not interchange
• ./references/a2ui-openui.md -- A2UI surfaces over AG-UI and the third-party OpenUI integration
• ./references/tokens.md -- the shared token arrays and how to restyle or override the vocabulary

Which generative UI pattern?

Two questions separate the patterns: does the model compose the layout or do you bind it ahead of time, and does the UI originate from a tool call or from a part your backend emits.