Generative UI is an interaction pattern where an AI system returns structured, interactive interface elements instead of only text. An agent might generate a form, table, chart, card, review panel, configuration wizard, or approval component that the host application renders for the user.
In agent systems, generative UI is often used when plain chat is too weak for the task. A user may need to inspect data, edit fields, compare options, approve a step, or manipulate a visual object. The UI component becomes part of the workflow, not just decoration around a text answer.
Generative UI can be delivered through several mechanisms:
- MCP Apps, where an MCP tool returns an interactive UI resource;
- AG-UI events that connect agent state to frontend components;
- A2UI streams that describe UI surfaces and component updates; or
- application-specific JSON schemas rendered by a trusted client.
The key technical requirement is validation. A host should treat model-generated UI descriptions as untrusted data. It should validate component types, properties, actions, data bindings, and permissions before rendering or executing anything.
Generative UI is related to structured output, but it has a user-interface goal. Structured output may feed an API or database. Generative UI feeds an interactive surface.
OpenUI describes generative UI as rendering interactive components such as tables, forms, charts, and cards inside an agent interface in its Generative UI documentation.
The LLM Knowledge Base is a collection of bite-sized explanations for commonly used terms and abbreviations related to Large Language Models and Generative AI.
It's an educational resource that helps you stay up-to-date with the latest developments in AI research and its applications.
Previous