A2UI new

What it is

A2UI is Google’s answer to “how does an agent show a form, a card or a chart without generating code the client has to execute?”. The agent returns a JSON description of components and a data model. The client maps each component to its own widgets. Nothing executable crosses the boundary, which is the point.

How it works

An A2UI response contains:

• A component tree: buttons, text, inputs, lists, cards, with properties.
• A data model the components bind to.
• Actions that send user input back to the agent.

The official renderer is Lit (web components), with Flutter supported through Google’s GenUI SDK. React and Angular appear in the demos, and React, Jetpack Compose and SwiftUI are on the roadmap. The payload travels over AG-UI for a custom frontend or A2A between agents; REST is planned.

When to use it

Your agent needs to collect structured input or present structured output, and you want that to look native in each host app. If a text response is enough, skip it.

Watch out for

• Public preview. v0.9 is the stable line and v1.0 is a release candidate, so expect the component catalog to move.
• A2UI and AG-UI are often confused. AG-UI is how events get to the UI. A2UI is what UI the agent asks for.