AI News

Google Releases AI Agent UI Protocol A2UI 1.0

gihyo.jpUpdated Nov 12 21:35

AI summary

Google's A2UI team released version 1.0, the first stable release of A2UI, an open UI protocol for AI agents, on November 11, 2026. Agents can use JSON to describe interactive interfaces such as forms and charts based on the component catalog provided by developers, and these are rendered by an in-app renderer. Version 1.0 adds the ability for agents and renderers to call functions on each other and return results or errors, supports combining components from multiple catalogs on the same screen, and adds accessibility support and extension metadata. The official SDKs include React, Angular, Lit, Flutter, and SwiftUI rendering SDKs, as well as Python, TypeScript, and Dart agent SDKs; the Android Jetpack Compose renderer still corresponds to 0.9.1, with 1.0 support planned.

Why read this

SelectedAI score 77

A2UI 1.0 turns the interfaces returned by agents from plain text into interactive components and adds bidirectional function calling, allowing readers to judge how generative UI will be implemented.

Topics

Tags

Body · AI translation

On November 11, 2026, Google's A2UI team released version 1.0, the first stable release of A2UI, an open protocol that lets AI agents return interactive UIs such as forms and charts. In response to a user's request, the agent describes the UI's structure and displayed content, and the app can render it on screen based on that description. Version 1.0 adds features such as the ability for the agent and the rendering side to call each other's functions.

A2UI 1.0: Build delightful, fast, reliable, agentic experiences - Google Developers Blog

A2UI is an open-source project developed by Google, with contributions from CopilotKit and the open-source community as well. The source code is published under the Apache License 2.0 at a2ui-project/a2ui on GitHub.

Today we are releasing A2UI 1.0, now faster, easier, and stable, an open protocol and framework for generative UI across web, Android, iOS, wearables, TVs, and any other surface.

👇 Check out what shipped https://t.co/Hm5PhCIGtO— Google Cloud Tech (@GoogleCloudTech) November 11, 2026

Composing in-app UIs in response to user requests

In interactions with AI agents, users often convey in text the points they want changed or the options they choose from among candidates or plans presented in text. With A2UI, the agent can compose a UI to display within the app by combining components such as buttons and input fields prepared by the developer, in response to the user's request.

In the official flight booking example, the user tells the agent the destination, dates, and number of people in the app's chat. In response to that request, the agent returns a UI that combines flight information with a button to change the number of people and a button to confirm the booking. The user checks the flight information displayed in the app, changes the number of people as needed, and presses the confirm button to convey the intent to book.

Example flight booking UI. The developer provides the app foundation such as chat, and the UI inside the green dashed line is rendered with A2UI (from the Quickstart on the official A2UI website).

To display such UIs, developers prepare UI components on the app side. The available components and functions and how to use them are compiled into a definition file called a catalog, which is shared with the agent side.

An agent that receives a request while the app is in use describes the combination of components and the information to display based on the catalog, and sends it to the app as JSON-format data. Unlike approaches that send HTML or JavaScript code for the screen, a rendering process (renderer) built into the app reads this data and displays it using the prepared UI components. This mechanism can be used in web apps as well as native mobile and desktop apps.

In addition to the screen structure, the agent can also specify the behavior when a button is pressed. The booking confirmation button can be configured with an action that sends information such as the flight and number of people to the agent. When the user presses a button with this configuration, the app sends the intent to book and the specified information. The agent proceeds with processing based on the received information and can update the screen or data.

Support for bidirectional function calls and combining multiple catalogs

In 1.0, the agent and the renderer can now call each other's functions, and the called side can return execution results or errors [1]. For example, the renderer can call a function on the agent side and receive the result of checking flight availability.

Also, components and functions contained in multiple catalogs can now be combined and used on a single screen. The catalogs being combined must support the same A2UI specification version.

In addition, accessibility support and extension metadata for passing additional information were added to the specification. Changes from the 0.9 series are summarized in the migration guide.

SDKs to support UI rendering and agent development

As rendering SDKs compatible with A2UI 1.0, versions for React, Angular, and Lit are provided for the web. Flutter for mobile, web, and desktop, and SwiftUI for iOS are also available.

Meanwhile, the Jetpack Compose renderer for Android supported 0.9.1 at the time of announcement, with support for 1.0 planned for the future. Also, Gemini Enterprise already allows use of agents that use A2UI, and Google is working on A2UI 1.0 support in that service.

SDKs to support agent-side development are also provided, with TypeScript and Dart versions added to the existing Python version. These SDKs generate instructions based on the catalog to tell the model used by the agent how to describe the UI. They also validate whether the UI description output by the model conforms to the catalog definitions and convert it into A2UI messages to send to the app.

When the model describes the UI, using the Express format, which is shorter to write than JSON, reduces the number of output tokens. Because the SDK converts the Express-format output into JSON-format A2UI messages before sending them to the app, the app-side rendering process can be used as is.

Developers can also use Macros in the Python agent SDK to define functions on the server side that assemble reusable UIs. When the model outputs a description that calls such a function, the SDK executes the function on the server side and generates the description of the UI components to be rendered in the app. This means the model does not have to write out all the component descriptions needed for that UI every time.

Displaying UIs within ChatGPT and Claude conversations through MCP Apps

MCP Apps is a mechanism for displaying interactive UIs within AI assistant conversations. ChatGPT and Claude also support this mechanism. If a developer embeds an A2UI renderer into an MCP App, A2UI screens can be displayed within AI assistants that support MCP Apps.

As one example, Google introduced its use by Alation, which provides enterprise data catalogs and other services. Metric cards and graphs can be displayed within Claude conversations, and can be viewed in nearly the same layout within the Alation app as well.

Source

gihyo.jp