Google open-sources A2UI 1.0, a cross-platform protocol letting agents reply with native UI
Google released A2UI 1.0 on November 12, the first stable version of its generative UI protocol, adding bidirectional function calling and accessibility extensions, with web, Android and iOS renderers plus agent SDKs, and adoption by Atlassian, Bloomberg and ByteDance.
A2UI is an open generative UI protocol Google announced in late 2025, letting agents reply with interface content rather than only text to users. Google formally released A2UI 1.0 on November 12, the protocol's first stable version. Developers can now pair it with open-source web, Android and iOS renderer SDKs and agent SDKs for Python, TypeScript and Dart. Atlassian, Alation, Bloomberg and ByteDance have already started building with it.
According to Google, in A2UI the agent describes what should be displayed while each application draws the interface with its own native components. Whether a user brings the agent into an in-house application or reaches it through a third-party assistant hosting Model Context Protocol apps such as Gemini or Claude, the result is the same low latency, high trust and full native fidelity. The official Google blog notes that A2UI 1.0 adds bidirectional function calling along with expanded support for accessibility and additional metadata.
The open renderers can draw A2UI screens on the web with React, Angular and Lit, render on mobile, web and desktop through Flutter, and use SwiftUI on iOS and Jetpack Compose on Android. The web and Flutter renderers also run inside Model Context Protocol apps, so A2UI content can appear in AI assistants that support the protocol. The protocol addresses the problem of restoring an agent's structured interface description to native controls in the host application, decoupling interface description from rendering implementation so the same description can be reused on web and mobile.