Source · release
A2UI 1.0: Build delightful, fast, reliable, agentic experiences
Summary
A2UI, an open protocol for agents to answer with interactive UI instead of text, reached a stable 1.0 on 2026-11-11. The agent describes a view (a form, a confirmation dialog, a chart), and each app renders it with its own native components; only components the developer has registered can appear. Version 1.0 adds bidirectional function calls, accessibility support, and extensions.
Key points
- Renderers: React, Angular, and Lit for the web (
@a2ui/react,@a2ui/angular,@a2ui/lit), Flutter (a2ui_flutter), SwiftUI, and Jetpack Compose (still v0.9.1, 1.0 "coming soon"). The web and Flutter renderers also work inside MCP Apps. - Agent SDKs: TypeScript (
@a2ui/agent), Python, and Dart, which help generate skills, translate messages, and correct errors. Python adds A2UI Macros for component expansion and data binding. - Bidirectional function calls: a rendered component can call back to the agent, for example to check availability while the user picks a slot.
- Upgrading from 0.9: follow the evolution guide in the repo
(
specification/v1_0/docs). Docs and quickstart are at a2ui.org; the repo is github.com/a2ui-project/a2ui. - The post does not state the licence or how A2UI relates to ADK or A2A beyond MCP Apps; the Python and Dart package links on the page point to the same PyPI package, so check package names before installing.
Why it matters
Generative UI lets agents collect structured input and show results without free text. Because the client controls the component catalogue, it avoids running agent-written HTML or code in the app.