A2UI Protocol Demo

A2UI (Agent-to-User Interface) is an open-source protocol by Google that standardizes how AI agents generate user interfaces through declarative JSON. Instead of returning plain text, agents emit structured surfaces—forms, cards, lists, notifications—that clients render natively in any framework. This solves the “Chat Wall” problem and enables rich, interactive agent experiences without sacrificing security.

While AG-UI provides the imperative runtime for manipulating existing interfaces, A2UI provides the declarative specification for generating new ones. Together, they form a complete stack for agent-driven UI.Learn about our GenUI + AG-UI service →

a2ui.org GitHub

What is A2UI (Agent-to-User Interface)?

An open-source protocol by Google for standardizing how AI agents communicate and generate user interfaces through declarative JSON.

Core Design Principles

• Declarative JSON: Agents request pre-approved UI components from a trusted catalog—no executable code
• Decoupled Architecture: UI structure (agent-generated) is separate from UI implementation (client-rendered)
• LLM-Friendly: Flat, streaming JSON enables incremental generation and progressive rendering
• Framework-Agnostic: Same A2UI JSON renders natively in Angular, Flutter, React, or SwiftUI
• Security-First: Declarative-only approach prevents UI injection attacks

Protocol Versions

• v0.8 Stable: Core surface types, component catalog, streaming primitives
• v0.9 Draft: Adds createSurface, client-side functions, custom catalogs

The Problem A2UI Solves

• Breaks past the “Chat Wall”—agents generate structured UI (forms, date pickers, cards) instead of text-only responses
• Eliminates platform-specific agent code—one JSON spec, every client
• Provides a security boundary between agent output and client rendering

A2UI vs AG-UI: Complementary Protocols

A2UI and AG-UI address two different sides of agent-driven interfaces. They are designed to work together:

A2UI = Declarative Spec

Defines how agents describe UIs. The agent emits structured JSON that declares what components to render, with what data, in what layout. The client maps JSON to native components.

AG-UI = Imperative Runtime

Defines how agents manipulate existing UIs. The agent sends commands to show, hide, modify, or animate interface elements that already exist on the page.

How the Interactive Demo Works

1. Agent Generates JSON

Given a user prompt, the agent streams a declarative A2UI surface as flat JSON tokens.

2. Progressive Rendering

As each component is parsed, the client renders it immediately—no waiting for the full payload.

3. User Interaction

Rendered components are fully interactive. User actions can trigger the agent to update the surface.

Ship agent-generated UIs with confidence

We integrate A2UI surfaces with your design system, add guardrails, and connect to your agent runtime.

Explore A2UI integration

Sources & References

Official resources for the A2UI protocol

Official

• a2ui.org — Protocol specification and documentation
• github.com/google/A2UI — Reference implementation and examples

Related Demos on Virgent

• AG-UI Control Demo — The imperative side of agent-driven interfaces
• GenUI + AG-UI Service — Our production offering for agent UI