A2UI (Agent-to-UI Protocol)

What it is

A2UI is a declarative, JSON-based protocol that lets an agent describe UI it wants shown to the user, while the client application decides how that UI is actually rendered. The client maintains a catalog of trusted components (Card, Button, DatePicker, Form, and so on), and the agent can only request components from that catalog with props — it cannot inject code. Released under Apache 2.0 by Google with contributions from CopilotKit and the open-source community. The v0.9 draft is framework-agnostic across web, mobile, and desktop.

Why it matters

Text-only agents create friction: users read a paragraph and then have to hunt for the right button in the app. A2UI lets the agent surface the exact UI needed at the moment it's needed — a date-time picker, a confirmation dialog, a pre-filled form — right inside the conversation. Because the client owns the component catalog, brand consistency and security are preserved. For enterprise Salesforce deployments, the agent can embed native Lightning components instead of dumping rendered HTML, keeping the experience inside the platform's design system.

Key components

• Declarative JSON schema — the agent sends a data payload describing what UI to show, not executable code
• Client component catalog — only pre-approved components (Card, Button, Form, etc.) can be requested
• Framework-agnostic (v0.9) — renders across web, mobile, and desktop
• Decoupled UI structure from implementation — the agent says *what*, the client decides *how*

How it works

1. The client exposes a catalog of renderable components to the agent at session start
2. The agent responds with an A2UI JSON payload referencing catalog components and their props
3. The client validates the payload against its catalog and rejects anything unexpected
4. The native renderer instantiates the components with the agent's data, inside the client's design system