Developer Guide for Integrating Gemini Enterprise with A2UI

Existing AI agents could mostly only return text or markdown, which caused problems such as requiring multiple rounds of conversation or degrading the user experience when performing complex tasks like selecting a date or displaying a map.

To solve this, Google is introducing the open protocol A2UI, jointly developed with the Flutter and Gemini Enterprise teams. A2UI has agents return a JSON payload that describes a UI structure instead of text, allowing agents to directly render rich interfaces such as buttons, cards, and a ChoicePicker.

The key features of A2UI are as follows:

• Declarative: Because it sends data rather than executable code, it is highly secure, and the client only renders an approved component catalog.
• Streaming-friendly: Since it takes the form of a list of small JSON messages, the client can draw them on screen immediately as the LLM generates them.
• Framework- and transport-agnostic: It works across various frameworks such as Lit, Angular, and Flutter, and can be used with existing communication methods like A2A, WebSocket, and SSE as-is.

Using a restaurant search agent built with Google Agent Development Kit (ADK) and the A2A protocol as an example, this guide covers the process of implementing an interactive UI within the Gemini Enterprise environment.