Create

Sign in to ReadmeX

or

How we handle your data: Privacy Policy

ReadmeX

Your AI reading companion

What would you like to catch up on?

Pick a briefing, or ask your own question.

Your community and people briefings stay personal to you.

Story

A2UI technical selection: GenUI client and Genkit TS server

AI summary

A technical article recommends GenUI for a Flutter client and Genkit TS for server-side agent orchestration, tool calls, catalog injection, validation, and streaming A2UI messages. It presents a three-frame JSONL TODO example built on the v0.9 envelope and basic catalog: createSurface to open a surface, updateComponents to define its interface, and updateDataModel to populate it. The article states that A2UI 1.0 had reached general availability while client renderers still defaulted to v0.9, and notes that teams already using Python with Google ADK may prefer to retain that stack.

In-depth insight

The article provides a concrete division of responsibilities for server-driven AI interfaces: Genkit TS prepares and streams structured messages, while GenUI renders catalog-approved components and returns user actions. Its example also shows how component structure and data can be updated separately, allowing A2UI output to be validated and rendered as protocol-defined frames rather than unstructured text.

A2UI 落地:技术选型——客户端 GenUI,服务端 Genkit TS

前几天在 DeepSeek 网页端问了个问题。答案还在一个字一个字往外蹦的时候,我没盯着页面,而是打开了 DevTools 的 Network 面板——那条还在传输的响应里,一行行 data: 开头的 JSON 正在滚出来。

那一刻我突然反应过来:所谓"流式输出",在传输层从来就不是打字机效果,而是一行一行的结构化帧。前端把每一帧拆开、增量渲染,才拼出了"AI 在打字"的观感。

A2UI 做的事,可以粗着理解成把"模型的输出是结构化帧"这件事标准化了:帧里装的不再是聊天文本的增量,而是界面本身——先给画布,再给组件,再给数据。思路与 Server-Driven UI 一脉相承:界面描述由服务端下发,客户端照着渲染,不掺私货。

上一篇《A2UI 初想:Agent 想画 UI,先得有一块画板》里,我们把 A2UI 的生命周期过了一遍:createSurface 开画布、updateComponents 上组件、updateDataModel 填数据、deleteSurface 收摊。文末留了个问题:真要跑一个 demo,两端各用什么框架?这篇把选型定下来。

#A2UI#GenUI#Genkit TS#Genkit Go#Genkit Python#Google ADK#Flutter#A2A

Original · 10 min readRead the original
Group chat

No comments yet. Start the conversation.