Introduction

a2tea is the bridge that lets an AI agent drive a terminal UI with A2UI. It recognizes A2UI in a model's reply and draws it — instead of dumping raw JSON.

:::note Early, but real. Rendering works for the core catalog: all five input components are editable, modals open and close, and ChildList templates expand from the data model. The main remaining gap is interactive tab switching. See the roadmap in the README. :::

What is a2tea

a2tea parses the A2UI messages an agent emits — interleaved with conversational text in an LLM response — and renders the described surfaces as Bubble Tea models. The consumer is Joe's fork of charmbracelet/crush: a2tea is the bridge that lets crush spot A2UI in a reply and render it as a live surface.

Why a bridge

A host should not hand-roll detection of A2UI payloads in model output. It calls Scan (or the cheap Contains probe) and gets back the text and the typed A2UI messages, using the real A2UI wire format — then hands the messages to Render for an embeddable surface.

The two standards

A2UI · Google

A declarative protocol for agent-driven UIs. Agents send component descriptions — not code — that clients render with their own widgets. a2tea targets v0.9.

Bubble Tea · charm.land

The Elm Architecture, in Go: a model, an update function, a view. Styled with Lip Gloss. a2tea renders A2UI surfaces as embeddable tea.Model values.