As part of today's new wave of Jetpack artifacts, we got androidx.a2ui and androidx.a2ui.compose as new artifact groups. These appear to contain a 1.0.0-alpha01 edition of Google's official A2UI engine in Compose.

A2UI is Google's own specification for agent-defined UI. Server-defined UI is where a backend system sends a description of a UI that a client renders. Agent-defined UI is the same thing, except that the backend system happens to be an LLM.

The overall objective of A2UI is to provide a means for an agent powered by an LLM to interact with users in a richer form than simple text chat and without locking the LLM into some specific set of screens/modals/whatever. Just as an LLM can invoke a tool by providing a response that contains the JSON for a tool call, an LLM can request a GUI interface by providing a response that contains A2UI.

A2UI itself is a JSON Lines (JSONL) document. If you are unfamiliar with JSON Lines, it is a text file, with each line (newline-delimited) representing a JSON object. That document describes what the UI should look like, both initially and after some updates (e.g., in response to a user interaction).

I have been following A2UI for some time. Beyond its intended use by agents, A2UI is one of the few specifications for general-purpose server-defined UI. There is nothing stopping somebody from using A2UI as a server-defined UI approach, with no LLM in the mix.

In addition to the artifacts in those two new a2ui groups, androidx.compose.material3:material3-a2ui showed up. My assumption is that the two a2ui groups will be a design-system-agnostic framework for implementing A2UI support in Compose, and Compose Material3 A2UI will plug into that framework, offering an out-of-the-box Material3 implementation.

Hopefully, Google will post something soon about their new A2UI for Compose engine. And, with luck, JetBrains will explicitly state their plans regarding A2UI for Compose Multiplatform, whether those plans are to support it or to delegate it to the community.