A2UI

Build delightful, fast, reliable agentic experiences

A2UI (Agent-to-User Interface) is an open-source generative UI framework, complete with a protocol optimized for streaming dynamic user interfaces at low latency, native renderers for web and mobile frameworks, and agent SDKs in several popular languages. With A2UI, your agents can communicate with your users using natively rendered widgets, in addition to text.

Examples of UI compositions that A2UI can express.

Quick links

Main website.
Try it in your browser using A2UI Composer.
Supported renderers.
Supported agents.
Protocol specification.

Introduction

Large language models (LLMs) and decision models (e.g. Jev) excel at creating text and code and at making fast decisions on top of raw structured data (e.g. JSON). A2UI gives these models a specialized vocabulary for expressing user interfaces.

A2UI expresses UI using data, not code. A schema for the data exchanged between the agent and the renderer enforces strict capability boundaries for the agent, keeping agent-generated experiences safe and aligned with the business requirements of your application. A core library, a native framework adapter, and a catalog of components provide access to native components and form the building blocks that the agent composes into a final user experience, making the result expressive and high-performance.

When to use A2UI

Use A2UI when any of the following apply to your application:

Parts of your application are agent- and/or server-driven. By design, A2UI supports remote rendering.
You want the UI to be composed dynamically, tailored to the user's immediate need. Instead of making your users navigate through multiple screens to solve their problems, have the agent build the right flow of UI that guides users directly to solutions.
You need the highest performance and visual fidelity. A2UI renders UI using the framework of your choice, typically one that the host app itself is written in. WebViews or HTML are not required.
The same agent streams UI to multiple platforms or frameworks, such as Flutter, React, SwiftUI, and Jetpack Compose. A2UI helps keep the user experience compatible and consistent across all screens.
The agent is remote, third-party, or operates across a trust boundary. For example, the agent must not be able to handle certain kinds of data (credit card/account numbers, PII, financial data). A2UI helps enforce the trust boundary and protect your app's critical functionality from hallucinations using a strict schema, and by expressing UI using validated data rather than executable code.

A2UI core philosophy

We develop A2UI under the guidance of the following core principles:

Security first: The app developer controls what the agent can and cannot do in the app. A2UI provides a framework for specifying and enforcing the necessary guardrails, such as expressing UI as data, validating inputs and outputs against a schema, and disallowing arbitrary executable code. Every capability granted to the agent is either a safe standard capability built into the framework, or a capability explicitly granted by the app developer.
Model-friendly and server-driven: The A2UI protocol is designed to be efficiently generated by language and decision models and transported over the network, and to support progressive rendering (or "streaming") for a responsive user experience. An agent can adapt the UI based on where the user takes the conversation. A2UI supports swappable inference formats, so you can tailor inference to specific models and use cases.
Native rendering and portability: A2UI renders UI using the framework native to your app, with renderers for many major UI frameworks. No need to change frameworks unless required for other reasons.

Architecture

A2UI's architecture is made of multiple layers spanning the full stack that connects the agent to the application:

Inference: An agent (typically backed by a language or a decision model) instructed by an A2UI skill (created using an A2UI Agent SDK) generates the description of UI and data using an appropriate inference format (e.g. A2UI Express, or a decision tree).
Transport: An A2UI Agent SDK, embedded in a larger agent architecture, parses and validates the model's output, expands macros, and converts the UI to the wire-stable A2UI Transport protocol messages, which are sent to the renderer.
Rendering: A renderer is made of an A2UI Core library (e.g. the @a2ui/web_core npm package, or a2ui_core Dart package), a UI framework adapter (e.g. @a2ui/react React adapter, or a2ui_flutter Flutter adapter), and a widget catalog. The core library parses the incoming messages and creates a framework-agnostic in-memory representation of the UI and data. The framework adapter maps the in-memory representation to concrete framework-specific UI widgets. The widget catalog provides access to the widgets the agent is allowed to compose.

Getting started

Pick the path that matches where you want to start:

Path	What you get	Time
🍜 Quickstart	A full-stack A2UI app with a backend and frontend. Learn A2UI end-to-end and customize to your use case.	~5 min
🎨 A2UI Composer	Generate A2UI JSON from a visual editor and paste it into any agent prompt — no install required.	~1 min
🎬 A2UI Theater	Step through pre-built A2UI streaming scenarios across Lit, React, and Angular renderers — no install required.	~1 min

Packages

These packages are developed in this repository. Packages marked "Not published yet" can be used from source.

Python

Package	Description	Version
a2ui-core	Protocol messages, data model, data binding, expression evaluation, and catalog validation.
a2ui-agent-sdk	Agent SDK: negotiates catalogs, generates prompts and skills, and compiles model output into A2UI.
a2ui-a2a	Connects the agent SDK to the Agent-to-Agent (A2A) protocol.
a2ui-adk	Connects the agent SDK to the Agent Development Kit (ADK).

TypeScript

Package	Description	Version
@a2ui/web_core	Protocol messages, state management, expressions, and basic catalog shared by renderers and `@a2ui/agent`.
@a2ui/lit	Lit (web components) renderer.
@a2ui/angular	Angular renderer.
@a2ui/react	React renderer.
@a2ui/markdown-it	Markdown rendering with markdown-it and DOMPurify, shared by the web renderers.
@a2ui/agent	Agent SDK: negotiates catalogs, generates prompts and skills, and compiles model output into A2UI.
@a2ui/catalog-mcp	MCP catalog: calls MCP tools and renders A2UI messages returned in tool results and UI resources.
@a2ui/catalog-vegachart	VegaChart catalog: renders Vega and Vega-Lite chart specifications on A2UI surfaces.	Not published yet

Dart and Flutter

Package	Description	Version
a2ui_core	Protocol messages, data model, data binding, and expressions for renderers and agents.
a2ui_agent	Agent SDK: negotiates catalogs, generates prompts and skills, and compiles model output into A2UI.
a2ui_flutter	Flutter renderer.
a2ui_cli	Developer tool that generates typesafe component code from a catalog.	GitHub Releases

Swift

The Swift libraries are distributed via Swift Package Manager from Package.swift at the root of this repository.

Package	Description
A2UISwiftCore, BasicCatalog	Message processor, data model, expression evaluation, and basic catalog definitions.
A2UISwiftUI, BasicCatalogSwiftUI	SwiftUI renderer and basic catalog views.

Roadmap

Protocol v1.1: We are collecting requirements for the next iteration of the A2UI protocol.
Inference: Support for decision models (e.g. Jev).
Templates: A cross-framework templating engine and a set of standard primitives for rendering reusable rich UI widgets.
Sandboxed logic: A way to securely attach logic to A2UI components.
Agent SDK: Skill generation, better diagnostics, and integrations with more agent frameworks.
Evals: A set of tools for evaluating generated UI quality.
MCP Apps: Embedding MCP Apps inside A2UI, and embedding A2UI experiences into MCP Apps.
CLI: Schema validation, codegen, local preview, and integration with coding agents.

Contribute

A2UI is an Apache 2.0 licensed project. We believe the future of UI is agentic, and we want to work with you to help build it.

See CONTRIBUTING.md for details on how to get started.