A2UI

A2UI is an open-source protocol and set of libraries designed to help AI agents "speak UI." While generative AI excels at producing text and code, agents often struggle to present rich, interactive interfaces safely. A2UI bridges this gap by enabling agents to generate native-quality interfaces that work across web, mobile, and desktop environments without the security risks of executing arbitrary code.

Currently in public preview (v0.8), A2UI provides a standardized way for agents to send declarative UI descriptions. Instead of rendering a static block of text, your application can now display interactive forms, charts, and maps that feel like a seamless part of your product’s native experience.

Key Features

• 🛡️ Secure-by-Design Architecture: A2UI uses a declarative data format rather than executable code. Because agents can only trigger pre-approved components from your specific catalog, you eliminate the risk of UI injection attacks and maintain total control over your application's security perimeter.
• 🤖 LLM-Optimized Structure: The protocol uses a flat, streaming JSON structure specifically designed for how Large Language Models (LLMs) process information. This allows agents to build complex interfaces incrementally, ensuring the system remains robust even if the model doesn't generate a perfect, massive payload in a single pass.
• 📱 Framework-Agnostic Portability: You can write your agent logic once and render the output anywhere. Whether your frontend is built on Flutter, Angular, React, or native mobile components, A2UI maps abstract agent descriptions to your specific native widgets, maintaining visual consistency across all platforms.
• ⚡ Progressive Real-Time Rendering: Users don't have to wait for a complete response to see results. A2UI supports streaming updates, allowing the interface to build and populate in real-time as the agent generates data. This creates a highly responsive, "live" feel that mirrors the speed of natural conversation.

Use Cases

• Dynamic Contextual Data Collection: An AI travel assistant can generate a bespoke booking form with date pickers, seat maps, and price sliders on the fly, specifically tailored to the destination and preferences discussed in the chat.
• Adaptive Enterprise Dashboards: A business intelligence agent can respond to a query like "show me last quarter's churn" by instantly rendering an interactive bar chart and a filtered data table, rather than just describing the numbers in text.
• Remote Sub-Agent Integration: A primary orchestrator agent can call upon a specialized "billing agent" which then returns its own secure UI payload. This allows complex, multi-agent workflows to be displayed within a single, unified user interface.

Why Choose A2UI?

A2UI solves the fundamental tension between UI expressiveness and system security. Traditionally, developers had to choose between safe but limited text-based responses or powerful but dangerous code execution (like generating and running JavaScript).

A2UI introduces a third way: the "Universal UI Language." By decoupling the intent of the UI (the agent's job) from the implementation of the UI (the client's job), you gain the flexibility of generative interfaces with the safety of a managed component library. This architecture ensures that even as AI models become more autonomous, your application remains stable, performant, and secure.

A2UI represents a significant step toward a future where user interfaces are as fluid and intelligent as the agents driving them. By adopting a declarative, component-based protocol, you can empower your AI agents to provide rich, actionable experiences that go far beyond the chat bubble. As the project moves toward a v1.0 specification, it offers a robust foundation for anyone building the next generation of agentic software.