A2UI: A Protocol for Agent-Driven Interfaces (a2ui.org)

Google’s A2UI is an open-source UI toolkit that lets LLM-driven agents send interactive UI components (not just text) using a high-level, framework-agnostic format. Built on the Agent2Agent (A2A) protocol, A2UI transmits component descriptions as streamable JSON Lines (JSONL), enabling progressive rendering and real-time UI updates across surfaces. Initial reference clients include JavaScript Web Components and Flutter, but the format is intended to be portable so the same agent output can be rendered natively on web, mobile, and other platforms. This shifts UI development toward dynamic, AI-driven interfaces: LLMs can adapt layouts, controls, and flows at runtime, speeding prototyping and delivering personalized experiences. Key technical properties include a secure, performant transport (A2A), and a deliberate separation of structure and data to reduce UI-injection risks. For the AI/ML community, A2UI means new opportunities for model-driven UX, streaming generation, and cross-platform rendering, while also raising implementation considerations—validation, permissioning, and guardrails—to keep generated interfaces safe and deterministic in production.