
A2UI is an open standard and set of libraries that lets AI agents “speak UI.” Instead of emitting HTML or running code, agents stream declarative JSON messages describing layouts, components, and data bindings. Client renderers (web, mobile, desktop) turn those messages into native, brand-safe interfaces—so assistants can present forms, tables, charts, and multi-step flows securely across trust boundaries. The project is Apache-licensed, community-driven, and currently in public preview (v0.8) with quickstart demos.
Agents describe intent as data: structure, components, props, and bindings arrive as JSON messages that the host app renders with its own trusted component library. No arbitrary code executes from the agent, reducing XSS/RCE risk and keeping styling consistent. Because UIs are data, they can be diffed, replayed, logged, versioned, and audited—useful when assistants operate across accounts, tenants, or regulated workflows where provenance matters as much as speed.
Client libraries render the same A2UI JSON into native surfaces for the web (e.g., Angular/Lit), mobile (Flutter), and desktop. Hosts control theming and accessibility while agents focus on task logic. This separation lets one agent interface travel anywhere your product does—support chat, sidebars, and full-screen apps—without re-authoring UI code. Teams ship consistent, brand-aligned experiences for many platforms from a single protocol instead of bespoke UI per client.
A2UI streams UI changes over time, so assistants can add steps, refine forms, and update results live as plans evolve. A distinct data model enables reactive updates: inputs flow back to the agent; state changes re-render without full refreshes. Developers get predictable lifecycles for mount/update/unmount and clear contracts for events. The result is conversational software with the responsiveness of native apps—great for iterative tools like planners, search, and configurators.
Because agents send data, not code, hosts enforce permissions, sandboxing, and design rules. Components whitelist allowed inputs; transports control who can speak; logs preserve every message for review. This model helps enterprises adopt agentic UX without surrendering brand or security posture. It also simplifies threat modeling across third-party or remote agents, letting security teams reason about surfaces as structured messages instead of opaque script blobs.
Released under Apache-2.0, A2UI invites community renderers, transports, and components. Quickstart demos (e.g., a restaurant finder) get teams hands-on in minutes, while docs cover concepts, lifecycles, theming, and custom widgets. Active GitHub issues/discussions track naming, schema polish, and new client libraries. The outcome is a shared foundation others can extend—avoiding duplicative “agent UI” efforts and accelerating interoperable tools across vendors and stacks.


Teams building agentic products: platform and frontend engineers who want secure, brand-controlled interfaces for assistants; enterprise developers standardizing multi-agent UX; researchers prototyping tools that must run on web, mobile, and desktop without bespoke UI code. Also a fit for startups that need faster iteration on assistant surfaces while keeping governance, accessibility, and theming in the host app.
Replaces ad-hoc iframe widgets and unsafe code generation with a portable, reviewable protocol. It solves three pains: security (no arbitrary code execution), consistency (host renders with native components and brand styles), and portability (one JSON format across platforms). With streaming updates, data binding, and open-source renderers, teams move from text-only chats to rich, auditable interfaces—scaling agent UX without rebuilding the same UI scaffolding for every client.
Visit their website to learn more about our product.

Grammarly is an AI-powered writing assistant that helps improve grammar, spelling, punctuation, and style in text.

Notion is an all-in-one workspace and AI-powered note-taking app that helps users create, manage, and collaborate on various types of content.
0 Opinions & Reviews