Discover and explore top open-source AI tools and projects—updated daily.
googleAgent-native UI generation and rendering framework
Top 5.3% on SourcePulse
A2UI provides an open standard and libraries for agents to generate and populate rich, updateable user interfaces. It addresses the challenge of presenting complex, interactive UIs from generative AI agents, especially in remote or cross-trust scenarios. The project benefits developers building agent-based applications by enabling safe, expressive, and interoperable UIs across different platforms and frameworks.
How It Works
A2UI employs a declarative JSON format where agents describe UI intent rather than executing code. Client applications maintain a catalog of trusted, pre-approved UI components (e.g., Card, Button) and render the agent's JSON description using these native components. This approach ensures security by design, as agents can only request rendering of known components. The UI representation is a flat list of components with ID references, making it LLM-friendly for generation and incremental updates, facilitating progressive rendering and responsive user experiences.
Quick Start & Requirements
To get started, clone the repository (https://github.com/google/A2UI.git), set up your Gemini API Key, and run the provided agent and client samples.
Highlighted Details
Maintenance & Community
A2UI is currently in an "Early Stage Public Preview" (v0.8), with the specification and implementations evolving. The project is open for collaboration, feedback, and contributions, with details available in CONTRIBUTING.md.
Licensing & Compatibility
The project is licensed under the Apache 2.0 license. Its framework-agnostic design promotes broad compatibility, though developers are responsible for implementing secure rendering logic within their host applications.
Limitations & Caveats
As an early-stage preview, the specification and implementations are subject to change. A host application built in a supported framework (currently Web or Flutter) is required. Additional renderers (React, SwiftUI) and transports are planned but depend on community contributions.
2 days ago
Inactive
MCP-UI-Org