GenUI: Native SwiftUI Interfaces Generated by AI Agents
GenUI validates AI-agent-generated A2UI messages against an allowed catalog and renders them as native SwiftUI interfaces without executing agent code.
GenUI is an experimental Swift workspace that renders AI-agent-generated interfaces natively with SwiftUI. Agents emit declarative A2UI messages, clients validate them against an allowed component catalog, and SwiftUI draws the result without ever executing agent-supplied code—an architecture choice that matters for safety and predictability in LLM-driven UI generation.
The project follows a multi-repo structure, with the root repository serving as the architecture and workflow hub while iOS/macOS apps, an offline Playground component gallery, shared design-system packages, a cloud agent client SDK, a versioned protocol definition, and a provider-agnostic Cloudflare-based agent runtime are wired in as Git submodules. Optional desktop runtimes support Claude and Codex agent SDKs as well.
The project remains experimental: the local renderer, Playground, and native app builds work, but the hosted gateway still lacks production-grade authentication, rate limiting, and session-resume guarantees, and open-source licensing and release tagging are still in progress. For engineers, the notable design decision is decoupling agent output from executable code entirely, connecting it to native UI through a declarative, validatable protocol instead.