interactionactive
AG-UI
An event-based protocol for connecting agent runtimes to user-facing applications with streaming, state, tools, and human interaction.
Independent · Source-verified · Open
Choose the right layer for your Generative UI stack.
A field guide to the protocols, declarative specifications, renderers and host surfaces behind interfaces that agents can shape at runtime.
LIVE EVENT TRACERUN_8F2A
01
RUN_STARTEDagent → client02
TEXT_MESSAGE_CONTENTstreaming03
TOOL_CALL_STARTrender_cardGENERATED SURFACEArchitecture recommendation ready
04
USER_ACTIONapprovedEcosystem / 01
Explore the ecosystem →One ecosystem. Four architectural questions.
These technologies compose across layers. Compare their responsibilities before comparing their APIs.
01
Agent ↔ frontend interaction
How do runs, messages, tools and state move?
02
Declarative UI & rendering
How is UI intent constrained and rendered?
04
Implementation & integration
What ships the product experience?
CopilotKitAgent frameworksmany implementations
Profiles / 02
Compare all technologies →Core technology profiles
Seven source-backed guides using the same decision framework—without a winner-takes-all ranking.
declarativepublic preview
A2UI
A declarative, portable UI format that lets agents describe updatable interfaces rendered through trusted client-side component catalogs.
declarativeactive
OpenUI
A streaming-first Generative UI framework built around a compact language, React runtime, component libraries, prompt generation, and chat surfaces.
declarativeactive
json-render
A Generative UI framework where models produce predictable JSON specifications constrained by developer-defined components and actions.
declarativeemerging
Open-JSON-UI
An emerging declarative JSON UI concept documented by CopilotKit, without a separately verified first-party OpenAI specification repository.
embeddedstable
MCP Apps
An official MCP extension for serving sandboxed interactive HTML interfaces from MCP servers into compatible conversational hosts.
implementationactive
CopilotKit
A frontend and runtime platform for agentic applications, Generative UI patterns, and integrations built around the AG-UI ecosystem.
Decision engine / 01
Compose a stack, not a winner
Four constraints produce a transparent recommendation you can audit.
Boundary to verify: json-render constrains output to registered components and AG-UI supplies the interaction stream. Catalog design and validation remain your responsibility.
Boundary to verify: If interoperability matters more than framework ergonomics, evaluate A2UI; if the UI must live inside an AI host, evaluate MCP Apps.
Learning / 03
All learning paths →From mental model to production boundary
Three focused routes for understanding the stack, making an architecture decision, and operating it safely.
01foundation · 30 min
Map the stack in 30 minutes
Build a durable mental model of transport, declarative UI, embedded surfaces, and implementation frameworks.
Start path →02intermediate · 60 minChoose and architect a Generative UI stack
Turn product constraints into an explainable combination of protocols, specifications, renderers, and host surfaces.
Start path →03advanced · 75 minSecure and observe the generated surface
Design trust boundaries, event tracing, approval gates, recovery paths, and evaluation signals for production Generative UI.
Start path →Index / 04