Which agent UI?

A neutral index of agent frontend libraries and protocols, minus vendor spin.

Join the waitlist Read the matrix
Show

8 of 8 entries

ToolPlays with
Protocols
LangChain, Mastra, PydanticAI, CrewAI; carries A2UI

Transport protocol: agent backend to frontend with real-time state sync (by CopilotKit). Show it in the stack

Web and Flutter renderers; A2A and AG-UI transports

Declarative JSON UI format (Apache 2.0). The agent can only request components from your trusted catalog. Show it in the stack

MCP hosts

MCP extension (SEP-1865). Servers ship pre-built HTML via ui://, rendered in sandboxed iframes. Show it in the stack
Libraries
React, Angular, Vue, React Native, Slack, Teams

Agent frontend SDK for chat, generative UI, shared state, human-in-the-loop. Show it in the stack

LangGraph data UI; consumes A2UI over AG-UI

Composable React primitives. Generative-UI package with a 35-component default vocabulary. Show it in the stack

OpenAI agents

OpenAI’s chat UI kit. Tightly integrated with the OpenAI ecosystem. Show it in the stack

React, Vue, Svelte, Angular; any AI SDK provider

AI SDK UI hooks (useChat) plus AI Elements components. Provider-agnostic streaming chat and tool calls. Show it in the stack

NewReact, Solid, Vue, Svelte, Preact, Angular; AG-UI events

New: 1.0 shipped Oct 2026. Provider-agnostic TypeScript toolkit: agent loop, tools, framework chat hooks. Show it in the stack

Sources: project docs and repos, checked Oct 2026.

Why this exists

Agent backends multiply every month. Which frontend stack and protocol wire into them without buying a vendor’s worldview?

One matrix, one set of demos, one changelog.

Independent. Built by a frontend engineer who builds AI agent platforms. Not affiliated with any listed project.

How the stack layers

Focus or hover a box to trace what it connects to.

Diagram of three layers. UI libraries: assistant-ui, CopilotKit, TanStack AI, Vercel AI SDK, ChatKit. Formats and transport: A2UI, AG-UI, MCP Apps. Agent backends: LangGraph, Mastra, CrewAI, PydanticAI, MCP hosts, OpenAI Agents. The connections are listed below the diagram. assistant-ui CopilotKit TanStack AI Vercel AI SDK ChatKit A2UI AG-UI MCP Apps LangGraph Mastra CrewAI PydanticAI MCP hosts OpenAI Agents

assistant-ui consumes A2UI over AG-UI.

AG-UI is by CopilotKit and talks to LangGraph-class backends.

ChatKit sits on OpenAI Agents.

MCP Apps speak to MCP hosts.

What ships at launch

  1. Table

    Compatibility matrix

    Which UI library works with which agent backend and protocol. Updated when the docs change.

  2. Demos

    Same agent, every UI

    One reference agent wired into each library, side by side, so the comparison is visual, not marketing copy.

  3. Email

    Release changelog

    Breaking changes and new specs, tracked in one place. The feed behind the weekly email.

How the pieces fit

Four layers, kept distinct in the matrix.

Pipe

AG-UI

Agent backend to frontend transport, with real-time state sync.

Payload

A2UI

Declarative JSON UI; components only from your trusted catalog.

Resource

MCP Apps

Server-owned HTML via ui:// in sandboxed iframes.

UI kits

Five libraries

How you actually paint the surface.

What changed in agent UI

A sample of the feed behind the weekly email.

  1. A2UI

    v0.9 schema rewrite; bidirectional with AG-UI and A2A.

  2. OpenUI

    OUI-1 model aimed at generative UI composition.

  3. CopilotKit

    AG-UI client for React Native (generative UI, threads, human-in-the-loop).

Get on the list

You’ll also get “What changed in agent UI”: one short email a week on new releases, spec changes and breaking changes. Unsubscribe anytime.