Save your spot

Live eventOpenBot for Business

Copilot Kit Logo
  • Docs
  • Blog
  • AG-UI

    AG-UI (Agent–User Interaction) is the general-purpose, bi-directional connection between a user-facing application and any agentic backend.

  • Pricing
Get StartedTalk to an Engineer

Ship A2UI with the team that helped build it

We created AG-UI, the protocol A2UI 1.0 names as a standard transport, and co-launched A2UI with Google. Book 25 minutes with an engineer who built it.

  • Creators of AG-UI

    The two-way connection between agents and apps

  • Co-launched A2UI with Google

    At launch and again for 1.0

  • Built the A2UI Playground

    And collaborated on the A2UI Composer

  • Teach it on DeepLearning.AI

    A2UI, controlled components and MCP Apps

Loading the booking calendar...

Free on Google Meet · includes $400 in credits + 1k MAU for 3 monthsOpen in Calendly ↗

AG-UI, the protocol we created, runs across

Our role in the ecosystem

We didn't just read the spec. We helped write the stack around it.

A2UI describes what an agent wants to show. Something still has to carry it to your app, render it safely, and send what the user does back. That part is what we build.

transport: AG-UI

We created AG-UI, and A2UI 1.0 names it in the spec

AG-UI is the open, event-based protocol between any agent backend and a user-facing app. It carries messages, state, tools and generative UI. The A2UI 1.0 spec lists AG-UI as a standard transport, alongside A2A and MCP.

Read about AG-UI 1.0 →

A2UI 1.0 standard transports

A2A
agent to agent
MCP
model context
AG-UI
agent to user · created by CopilotKit

Co-launched A2UI with Google

We worked with Google on A2UI from the start and shipped first-class A2UI rendering in CopilotKit for the launch and again for 1.0, including two-way function calls.

What's new in A2UI 1.0

Built the A2UI Playground

One research agent, many interfaces. Ask a question and the answer comes back as a picture grid, a report, a live calculator or a chart, all built from A2UI 1.0.

Open the Playground

Collaborated on the A2UI Composer

A gallery of agent-generated A2UI components you can open and edit in the Composer. It's the fastest way to see what a catalog can do before you write one.

Open the Composer

We teach it on DeepLearning.AI

Atai Barkai, CopilotKit's CEO, teaches Build Interactive Agents with Generative UI on DeepLearning.AI. A2UI is one of the three approaches in the course.

Take the free course

What we help with

From first surface to production

Bring the part you're stuck on. These are the questions teams usually show up with.

  1. 01

    Picking the right approach

    A2UI, controlled components, or MCP Apps? We'll map your use case to the right spot on the generative UI spectrum, or a mix of them.

  2. 02

    Designing your catalog

    Which components your app trusts, how they nest with allowedChildren, and the instructions that teach the model to use them well.

  3. 03

    Wiring your agent

    ADK, LangGraph, CrewAI, A2A or your own backend. We'll get A2UI messages streaming to your app over AG-UI.

  4. 04

    Rendering in your app

    Render A2UI in React with CopilotKit, keep it on brand, and bind it to your data with @path and @call.

  5. 05

    Closing the loop

    User actions and two-way function calls back to the agent, so the UI is something people act in, not just look at.

  6. 06

    Getting to production

    Accessibility, guardrails at the agent boundary, and debugging the traces that only show up with real users.

A2UI in 60 seconds

A2UI describes it. AG-UI carries it. CopilotKit renders it.

A2UI is a declarative spec from Google. The agent picks components from a catalog your app trusts, so the result is safe to render on web, mobile and desktop. It sits in the middle of the generative UI spectrum.

Controlled

Pre-built components

You build each component and wire it up. Most control, least flexible.

Declarative · A2UI 1.0

The agent describes it

Your catalog renders it. Flexible layouts, with components you already trust.

Open · MCP Apps

Free-form surfaces

Sandboxed UI the agent fully controls. Most flexible, least on-brand.

FixedFree-form
Inside the A2UI Playground ↗ · 4:42
Full walkthrough of A2UI 1.0

How the call works

25 minutes with someone who's shipped it

You'll talk to a CopilotKit engineer, not a sales rep. We read what you send ahead of time and route you to the right person for your stack.

Bring

  • Your repo, a branch, or a screen to share
  • The error, trace or surface that's misbehaving
  • Your stack: agent framework, model, frontend

0 to 5

What you're building

Your product, your agent, and where generative UI fits.

5 to 20

Architecture, code, debugging

Catalog design, AG-UI wiring, rendering, or the bug you brought. Screen-share welcome.

20 to 25

Next steps

A plan for what to build next, and who to reach when you need more help.

Pick a time

Prefer to start on your own?

Everything we'd point you to on the call

Live demoA2UI PlaygroundAsk a question and watch an agent answer with A2UI 1.0 interfaces.Open the Playground ↗DocsA2UI in CopilotKitRender A2UI 1.0 in your React app with your own catalog.Read the docs ↗SpecA2UI 1.0 specGoogle's declarative spec for agent-generated interfaces.Read the spec ↗GalleryA2UI ComposerBrowse agent-generated components and edit any of them.Open it ↗

Talk to the engineers who built it

Free 25-minute call, plus $400 in CopilotKit credits and 1k MAU for 3 months.

Book a call