What is AG-UI?

• Agent-User Interaction Protocol
• Open, lightweight, event-driven protocol
• Universal translator for AI-driven systems
• Standardizes frontend-to-agent communication

"Think of it as USB-C for AI agents!"

The Problem AG-UI Solves

• Agents live in backend silos
• Custom WebSocket implementations for each framework
• No standard for real-time interaction
• Fragmented agent-to-UI communication
• Human-in-the-loop workflows are complex

Core Capabilities

• ⚡ Real-time interactivity
• 📡 Live state streaming
• 🤝 Human-in-the-loop collaboration
• 📝 Token-by-token text streaming
• 🔍 Tool execution transparency
• ↔️ Bidirectional communication

Event-Driven Architecture

• 16 standardized event types
• 5 categories: Lifecycle, Text, Tool, State, Special
• JSON events over HTTP/SSE
• Observable pattern for real-time updates
• Optional binary protocol for performance

Event Categories

Lifecycle (5)

• RUN_STARTED
• RUN_FINISHED
• RUN_ERROR
• STEP_STARTED
• STEP_FINISHED

Text Message (3)

• TEXT_MESSAGE_START
• TEXT_MESSAGE_CONTENT
• TEXT_MESSAGE_END

Tool Call (3)

• TOOL_CALL_START
• TOOL_CALL_ARGS
• TOOL_CALL_END

State Management (3)

• STATE_SNAPSHOT
• STATE_DELTA
• MESSAGES_SNAPSHOT

Special (2)

• RAW
• CUSTOM

Framework Integrations

LangGraph

✅ Supported

Graph-based agent orchestration

CrewAI

✅ Supported

Multi-agent workflows

Mastra

✅ Supported

TypeScript-first agents

AG2

✅ Supported

Open-source AgentOS

Bedrock

🚧 Coming Soon

AWS managed agents

Protocol Ecosystem

MCP

Model Context Protocol

By: Anthropic

Purpose: Agent ↔ Tools communication

A2A

Agent-to-Agent Protocol

By: Google

Purpose: Agent ↔ Agent communication

AG-UI

Agent-User Interaction Protocol

By: CopilotKit

Purpose: Agent ↔ Human interaction

These protocols complement each other - not compete!

Real-World Use Cases

💻

Live Code Pairing

AI writes code token-by-token, human can interrupt and collaborate

📊

Data Analysis Dashboards

Real-time query execution with human oversight

🤖

Multi-Agent Orchestration

Human supervisors monitoring agent workflows

🎨

Creative Design Tools

AI generates designs with live previews and human feedback

Technical Benefits

• Sub-100ms latency for token streaming
• 60% smaller payloads with binary protocol
• Transport agnostic (HTTP/SSE/WebSockets)
• Framework agnostic
• State delta syncing reduces bandwidth
• Built-in security and error handling

Getting Started

📚

Documentation

docs.ag-ui.com

🧪

Live Demo

agui-demo.vercel.app

💻

GitHub

github.com/ag-ui-protocol/ag-ui

🛠️

SDKs

TypeScript and Python

Start building human-AI collaborative apps today!