Skip to main content

AG-UI: Agent-User Interaction Protocol

Lightning Talk - 10 minutes

🔗 Interactive Presentation: View Live Slides

📹 Video Recording​

🎯 What is AG-UI?​

AG-UI is the Agent-User Interaction Protocol - think of it as "USB-C for AI agents!"

  • Open, lightweight, event-driven protocol
  • Universal translator for AI-driven systems
  • Standardizes frontend-to-agent communication
  • Real-time human-AI collaboration

🚨 The Problem It Solves​

Current Pain Points​

  • Agents live in backend silos - no standard UI integration
  • Custom WebSocket implementations for each framework
  • No standard for real-time interaction between humans and agents
  • Fragmented agent-to-UI communication across platforms
  • Human-in-the-loop workflows are complex to implement

⚡ Core Capabilities​

  • 🔄 Real-time interactivity - Sub-100ms latency
  • 📡 Live state streaming - Watch agents work in real-time
  • 🤝 Human-in-the-loop collaboration - Interrupt and guide agents
  • 📝 Token-by-token text streaming - See AI thinking live
  • 🔍 Tool execution transparency - Monitor what agents are doing
  • ↔️ Bidirectional communication - True conversation flow

🏗️ Event-Driven Architecture​

16 Standardized Event Types across 5 Categories:​

Lifecycle Events (5)​

  • RUN_STARTED, RUN_FINISHED, RUN_ERROR
  • STEP_STARTED, STEP_FINISHED

Text Message Events (3)​

  • TEXT_MESSAGE_START, TEXT_MESSAGE_CONTENT, TEXT_MESSAGE_END

Tool Call Events (3)​

  • TOOL_CALL_START, TOOL_CALL_ARGS, TOOL_CALL_END

State Management Events (3)​

  • STATE_SNAPSHOT, STATE_DELTA, MESSAGES_SNAPSHOT

Special Events (2)​

  • RAW, CUSTOM

Transport Options​

  • JSON events over HTTP/SSE
  • Optional binary protocol for 60% smaller payloads
  • Transport agnostic - HTTP/SSE/WebSockets

🔧 Framework Integrations​

✅ Currently Supported​

  • LangGraph - Graph-based agent orchestration
  • CrewAI - Multi-agent workflows
  • Mastra - TypeScript-first agents
  • AG2 - Open-source AgentOS

🚧 Coming Soon​

  • Bedrock - AWS managed agents

🌐 Protocol Ecosystem​

AG-UI complements (doesn't compete with) other protocols:

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 work together to create a complete AI ecosystem!

🎯 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 - works with any agent system
  • State delta syncing reduces bandwidth
  • Built-in security and error handling

🛠️ Getting Started​

Resources​

Quick Start​

Start building human-AI collaborative apps today with standardized real-time interaction!

🔗 Related Content​

AG-UI represents the missing piece in the AI protocol ecosystem - standardized human-agent interaction that enables true collaborative AI applications.