Join waitlist
Back to products
AG-UI

AG-UI

Open protocol for interactive Agent-to-UI communication

Website ag-ui.com
Overview

What it is

AG-UI is an open, lightweight protocol standardizing AI agent to front-end app connections. Enables real-time interactivity, streaming, and human-in-the-loop workflows. Supports LangGraph, CrewAI, etc.

Intent

I need it when

Build agentic applications that work with multiple agent frameworks and protocols seamlessly

AG-UI integrates with LangGraph, CrewAI, Microsoft Agent Framework, Google ADK, AWS Bedrock, Pydantic AI, and others. It also handshakes with MCP (Model Context Protocol) and A2A (Agent-to-Agent) protocols, enabling universal compatibility across the agentic ecosystem.

Implement generative UI where agents propose UI components and applications validate and render them

AG-UI supports both static and declarative generative UI rendering. Agents can propose typed UI components or declarative trees with constraints; the application validates and mounts them, giving developers full control over UI presentation.

Connect a user-facing application to AI agent backends with standardized, event-driven communication

AG-UI provides an open, lightweight, event-based protocol that standardizes how agents connect to applications. It enables bidirectional, real-time interaction between frontends and agentic backends, allowing developers to ship reliable, debuggable agent features without complex ad-hoc wiring.

Handle long-running, streaming agent operations with intermediate state visibility and user interrupts

AG-UI's event-driven architecture supports streaming chat, multimodal inputs, state management (snapshots and deltas), and human-in-the-loop interrupts. Developers can pause, approve, edit, or retry agent execution mid-flow while maintaining full state continuity.

Debug and trace agent execution with visibility into reasoning, tool calls, and state changes

AG-UI provides structured events for lifecycle tracking, tool call streaming, reasoning steps, and state deltas. Developers can visualize intermediate reasoning from traces and tool events, enabling transparent debugging and monitoring of agent behavior.

Drop

Not a fit when

  • Building simple request-response REST/GraphQL APIs without agentic features
  • Developing non-interactive applications that do not require real-time bidirectional communication
  • Creating applications that do not need to handle long-running, nondeterministic agent workflows
  • Projects requiring synchronous, stateless backend interactions only
  • Applications that do not integrate with AI agents or LLM-based backends
  • Systems that cannot support WebSocket or Server-Sent Events transport mechanisms
Commercials

Pricing

Pricing not specified