Agent User Interaction Protocol

AG-UI (Agent–User Interaction Protocol) is an open, event-based standard that defines how AI agents communicate with user-facing applications. Instead of building custom, tangled connections between an agent runtime and a frontend, you plug both into AG-UI and let the protocol handle streaming data, state updates, and user interactions. Think of it as a universal socket for agentic apps — it replaces the old request/response model with a bi-directional, real-time channel that matches how agents actually work (long-running, nondeterministic, mixing text, voice, tool calls, and UI widgets).

The protocol runs over HTTP and WebSockets, with building blocks that cover the messy parts of agentic UX. You get streaming chat with token-by-token output, multimodal support (files, images, audio), generative UI widgets that the agent can propose and your app can safely render, a shared typed state store synced via event-sourced diffs, and human-in-the-loop interrupts for approvals or edits. AG-UI also supports sub-agent composition, tool output streaming, and custom events. It's designed to be framework-agnostic — integrations exist for LangGraph, CrewAI, Microsoft Agent Framework, Google ADK, AWS Bedrock, Mastra, Pydantic AI, LlamaIndex, and more.

This is a tool for developers building applications with AI agents — especially those shipping co-pilots, customer support bots, or any interface where an agent needs to show intermediate thinking, ask for confirmation, or hand off control to the user. It saves you from reinventing the wheel every time you want an agent to stream text, show a chart, or pause for approval. If you're wiring an agent backend to a frontend and finding it awkward, AG-UI is the clean abstraction you've been missing.