ag-ui-protocol/ag-ui

AG-UI is an open protocol that standardizes how AI agents connect to the apps that real people use. An AI agent here means a program that runs tasks on its own using a language model. The problem AG-UI addresses is the gap between that agent running on a backend server and the screen a user is looking at. Without a shared standard, every team has to invent its own way of streaming the agent's progress, updates, and requests into a chat window or other interface. AG-UI defines a common, event-based language so that work does not have to be redone for each project. The way it works is described in plain terms. While an agent runs, its backend sends out a series of events, and AG-UI defines around sixteen standard event types that those events can match. The agent can also accept a small set of simple inputs. Between the agent and the app sits a middleware layer that smooths over differences, so the same protocol works whether the events travel over Server-Sent Events, WebSockets, webhooks, or other transport methods. The project ships a reference HTTP implementation and a default connector so teams can start quickly. The README places AG-UI alongside two other agent protocols to show where it fits. One, called MCP, gives agents access to tools. Another, called A2A, lets agents talk to other agents. AG-UI's distinct role is bringing agents into user-facing applications. You can create a starter project with a single command shown in the README. The listed features include real-time chat with streaming replies, keeping state in sync in both directions between agent and interface, generating interface elements from the agent, adding live context, letting the frontend offer tools to the agent, and supporting human-in-the-loop steps where a person reviews or guides the agent. A large table lists the many agent frameworks and platforms that already work with AG-UI, including LangGraph, CrewAI, and several first-party and community integrations.