ag-ui-protocol/ag-ui

AG-UI: the Agent-User Interaction Protocol. Bring Agents into Frontend Applications.

Summary

AG-UI is an open, event-based protocol that standardizes how AI agents connect to user-facing applications. It defines roughly 16 standard event types, accepts a small set of compatible inputs on the agent side, and ships a middleware layer that works over transports like SSE, WebSockets, or webhooks. The project complements MCP and A2A by focusing specifically on the agent-to-UI layer of the protocol stack.

What it is useful for

AG-UI fits when an engineer is wiring an existing agent backend into a chat surface, a dashboard, or an in-app assistant that has to stay in sync while the agent runs. The README lists concrete capabilities: real-time streaming chat, bi-directional state synchronization between agent and frontend, generative UI where structured agent messages render as components, real-time context enrichment, frontend-registered tools, and human-in-the-loop collaboration where a user can interrupt or correct a running agent.

It also helps teams that want to keep their current agent framework. The README documents support for LangChain, CrewAI, Microsoft Agent Framework, Google ADK, Google Antigravity, AWS Strands, Mastra, Pydantic AI, Agno, LlamaIndex, and AG2. Related protocol work, including A2A, MCP Apps, and the Oracle Agent Spec, is also listed as supported.

How engineers can use it

The documented quickstart is a single command: npx create-ag-ui-app my-agent-app. Engineers then pair an agent framework with a frontend client. Documented clients include CopilotKit, a Terminal + Agent client, the Channels SDK for Slack and Microsoft Teams, and a React Native example.

For teams building a custom integration, the repository ships a reference HTTP implementation and a default connector. Community SDKs are available in Kotlin, Go, Dart, Java, Rust, Ruby, C++, and .NET. The README marks AWS Bedrock Agents, OpenAI Agent SDK, Cloudflare Agents, Nim, Flowise, and Langflow as "In Progress" rather than fully supported, so those paths require checking the repository's current state.

The README points to docs.ag-ui.com for the full event-type specification and transport details, and to the AG-UI Dojo for runnable examples that the project describes as 50-200 lines each. The README does not include the exact event schemas or SDK APIs, so engineers planning an integration should consult the docs and the AG-UI Dojo source before writing production code.