agui

Real-time streaming AI agents use Server-Sent Events (SSE) to push assistant responses from backend to frontend as they generate. AG-UI provides a protocol and reference implementation for bidirectional communication, tool execution, and session management across web clients, enabling live agent interactions without polling.

AG-UI is a protocol for web-based AI agents that supports streaming responses, backend and frontend tool execution, and human-in-the-loop workflows. Use it when you need real-time agent interaction, multi-client sessions, tool invocation, and human approval loops in a web interface.

Yes. AG-UI agents support dual tool execution—backend tools for server-side operations and frontend tools for client-side actions. This enables complex workflows like human approvals, session management, and distributed processing across your application stack.

An AG-UI server requires SSE streaming capability, tool invocation support, and session management. You also need a frontend client to render responses. AG-UI provides minimal quick-start examples and reference architectures to integrate with common frameworks.

Human-in-the-loop workflows in AG-UI agents involve streaming responses to a frontend client, pausing for user approval or input via frontend tools, and resuming backend execution. Session management preserves state across client interactions, enabling asynchronous human decision points.

Yes. AG-UI's session management and bidirectional communication support multiple clients connecting to the same backend server. Each client receives streamed responses and can execute frontend tools independently while sharing backend state through the session.