Amazon has introduced the AG-UI (Agent-User Interaction) protocol to enhance the user experience for AI agents built on Amazon Bedrock AgentCore. This protocol allows agents to render interactive UI elements such as charts and canvases directly within conversations, while also enabling shared state and human-in-the-loop interactions. AG-UI works with multiple agent frameworks and frontend libraries, ensuring that backend and frontend code remain decoupled. The protocol is part of the Fullstack AgentCore Solution Template (FAST), which provides a ready-to-deploy starter project for building interactive agent frontends on Amazon Bedrock AgentCore.
The AG-UI protocol integrates with Amazon Bedrock AgentCore Runtime to handle authentication, session isolation, and scaling. It supports multiple agent protocols, including the Model Context Protocol (MCP) and Agent2Agent (A2A). When an agent container is deployed with the AG-UI protocol flag, AgentCore acts as a transparent proxy, passing requests unchanged while managing security and observability. The protocol also includes a frontend parser that handles both Strands and LangGraph backends, simplifying the development process.
The solution includes two layers: AG-UI in FAST provides two new agent patterns and a single frontend parser, while CopilotKit + FAST is a standalone sample that replaces the built-in chat UI with generative UI features. Both layers deploy on AgentCore Runtime with Cognito authentication, AgentCore Gateway for MCP tool connectivity, and AgentCore Memory for persistent conversations. The frontend communicates with AgentCore Runtime through AG-UI events, and the agent runtime translates framework-specific events into the AG-UI protocol.
Source: awsml
