fahmidme/agentcore-agui-starter

AgentCore AG-UI Starter is a reference project that shows how to connect an AI agent running on Amazon Bedrock to a chat interface built with Next.js. It focuses on one complete path: a Python agent talks to AWS, and a web browser shows the conversation as it streams in, including when the agent calls a tool. The project uses the Strands Agents SDK to build the agent, which runs on Amazon Bedrock AgentCore, a managed service from AWS for hosting and invoking agents. Messages and tool calls travel over a protocol called AG-UI, sent as a stream of events so the chat updates in real time instead of waiting for a full response. On the browser side, CopilotKit renders the chat and displays tool activity as it happens. A simple add_numbers tool is included so you can watch the tool call and its result appear step by step. To run it locally, you need an AWS account with access to Bedrock models, AWS credentials set up on your machine, the AgentCore command line tool, the uv Python tool, Python 3.12 or newer, and Node.js 20.9 or newer. Setup involves cloning the repo, copying example config files, installing dependencies for both the Python agent and the Next.js web app, starting the agent locally, then starting the web server. Once running, you can chat with the assistant and watch tool calls happen live. The project explains that its memory is not permanent. Conversation history is only kept for the current browser session, sent along with each new message so the agent has context. Closing the browser or starting a new session means the agent forgets everything. The README notes that durable memory across sessions would require a separate AWS feature called AgentCore Memory, which this starter does not set up. This is meant as a teaching example rather than a full application, useful for developers who want to understand how the pieces of an AI agent system fit together. It is released under the MIT License.