AgentOS: AMD-Powered Multi-Agent Workflow Console

AgentOS is a multi-agent workflow console I built for the AMD Developer Hackathon to test how practical AI agents can run on AMD GPU cloud infrastructure. The project goes beyond a normal chatbot. A user enters a task, and the system routes it through a structured workflow that includes planning, execution, validation, and memory recall. The frontend is built with React and Vite, while the backend is built with FastAPI. The backend connects to Mistral 7B served through vLLM on an AMD GPU cloud instance. AgentOS is designed for developers, small software teams, hackathon builders, and technical teams who need more than one-shot AI answers. It can help generate validation checklists, demo scripts, product strategies, readiness reviews, and implementation plans. The interface shows the final answer, plan, research status, validation result, memory context, workflow duration, and production connection status. The project uses ChromaDB for memory recall and includes a response-quality guard to reduce incomplete or placeholder outputs during live demos. The current demo runs through a working production path: React frontend → FastAPI backend → AMD vLLM endpoint → Mistral 7B response. This project fits Track 1: AI Agents & Agentic Workflows because it demonstrates a real agent workflow system powered by open-source model inference on AMD GPU cloud infrastructure.