# I Built a DevOps Chatbot That Checks My Live App for Failures — Here's How It Works

Why I Built This Every DevOps engineer has had the 2am moment. Something is broken. You don't know what. You SSH in, check logs, Google the error, open five tabs, still nothing clear. Thirty minutes later you find it — a config flag someone changed, a slow query, a dependency timing out. I wanted to ask an AI instead. Not a generic ChatGPT that gives you textbook answers, but an AI connected to my actual running system that can check what's broken right now. So I built the AI DevOps Copilot — Project 01 of my 30-project AI + DevOps YouTube series. How It Works The system has four layers: 1. LangChain agent (the brain) Uses create_tool_calling_agent with Llama 3.1 via Groq. When you ask a question, the agent decides whether to answer from knowledge or call a tool. General DevOps questions → instant answer. Questions about the live app → tool call. 2. ChromaDB RAG (the knowledge base) Nine runbook documents embedded into a vector database — Docker troubleshooting, AWS debugging, Kubernet