How an AI agent investigates production incidents the way a seasoned engineer would — by gathering evidence, correlating it, and reasoning toward a root cause.

Welcome to HolmesGPT: Agentic Troubleshooting for Cloud-Native Systems.

HolmesGPT is an open-source AI agent, originally built by Robusta and now developed with Microsoft and accepted as a CNCF Sandbox project, that investigates production incidents and finds their root causes. Monitoring tells you that something is wrong; HolmesGPT works out why. It does this through an agentic loop that lets a large language model decide which observability tools to call, pull live data from Kubernetes, Prometheus, logs, Grafana and dozens of other sources, correlate the evidence across them, consult your runbooks, and produce a written root-cause analysis. This book follows that idea from the ground up. We begin with the problem of manual on-call debugging and what an investigating agent changes. We then open up the agentic loop at the heart of the system, the toolsets that give it eyes and hands, the runbooks that encode your team's knowledge, and the way it plugs into AlertManager, PagerDuty and Slack to close the loop on real alerts. We cover the language models it can run on, how to operate it as a CLI, a server, or a cluster operator, and we end with the practices that make it trustworthy in production. Throughout, the emphasis is on how the pieces actually fit together.

This title is part of the ShriIra library and is free to read in full, right here — our small contribution to making world-class knowledge easy to reach.

A note on reading it: open the Contents menu at the top of the reader to jump between chapters, use the Aa menu to set a comfortable text size, theme (light, sepia, or night), and single- or two-page layout. Your place is saved automatically, so you can always pick up where you left off.

We hope it serves you well.

— Shriira Press

Contents

1. Chapter 1 — The Problem of the 3 a.m. Page
2. Chapter 2 — The Agentic Loop
3. Chapter 3 — Toolsets: Giving the Agent Senses
4. Chapter 4 — Runbooks and Knowledge
5. Chapter 5 — Alerts, Integrations, and Closing the Loop
6. Chapter 6 — Models and Deployment
7. Chapter 7 — HolmesGPT in Practice