Let a Bot Explain Why Your Pods Are Dying

What it does This is a Robusta playbook that catches Prometheus alerts via webhook, ships them to OpenAI’s text-davinci-003 model, and posts the resulting explanation to Slack. You get a button on each alert; press it, and the bot offers a pithy diagnosis of why your cluster is unhappy. It is not a standalone service, but a thin prompt layer on top of the Robusta.dev alert-response platform.

The interesting bit The README cheerfully confesses that “ChatGPT” is a marketing convenience: under the hood it is the GPT-3.5-based text-davinci-003. The real hook is the honest admission that the bot is currently flying blind—it does not yet feed pod logs or kubectl get events into the prompt, even though Robusta already collects that data, leaving the answers only as informed as the alert labels allow.

Key highlights

• Surfaces LLM-generated triage advice directly inside Slack threads
• Implemented as a Robusta playbook, leveraging existing Kubernetes alert plumbing
• Uses text-davinci-003, not ChatGPT proper, despite the repo’s name
• Explicitly deprecated by its maintainers in favor of the broader HolmesGPT successor
• Designed to enrich Prometheus alerts, though richer cluster context remains unimplemented

Caveats

• The project is officially deprecated and replaced by HolmesGPT; expect no further maintenance
• It is essentially integration glue between Robusta, OpenAI, and Slack rather than a standalone debugging tool
• Answers are generated from alert metadata alone; the README identifies pod logs and events as a “future improvement,” so expect generic guidance

Verdict A neat proof-of-concept for LLM-augmented on-call workflows, but only if you enjoy maintaining abandoned glue code. New adopters should head straight to HolmesGPT or roll a richer prompt pipeline themselves.