Introduction I have Prometheus and Grafana monitoring my homelab. I have Alertmanager sending Discord notifications when a node goes down. But there was a gap in the middle that kept bugging me. Prometheus tells me that something is wrong. CPU is high. A container restarted. A scrape target is unreachable. What it does not tell me is why . For that, you need to read the logs. And reading Docker logs across multiple containers, multiple times a day, is the kind of task that feels productive for about ten minutes before you start skimming and missing things. So I built something to read them for me. A Python script that runs every 15 minutes, pulls Docker container logs, checks for anything that looks critical, and sends the critical stuff to a small language model running on my Oracle Cloud instance. The model reads the raw log entry and writes a plain-English summary. That summary gets posted to a Discord channel.…