Running several products on a few servers creates a monitoring problem that the monitoring industry does not want to solve. Datadog and PagerDuty are priced for teams; the free tiers expire exactly when you start needing them. What I wanted was one screen that told me the truth about every box.
Atlas is a terminal application that runs 24/7 in tmux and collects over SSH. There is nothing to install on the machines it watches. It discovers what each app is by convention rather than configuration: containers by name, databases by image, certificates from the nginx vhosts, backups by looking for the files.
Deploys go through it too, and this is the part I am most careful about. Preflight shows the deployed commit against origin, you type the app name to confirm, output streams live, and verification runs whether or not the deploy claimed success. It is honest about rollback: if your deploy script cannot roll back, Atlas does not pretend it can.
It is MIT licensed and public, which changes how it gets written. Nothing in it can assume my fleet.