Nightwatch, The open-source, read-only AI SRE

Nightwatch is a local-first, read-only AI SRE that sits on top of your monitoring stack and does one thing exceptionally well: it stops you from losing your mind during a 3AM PagerDuty nightmare. When alert storms hit—and they always hit at the worst possible moment—Nightwatch clusters those alerts into actual incidents, flags the noisy checks that are just wasting your time, and deploys a little "owl" agent into each of your environments to investigate what's actually broken. The owl keeps credentials local, dials outbound to a central brain, and leaves zero inbound holes in production. That's the security model. Read-only for now—these creators aren't naive enough to trust AI with production changes on day one.

The origin story is classic: a Kubernetes upgrade went sideways, rollback became impossible, and they had to fix it live at 2 AM while chaos piled up. They run on-prem and multiple K8s clusters, and in that moment they realized the real problem isn't the fix—it's figuring out what's actually broken in the first place. So they built a pair of digital eyes that lives in each environment, gathers evidence using read-only skills, and hands the on-call engineer a root-cause hypothesis instead of a blank terminal. No more starting from zero. No more guessing. Nightwatch just made its debut as an open-source weekend project, and if you've ever stared at 47 alerts at 3 AM wondering which one matters, you already know why this exists.