
Most production incidents aren’t new. They’re the same few problems, showing up at the worst possible time. 🤔
A crashing pod, a stuck rollout, a bad deploy on a Friday evening. Known problems, known fixes, and someone still gets paged.
During our Impact Days, one team asked: what if an agent took the night shift?
The result is KGuard, an agent on call for Kubernetes clusters. It watches traces, metrics and pods, collects evidence from logs and events, and finds the root cause. Then it decides who should fix it: KGuard itself, a coding agent that opens a pull request, or a person.
The AI only proposes. Fixed rules decide what it may change. It has read-only access, tightly scoped permissions, never touches databases or secrets, and every fix passes a policy check before it runs. Afterwards, KGuard verifies that the fix actually worked. New fixes go into the runbook, so it gets better every time the team solves something new.
Today, KGuard handles safe restarts, rollbacks and clean-ups on its own. And by design, the language model never decides freely. It follows the runbook, and the rules have the final say.
What this means for the team: no night shifts for routine incidents, and no dropping everything during the day because a pod restarted. Instead of hunting for issues in the cluster, people can stay focused on their actual work.
The biggest lesson: a good knowledge base matters more than the model. Routine incidents get handled within clear limits, and people can focus on the problems that actually need them.
KGuard keeps things running. Your team keeps things moving.👋
Curious how it works in practice? We’re happy to walk you through the demo.
#Kubernetes #AgenticAI #SRE #ImpactDays #arconsis