Operation of infrastructures and applications is hard - number of monitored clusters, applications and layers is ever increasing, complexity of systems is sometimes overwhelming while number of knowledgeable engineers seems to be always lower than needed.

How to handle such increasing complexity without hiring new staff? And how to do it in a secure way, so no sensitive data is exposed to untrusted 3rd parties?

In this talk we describe how Sovereign AI Ops we built is making our Day 2 operations life easier.

You know the drill: There was a temporary outage during the night, you come to office in the morning and start your day by searching for known bugs of software you are using which might cause observed behavior, probably using some non-trusted AI. Try to imagine that instead of doing so by yourself, Sovereign AI Ops does it for you over the night and when you came to office, analysis is already prepared for your review, including proposed action items (commands). And all that happens without sending any sensitive data to untrusted 3rd parties.

First part of the talk explains what happens with your data when you are using AI which is not sovereign.

In second part we describe architecture of Sovereign AI Ops we have built and have been using on a daily basis for watching our Monitoring Dashboard and proposing actions. It involves open source projects only, including OpenStack, Kubernetes, vLLM, Hermes Agent and software developed within ALASCA and Sovereign Cloud Stack (SCS) non-profit consortiums.

Third part is a live demo, we will intentionally break (non-production 😀) workload and show how Sovereign AI Ops reduces our reaction time.