The problem
Managed services hide the layers that matter. I want to run private AI bots on my own Kubernetes cluster, with my own monitoring, and know every layer underneath.
What I am building
Stage 1 of 6 is underway: building the cluster by hand. Monitoring with Prometheus and Grafana, one bot, GPU support and codification are later stages. They are plans, not completed capabilities.
Decisions and trade-offs
Build by hand first, add monitoring by hand, then run one bot. GPU support comes later. Only then codify the setup and test tearing it down and rebuilding from code.
This sequence puts understanding before automation. It takes longer, but gives me a chance to explain what the automation is doing.
What I would do differently
I’ll record the failures and changes at each stage.
Links
Follow the Now page for the current focus. There is no public repository linked yet.