building
Building the Foundation — Talos, Nodes, and Cilium
Bootstrapping a Talos Linux cluster with Omni, configuring node labels and zones, and replacing Flannel with Cilium's eBPF networking.
Field notes from a Kubernetes homelab
Build it. Operate it. Understand it.
A heterogeneous cluster, local AI, and the decisions that hold them together. Follow the builds, investigate the failures, and take the useful parts home.
Start here
Three ways into the project
operating
Operating on Green
Five ways a healthy dashboard lies — wrong artifact, unconsumed signal, out of scope, stale view, never applied — and the command that checks each
papers
Why Run Your Own Cluster in 2026?
Three philosophies dominate infrastructure decisions in 2026: cloud (pay the bill, ignore the metal), managed homelab-as-code (own the iron, outsource lifecycle), and DIY…
What are you working on?
All topicsDebugging KubernetesUnderstand what a healthy status actually proves, and find the checks that matter.Running local AIPut mixed GPUs to work, serve models, and manage inference workloads.Securing and operating agentsWorkstations, orchestration, identity, and the boundaries around autonomous tools.Choosing infrastructureNetworking, storage, recovery, and delivery decisions under real constraints.
Latest writing
Retrieval on the Idle iGPUs
Embeddings and reranking on my control plane's idle Arc iGPUs via Kubernetes DRA, plus the CPU control arm that decided whether the accelerator earned it
Operating on Green
Five ways a healthy dashboard lies — wrong artifact, unconsumed signal, out of scope, stale view, never applied — and the command that checks each
The Metrics API — Teaching Frank to Read His Own Vitals
kubectl top was dead for months because nothing served metrics.k8s.io. The fix was one Talos flag and a decision I almost got wrong: build the fancy thing, or the right thing.
Operating the Metrics API — kubectl top, HPAs, and the Silent-Empty Failure
How to check the resource Metrics API is healthy, add a CPU/mem HPA, and diagnose the one failure mode where the pod is Ready but top is empty.
Recently updated
Why Build a Kubernetes Homelab?
The motivation behind Frank, the Talos Cluster — learning enterprise infrastructure and building interesting projects on your own hardware.
Paperclip — An AI Agent Orchestrator on Frank
Deploying Paperclip — an AI orchestrator that organises agents into virtual companies with org charts and budgets — alongside Sympozium, to compare two fundamentally different agentic paradigms.
Operating on Paperclip
Checking Paperclip health, database access and secret sync, handling the RWO PVC constraint, and bootstrapping the pure-upstream instance after the 2026-10 fresh start.
Operating on Cluster & Nodes
Day-to-day commands for checking cluster health, managing Talos nodes, and debugging Cilium networking on Frank — including the real failure patterns the gotcha docs cover.
