HomeLab update
My new HomeLab. Fewer VMs, more purpose.
From around 30 Hyper-V VMs to a smaller platform built around delivery, observability, recovery, and AI experiments. The first HomeLab 3.0 update.
The old lab is gone. Around 30 VMs on Hyper-V. Plenty of Windows, Linux, infrastructure, and virtualisation practice. Also plenty of things I had configured by hand and would rather not have to reconstruct from memory.
I started rebuilding it, changed my mind along the way, and took a different route. HomeLab 3.0 is the result.
A smaller platform, with a clearer job
The goal is a small environment where I can practise DevOps, SRE, and platform engineering properly. Deploy something, observe it, break it, recover it, and rebuild it from code.
VMware is the direction for the new foundation. Azure DevOps Services, local Linux agent pools, and AI-assisted engineering are coming together as the delivery workflow. The rest is a rollout, not a claim that every service in the diagram is already running.
What I am building around
- Infrastructure as code: Terraform, Bicep, and Ansible. Version the definitions and keep configuration repeatable.
- Containers and GitOps: Kubernetes or k3s, with Argo CD. Practise the route from a commit to a deployment, including rollback.
- Observability: Grafana, Prometheus, Alloy, VictoriaMetrics, and VictoriaLogs. Metrics and logs should help explain what happened.
- AI projects: local LLMs, LangGraph, Qdrant, and Neo4j. My own platform for agents, retrieval, and experiments.
- AI workflow visibility: OpenTelemetry and operational context for understanding how the workflows behave.
- Recovery: deployments, failure scenarios, and rebuilds. Something worth testing while the consequences are still just a broken lab.
Where things stand
The foundation is being rebuilt and the Azure DevOps workflow is coming online. Kubernetes, GitOps, broader observability, and the local AI platform are areas I am building towards. I will document what works as each piece is deployed and checked.
The diagram shows the planned layers and the current VMware direction. It is a structure to build towards, not a claim that every service is already live.
AI in the toolbox
GPT and Claude help with architecture reviews, infrastructure code, and automation. I define the requirements, review the changes, and own the result.
For AI-assisted operations, the design includes human approval, dry-run, and rollback mechanisms. Those are things to implement and test before handing an agent a destructive operation.
What comes next
Get the foundation and agents into a repeatable state. Build the first delivery path. Add enough telemetry to see what it is doing. Then test what happens when something fails.
There will be broken clusters along the way. That is part of having a lab.