Home lab / Deployment notes
From an idea to a maintained service.
Practical notes from building my home lab: order of work, dependencies and checkpoints. This describes my approach; the Home AI demo configuration is being prepared separately.
Start with roles and dependencies.
I start with responsibilities: the network provides access, the NAS holds persistent data and services, and the PC supplies AI resources. For each service I record dependencies, configuration location, data to preserve and a way to check that it works.
Network and access before applications.
Stable address assignments and local DNS make communication predictable. WireGuard provides access to private services. A working tunnel or ping is only the beginning; I then check the application response and the correct HTTPS entry point.
Containers are replaceable; data must persist.
Compose projects have separate configuration and persistent data. Before an update I check the image, volumes and how to restore the previous state. A data backup is useful when its restoration process is understood.
HTTPS with persistent certificate state.
Caddy serves the public site and renews its Let’s Encrypt certificates. Website files are separate from proxy configuration and certificate storage. Updating content does not require recreating Caddy or deleting TLS data.
One entry point for personal services.
Homepage organises links to private services. Entries and widgets should reflect the current installation. The portal belongs to the home environment; the public portfolio remains a separate site.
Visibility before more automation.
I check availability, logs and the actual service function. Uptime Kuma, MySpeed, Grafana and Loki show different aspects of the environment. After a change I compare the signals and check for the expected outcome.
AI services and the Windows session.
Home AI startup is tied to Windows sign-in. I check Docker Engine, model servers and execution services. API availability and desktop interaction are separate conditions; GUI operations need an available session.
The whole flow, from knowledge to result.
HomeAI Knowledge provides documentation and RAG retrieval. PC Operator handles files, PowerShell, jobs and the Windows interface. An integration check should cover source selection, the tool call and verification of the actual result.
Release, verify and keep a way back.
Before a change I preserve configuration and required data. I deploy a limited scope, check functionality and retain a rollback path. The public site has its own release directory, so updates are separate from the agent services.
Public documentation has its own scope.
I describe service roles, dependencies and verification methods. Private data, keys and access configuration stay outside the website. Reproducible project materials will be published with a separate demo configuration.