Home lab / Infrastructure
The infrastructure behind Home AI.
My home lab connects the home network, a NAS and a GPU-equipped PC. Each has a defined role: access, persistent services or compute. I develop AI integrations in an environment I manage and observe.
Three layers, one environment.
The NAS runs services independently of the desktop session. The PC provides resources for models and Windows tools. Separating these roles makes updates, dependency diagnosis and data management easier.
Access matched to each service.
MikroTik handles networking, DNS and WireGuard connections. Private services are accessible according to their purpose; selected public sites are served by Caddy over HTTPS. Running in the home lab does not automatically make a service Internet-facing.
Everyday services and persistent data.
On Synology I run a service portal, monitoring, a music library, a password vault and this public website. Application data lives beyond individual containers. When making changes, I distinguish code, configuration and state that must be preserved.
- Homepageentry point for private services
- Uptime Kumaavailability and history
- MySpeedconnection measurements
- Grafana · Loki · Alloylogs, event collection and exploration
- Navidromemusic library
- Vaultwardenpassword manager
- CaddyHTTPS and website hosting
Watching services and the impact of changes.
Availability, connection measurements and logs answer different questions. Uptime Kuma checks whether an endpoint responds. MySpeed records throughput history. Grafana and Loki help explore events and compare them over time.
A green status is one signal. I also check functionality from the user’s perspective, such as an actual API response or a task result.
Home AI builds on this foundation.
Open WebUI, models and PC Operator run on the PC. Knowledge services and integrations connect them to documentation and selected environment functions. The full flow is described in a separate case study.
The PC also runs ComfyUI. The GPU controller queues its own gateway traffic in cooperative mode; other applications can still access the GPU.
This website runs on the NAS, too.
The portfolio is a static website served by Caddy. Public release files are separate from private service configuration. Each version is deployed in its own directory, with a way back to the previous release.
Public HTTPS uses Let’s Encrypt certificates renewed by Caddy. The site does not call the agent’s private APIs or provide visitors with administrative tools.
Engineering decisions and lessons