A server in my house, running since February 2023. It hosts 8+ services in Docker on TrueNAS, used by 4–6 people every day from 6+ devices.
What runs on it
Immich for photos.
Navidrome for music.
Open WebUI for local LLMs.
Pi-hole for network-wide ad blocking.
Jobhub, my job search pipeline, which has its own case study.
How people reach it
Everything goes through Tailscale, a private VPN, so nothing is exposed to the public internet. That covers phones and laptops, including other people’s.
Knowing when it breaks
Other people depend on this, so I shouldn’t be the last to find out it’s down. Uptime Kuma checks each service every 60 seconds and sends a Telegram alert the moment one drops. It has held 99%+ uptime.
Choosing a local model
I benchmarked quantized LLMs in Open WebUI, comparing Qwen 3.8 27B against a smaller, more heavily quantized variant on speed and on a test set of my own. I kept the 27B for its answer quality.