You don't need a datacenter to own your data. A single mini PC, an old laptop, or a used office desktop is enough to run media servers, photo libraries, note-taking apps, VPNs, and even AI models — for the cost of the electricity they draw. Every plan is free forever; the only recurring bill is the power company's, and their support is famously unresponsive.
Why self-host
- Ownership. No service you rely on can shut down, change terms, or raise prices overnight. Your data lives on your disks, in formats you can move.
- Privacy. Viewing habits, documents, photos, and AI prompts stay inside your walls. Nothing phones home unless you tell it to.
- Learning. Running real services on real hardware teaches networking, storage, containers, and automation better than any course.
- Cost. One upfront hardware purchase replaces a drawer full of monthly subscriptions. Power-efficient hardware pays for itself.
- Resilience. With good backups, your services survive cloud outages, account lockouts, and discontinued products.
What it takes
- A hypervisor. Proxmox VE turns one machine into many — isolated containers and VMs, snapshots, ZFS storage.
- Orchestration. Komodo manages Docker stacks across machines from one panel, with git-backed configs and push-to-deploy.
- Automation. A media automation layer fetches and organizes media with rules instead of manual work.
- Intelligence. Local LLMs and an API gateway keep AI workloads on your own GPUs.
- Safety nets. Layered backups — local snapshots, offsite restic/borg, tested restores.
- A front door. A reverse proxy with Let's Encrypt, an auth layer, and DNS — see reverse proxy & exposure.
The guides
Each service below gets its own page: what it does, why people run it, and what to know before starting. Everything is written generically — pick the pieces you need.
- Proxmox VE — the foundation
- Komodo — Docker orchestration
- Local LLM inference & gateways
- Photo management
- Backups that actually restore
- Reverse proxy, certificates & auth
- Home networking & DNS
Rules that keep it sane
- Config in git. If a stack's compose file isn't in a repo, it doesn't exist. Push-to-deploy beats hand-editing.
- Bind mounts over volumes. Data in plain directories is visible, movable, and backup-friendly.
- One job per container. Troubleshooting gets 10x easier when each unit does one thing.
- Backups are only real after a restore test. Schedule restores, not just backups.
- Power matters. Efficient hardware and sensible power limits turn an always-on server into a rounding error on the electricity bill.
Start with one service you actually use. The rest grows out of it.