Outcome
You will have a one-page fit decision and lab brief: a first task, a success test, a failure boundary, and a maintenance owner. The initial lab can be one existing computer. A rack, hypervisor, GPU, and PebbleRack device are not prerequisites. Buying hardware is only one possible answer.
Before you start
Choose a task you genuinely want to do at home: serve a private notes page, try a local AI model, back up family files, or learn virtualization. For this series, the example task is run a small local web service and be able to recover it. You need a notebook or local document. Do not put passwords or your full network diagram in a public document.
Steps
- State the job in one sentence. Write “I want to run ___ for ___ people, on ___ devices.” Say whether the task must work when the internet is down. Avoid “build a complete home cloud”; it cannot tell you when you are done.
- Choose the first user and data class. Is this practice data, personal data, or irreplaceable family data? Use disposable practice data for the first installation. If you need personal data, plan backup and access controls before importing it.
- Write a visible success test. For the example service: “From the host computer, a local request returns the expected page; after a planned restart, it returns again.” A test should take minutes, not require subjective judgment, and name the device from which you will run it.
- Compare routes before buying. An existing cloud assistant may be best for everyday writing, search, or coding when convenience and strong hosted models matter more than running anything yourself. An occasional API call can be cheaper and simpler than maintaining a machine. A rented GPU may suit a short, compute-heavy experiment. Your current computer may already run the model or service you need. Dedicated local equipment is more plausible when the exact task needs offline use, recurring local capacity, hands-on learning, or a carefully verified data boundary. None of those benefits is automatic; compare the same task on the options you can access.
- Write a reason not to buy. If a free or existing subscription meets the success test, or if the workload is rare and setup time matters, stop at that route. If data must stay under a particular governance policy, check that policy and the actual data flow; “local” by itself is not a security guarantee. PebbleRack has no validated hardware performance, price, or support commitment to use as a fit claim today.
- Set the scope boundary. List what is excluded from version one: public internet access, remote login from outside the home, automatic changes by an AI agent, and production storage. You can add these later after you can observe and recover the baseline.
- Set a maintenance budget. Decide who can change the lab, how often it may be offline, and when updates will be reviewed. Reserve time for backup tests and documentation, not just installation.
- Choose a stop rule. If the chosen device has failing storage, unreliable power, or no way to recover access, stop and solve that problem before loading valuable data. Likewise, if you cannot explain what an installer will erase, do not begin installation.
- Record one next decision. You might need to choose whether an existing spare computer is suitable. Leave unknown items marked “unknown”; do not convert an attractive vendor specification into a measured outcome.
Check it worked
A second person can read your brief and tell what the lab does, why you chose local or another route, how to verify the task, and what you will do if it fails. Keep a dated copy so later changes have a baseline.
If it fails / rollback
If the first goal proves too broad, shrink it to one service and one machine. If the data or uptime need is serious, make the first milestone a restore test rather than a new application. Nothing in this guide changes a device, so rollback is simply revising the brief.
Safety and data notes
Treat home-lab experiments as capable of erasing disks and exposing services. Do not store your access keys or personal inventory in a shared guide. An AI assistant may help explain a plan, but it should not decide that an unknown network or disk is safe to change.
Sources
- Ubuntu Server, basic installation — backup and destructive partitioning cautions; checked 2026-09-29.
- Ubuntu Server, security suggestions — updates, least privilege, and firewall baseline; checked 2026-09-29.
Next guide
Continue to guide 02, “Inventory and map what you already own,” if a home lab still fits. You can read the entire series without subscribing; PebbleRack updates are optional if you want to follow the developing hardware plans.