Start here: what you are building and in what order

TL;DR: You can run useful home services without becoming a full-time system administrator. An AI operator can help you plan, diagnose problems, and do the routine work. Start by making sure you can get back into your computers, keep notes outside chat, and recover your files. Then add the services you want.

What you will have when you finish#

Perhaps you want family photos, shared files, or a media library on computers you control. Getting an application to launch is usually the easy part. The harder questions come later: why did it stop after a reboot, where is its data, and what happens if the person who set it up isn't available?

This guide uses an AI operator to help answer those questions and maintain the setup. You explain what you want and what the household depends on. The AI operator investigates the computers, prepares changes, and keeps a record that the next session can use. You shouldn't have to learn every command it runs. You should understand the decisions you're making, what might be interrupted, and how to recover if a change fails.

Your admin computer is where you and your AI operator work together to configure and troubleshoot the rest of the system. This guide uses an always-on Mac mini for that role; similar concepts apply to a dedicated Linux or Windows PC. Your home services can run on one server that also stores your data, or on separate application and storage servers. You can use new or existing equipment that meets your needs; Server setup walks through that assessment.

By the end, you should be able to use your chosen services, restore a file from backup, and recover access after a common failure. The AI operator should be able to find how your setup works without asking you to reconstruct months of chat history.

This guide speaks to you. Blocks labeled Prompt to paste are instructions for your AI operator. Read the explanation before each prompt, then paste the block when you reach that step. Replace any placeholders with your own details. A prompt can ask the AI operator to prepare a change; it cannot give the AI operator permissions that your computers haven't granted.

The five service tiers#

Some things matter more than others during an outage. These five groups help you decide what to protect and what to recover first. They are this guide's shorthand, not a standard you need to memorize.

For example, don't make your only remote-access connection depend on the experimental server you're trying to fix. Don't let a media conversion fill the disk that holds your family files. The dependency map you build with the AI operator should make these relationships visible.

The build sequence#

Read in this order. Each chapter ends with a practical check so you can tell whether you're ready to move on.

Phase 0 — decide what you want to build#

Tell the AI operator what you want, what equipment you have, and which disks or services it must leave alone. Let it inspect only the computers you've authorized. You finish this phase with a short equipment list, unresolved questions, and an agreed scope of work.

Start with how to work with an AI operator.

Phase 1 — set up your admin computer#

Use your admin computer for the AI operator's sessions and the working copy of your setup records. Give the AI operator a separate account. Check that you can sign in yourself without the AI operator, and commit and push the records to your private GitHub repository.

Continue with setting up your admin computer.

Phase 2 — set up your servers#

Install Linux on its own disk, with valuable data disks disconnected during installation. Check the hardware and learn how to reach the server from your admin computer. Test a direct cable connection before you need it during a network problem.

Follow Server setup.

Phase 3 — set up networking and backup power#

Record how the router, Wi-Fi, DNS, and remote access fit together. Test remote access from a phone using cellular data. If you're adding battery backup, include the network equipment needed to reach the server, as well as the server itself.

Use the network and power chapter.

Phase 4 — set up storage, accounts, and backups#

Identify the disks and decide which files need backup. Choose a storage arrangement that suits your capacity and recovery needs. Set up accounts and sharing, then restore a sample file from a separate backup before moving valuable data.

See storage, accounts, and recovery.

Phase 5 — add household services#

Start with one small application. Save its settings and record where it stores data. Check it from the device someone will actually use, and learn how to restore it before adding more applications.

The household services chapter walks through this using Docker Compose.

Phase 6 — add media and home automation#

Add the services you care about, such as Plex or Home Assistant. Use real playback or automation problems to decide whether extra hardware would help. A GPU, a cluster, or a more elaborate network should solve a problem you've observed.

Phase 7 — practice maintenance and recovery#

Arrange alerts that tell you when something useful stops working. Practice restoring a service and connecting remotely after a restart. Find out which failures still require someone at home.

Finish with maintenance and recovery.

Decisions that become expensive to reverse#

Spend time on the things that would force you to move data or rebuild the setup later. Which files are irreplaceable? How much storage do you expect to need? Who should be able to read or change those files? Which services must work while you're away?

Other choices can wait. You can add a dashboard or a faster graphics card later. Changing a storage layout after filling it can require another place to hold all the files. The storage chapter explains the options before you commit to one.

The short first interview#

Start with the equipment and immediate goals. You don't need to answer every future storage or networking question today.

Prompt to paste:

Help me plan a home-services setup that you can help maintain. Ask a short
batch of questions about what I want to use, the computers and network I
already have, and any data or services you must not touch. Distinguish facts
you can inspect from choices I need to make. Do not make changes yet.

Create a brief setup note with the goal, equipment, protected data/services,
unknowns, and the next useful step. Put it in our private GitHub repository
and its local working copy; help me establish those if they do not exist.
Before inspecting any computer, name it and confirm the permitted scope.

Who may do what on day one#

At first, let the AI operator inspect and prepare changes. Keep your administrator login separate from the account it uses. A standard account limits what it can change, though it can still edit files it owns. Truly read-only access needs additional restrictions on the server.

You keep your passwords and recovery keys in your password manager, with a recovery copy that doesn't depend on this homelab. The AI operator's notes should say where to find them, such as “password manager, Home server recovery,” without containing the secret itself.

As you gain confidence, you can authorize routine work within specific limits. Do that because the limits have been tested, not because the AI operator promises to be careful. Chapter 1 explains the distinction.

One command means one complete change#

When a change needs your administrator account, the AI operator should prepare the work before asking you to act. Its instructions should name the computer, explain the expected result, and give you one complete procedure with checks. It shouldn't send a new unexplained command after every reply.

Before running it, you should know what will change, what could stop working, and how to undo the change. If you can't tell, ask for a clearer explanation. You don't need to decode a long command to discover that it will erase a disk.

When to move to the next phase#

Use the chapter's final check. A screenshot saying “healthy” isn't enough if you haven't opened the service yourself or restored the file you meant to protect. Save the result in your setup notes, including any remaining limitation. Then the next session can continue from something you've actually checked.