A guided, self-paced Build Path that walks with you step by step - from hardware that fits you and your budget to a home server shaped around your life: useful self-hosted apps, personal tools, even a personal AI agent at home. Any experience level, any starting hardware - and it grows with you.
Search "home server" and fifteen minutes later you're deep in Proxmox versus Ubuntu, contradicting forum threads, one-line install scripts, and advice written for people ten years in. Every answer sounds confident. None of them knows you, your hardware, or which step you're actually on.
The problem isn't missing information. It's overload - with no order, no context, and no next step.
Build My Homelab is built by Ilan Kushnir - the maker behind KWS Rack, the printed 10-inch rack system - out of a home server that runs for real, every day:
"AI can build you tools shockingly fast now. But a tool with nowhere to live stays a demo. Your home server is where those tools become part of your life."
These are real, daily uses - not a promise that everything takes ten minutes. Each one is a station on the path you'll walk yourself.
Home Assistant with backups and private remote access - automations that survive a power cut, not a fragile late-night install.
A folder of files turned into a real service: music and movies at home and on the road, without renting yet another interface.
Instead of another year of a paid generic app - a tool built for exactly how Ilan trains, running from his own server.
A task board that matches how Ilan thinks - instead of bending his work into someone else's template.
Signing workflows on open-source, DocuSeal-style tools. For some workflows that's enough, cheaper, and fully yours.
The lab lives in the printed rack Ilan designed. Same philosophy throughout: built deliberately, documented, made to grow.
Not promises - workloads. These five run at Ilan's home as you read this, and each one replaced an app, a subscription, or a compromise.
Including this page. The landing page you're reading - and the KWS Labs system behind it - is served from that same home server.
Ilan built Pumpy with Claude Code, and it runs from this very server - for him, his friends and his family. It works better for him than any generic fitness app for the simplest reason: it was built around exactly how he trains - and when his needs change, the app changes the same week.
See Pumpy live ↗Home Assistant runs the lights, the door lock, the electric blinds and the TV - alongside the routers, access points and a mesh of sensors and automations. And because the brain lives at home, not in a cloud, it keeps working even when the internet doesn't.
Media, photos, documents, bookmarks, file sync - the self-hosted world has a serious app for almost anything. One home server runs a surprising number of them at once, and each one can retire another monthly subscription.
Browse the catalog - selfh.st/apps ↗Ilan's personal kanban. Built around exactly how he plans his work - and wired to his AI agent, so tasks get added, sorted and closed as part of the conversation, not in yet another form.
The one that changes everything: Ilan's personal agents - Hermes and Astra - run from this server, connected to his projects, his tools, his task board and a long-term memory. Not a cloud chatbot with amnesia - a partner that knows the house it lives in.
This is where the path leads ↓
This is the whole point of the path: not "learn Linux" - end up with things like these, running quietly at your place.
You don't need a rack of servers to begin. A Raspberry Pi, an old laptop or a small mini PC is a real first server - and for many people it stays enough for years. The path adapts to your goals, your experience, the hardware you already own and the budget you're comfortable with.
Starting small closes no doors. The same path later grows toward a more capable machine - or a multi-server home setup - if and when you want it to. Nobody needs to start big.
The range the path covers - join at any point
A real first server for the cost of nothing - or close to it.
A common first stepQuiet, low-power, and enough for a surprising number of apps.
Room for a family's media, photos and serious storage.
A small home cluster, if your ambitions ever point that way.
None of these is required before you start. The path meets you at whichever step you're on - or helps you pick the first one.
"I spent seven years learning what belongs in a home lab - and what only adds noise."
I've been building home labs for about seven years. I started with Raspberry Pis, then tried more hardware and infrastructure approaches than I can count - including running Kubernetes at home. I made real mistakes along the way, and learned where complexity earns its keep and where it only gets in the way.
Through those same years I worked as a software development team lead, and later as a tech lead, at technology companies - architecture and infrastructure were my day job, and I brought the same seriousness home. Piece by piece, the lab became a server shaped around my work, my home and my family's real needs. Today it also runs systems other people use - including the KWS Labs pages you're reading right now.
On the physical side, I designed and 3D-modeled KWS Rack, a modular printable 10-inch rack system that grew into a community-supported ecosystem - print profiles, extensions and community add-ons. If a printed rack ever fits your hardware and your needs, you can build one too. It's never required.
Build My Homelab distills that journey into an ordered, personalized path - for builders at very different experience levels.
"One mini PC. A calmer business behind it."
Picture a small-business owner with no infrastructure background - a studio, a clinic, a workshop. The starting point could be exactly what's in the picture: one mini PC and one small network switch.
On a path shaped around that owner's needs, the first wins could be practical ones: appointment scheduling the business controls, customer inquiries collected in one organized place, shared documents and small automations - and, further along, a personal agent that takes over repetitive business chores.
The point isn't that every business gets these exact outcomes. It's that useful services can begin on modest hardware when the sequence fits the owner's real needs - with every step ordered and verified, so the builder never gets lost.
An example of the kind of path the program can shape - not a customer claim.
We believe anyone can build a home server - with a clear Blueprint shaped around their goals, their experience and the hardware they have. The missing ingredient was never talent. It's organized, personal guidance.
At the end of this path, a sentence is waiting for you:
"I built my own server at home."
In your voice, about your machine - earned one verified step at a time.
Not video chapters, not articles. One small loop - Brief, Decide, Build with AI, Verify, Record, Celebrate, Next - repeating from the first lesson to the last upgrade:
Why this step matters - one mental model, one diagram, two minutes.
A recommended default with its reasons. Not ten options - one, and why.
A ready prompt card for your agent, with context, boundaries and warnings.
One check you can see with your own eyes: it works, or it doesn't.
A field or two in your Server Passport. Your server stays documented, always.
A capability you didn't have yesterday, stated plainly.
Exactly one clear step forward. Not a menu. No getting lost.
And from Next - back to Brief. That's how a server gets built: loop after loop, no dangerous leaps.
A build map, not a syllabus. Every stage ends with something real running at your place - built together, verified together.
Hardware that fits your budget, a system that fits your situation, your coding agent wired in.
Running: a server you reach from across the room.Docker Compose done properly, and first apps picked from your personal plan.
Running: your first useful app - and you understand how.A dashboard front door, friendly names instead of IPs, private remote access via Tailscale.
Working: your dashboard on your phone, anywhere - privately.Monitoring, real backups - and a restore you've done with your own hands.
Verified: you restored it yourself - and know you can again.More capable hardware, heavier apps, maybe one deliberately published service.
Comes later, once the earlier stages are solidA home AI agent, a personal-apps lab, a physical lab around KWS Rack.
Continues from your finished first pathThe onboarding questionnaire learns what excites you, how much experience you have, your budget, and what hardware is already lying around - then shapes your KWS Blueprint: the same broad path, re-sequenced so your primary needs become the first milestones.
Nothing here is a fork in the road. Personalization is sequencing, not exclusion - whatever you don't start with stays reachable later, on the same path.
What should become useful first? Click to see how the plan re-sequences:
Every stage ships a prompt card written for your coding agent: your server's context, this stage's goal, and how to check the result. The agent does the hands-on building - the Build Path keeps the plan coherent and makes sure each result is actually implemented and verified.
What you'll need: access to a coding agent such as Claude Code, Codex or Cursor. You don't have to be an expert with it - it's a tool you can arrange before you start, and the path guides how you build with it.
You are my build companion for setting up Tailscale on my home server. ## Context - OS: Ubuntu Server LTS (fresh install, updated) - Apps run with Docker Compose under /opt/kws/stacks - My dashboard (Homepage) works on the local network - I am a beginner: explain every command in one sentence before I run it ## Goal Reach my dashboard from my phone, away from home, privately. ## Hard rules 1. Do NOT suggest port forwarding or exposing anything to the internet. 2. Do NOT ask me for passwords, keys, or tokens. If a login is needed, tell me to complete it myself in the browser. 3. One step at a time. After each step, tell me exactly how to verify it worked before we continue. Start by checking what we need to install, and wait for my output.
One habit the cards teach from day one: facts about your server go into the chat - secrets never do.
Guided doesn't mean done for you. Before you join the waitlist, check the sheet - honestly, in both directions.
The path includes explanations, prompt cards, verification checks and a structured Stuck Report for when something fails. It is not unlimited personal support, and it doesn't promise that Ilan will personally solve every router, ISP or hardware problem.
You should - a lot. This path is built around AI, not against it. But answering questions and building a system are two different jobs:
In short: your agent builds. The path gives it direction, context and sequence - and checks that each result actually works before you move on.
The guided path is being prepared for its official opening. Joining is free and commits you to nothing - and if you're not sure it's for you, ask right here.
Honestly: there's no launch date to promise yet. Joining simply means you hear first.
No. Build My Homelab is a separate product - a guided Build Path from the same maker. KWS Rack stays exactly where it lives, and nothing about its published files changes. A printed rack is a beautiful home for your server, and the path includes an optional physical-lab stage - but on day one, a shelf works just as well.
No. You need technical curiosity and the willingness to run a command after it's been explained to you in one sentence. The path teaches each concept exactly when you need it - not in an intro lecture. If you've ever written code, managed a technical product, or set up a smart home, you have more than enough.
For most people: a used or new mini PC, or a computer already lying around. The Blueprint questionnaire recommends a hardware path for your budget and goals - including "don't buy anything yet, start with what you have". No hardware at all? There's a supported VPS path. A server closet is not required, promise.
It can be done calmly, and that's exactly how the path is built: every step is explained before you run it, steps come in an order that makes sense, and each one ends with a verification you can see. Risky or advanced actions are never treated casually - they arrive only after the basics that make them sensible. We won't promise zero risk; these are real systems, and you remain responsible for your own machine and data. The path's job is to make sure you're never guessing.
No. On the core path, remote access runs privately through Tailscale - no router ports are opened. Publishing a service to the public internet is an optional later stage, meant only for people who truly need it, and it comes with its own careful sequence.
Full transparency, so you don't find out midway:
"Not yet" always means "not now, and here's why" - never a locked door without an explanation.
Better to say it here than after you sign up. This path isn't for you if:
If one of those is you - you'd likely be disappointed, and we'd rather save us both the time.
You need access to a coding agent - a tool such as Claude Code, Codex or Cursor that can run commands and build alongside you. A regular chat app isn't enough on its own, because the path is built around an agent that does hands-on work. You don't need to be an expert with it: an agent-readiness check is the first thing on the path, and the prompt cards are written to work in any capable coding agent, not just one.
English. The lessons, the Blueprint and every prompt card are written in English - agents are most precise with English infrastructure prompts, so that's also what you'll paste into yours. An earlier Hebrew edition exists and may return later, but the path launches in English.
Every stage has an escape lane: a Stuck Report template that produces an orderly failure report - what you expected, what happened, and the relevant log with secrets stripped - ready to paste into your agent. Getting stuck is part of learning; staying stuck isn't.
Early in the path you build a backup and run a full restore drill - you break it on purpose and bring it back. After that, a mistake isn't a disaster, just another lap of the loop. Genuinely risky stages simply don't unlock before that safety net exists.
