How to Run AI on Your Own Computer: A Plain-English Guide to Local LLMs

You use a chatbot every day, and one evening you wonder where your questions go. A local LLM is the answer to that worry. It is an AI chat model that runs on your own computer, so your words stay on your machine. Free tools such as Ollama and LM Studio make the setup short on a recent laptop. The trade-off is simple: local models are smaller than the big cloud ones, so they know less and make more mistakes.

Who this is for: you are curious about AI, you have a fairly recent laptop, desktop or mini PC, and you want private, offline AI without a subscription. Who it is not for: anyone who needs the best possible answers for work today, or whose computer has little memory (the memory section below explains how to check). A cloud service will serve you better.

What a local LLM is, in plain words

LLM stands for large language model. It is the type of AI behind chat tools: a program that has read a huge amount of text and predicts good next words. The "model" is one big file, often several gigabytes. A "runner" is the app that loads that file and lets you chat with it.

When you use a cloud chatbot, the model runs in a data center. Your question travels over the internet, and the answer comes back. With a local LLM, the model file sits on your own disk. Your computer does the work. Once you download the model, you can unplug the internet and keep chatting.

You may also see the words "open model" or "open weights". That means the company that made the model lets you download it. Each model has its own license (the rules for use), so check the license on the model's page before you use it for a business.

Why run a local LLM?

• Privacy. Your notes, letters and family questions stay on your computer. The Ollama FAQ says, as of October 2026, that Ollama does not see your prompts or data when you run models locally.
• It works offline. On a plane, in a cabin, or during an internet outage, it still answers.
• No monthly fee for local use. Ollama and LM Studio are free to run with local models. Your cost is the hardware you already own and the power it uses.

What a local LLM cannot do (yet)

Be honest with yourself here. A model that fits on a laptop is much smaller than the models behind the big cloud services. In practice that means:

• More mistakes. Small models get facts wrong more often. The Home Assistant docs put it plainly: smaller models are more likely to make mistakes than larger ones. Check anything that matters.
• Old knowledge. A model only knows what it learned up to its training date. Most local setups do not search the web for you.
• Short memory by default. The "context window" is how much text the model can keep in mind at once. As of October 2026, Ollama's docs say it uses a 4k context (about a few pages of text) on machines with under 24 GiB of graphics memory. You can raise it, but that uses more memory.
• Speed depends on your machine. A model that answers quickly on one computer can crawl on another. Only a test on your own hardware tells you.

For drafting, summaries of your own notes and brainstorming, a local model is often good enough. For legal, medical or money decisions, no model is a source of truth.

Ollama vs LM Studio: which tool to start with

Both tools download models and run them on your machine. They feel different. These facts are as of October 2026. Check the vendor pages before you rely on them.

Factor	Ollama	LM Studio
What it feels like	A small background service you drive from a terminal (a text window for commands), plus a desktop app on Mac and Windows	A desktop app with a chat window, a model browser and settings
License	Open source, MIT license	Free, but closed source; terms allow personal and internal business use
Platforms	macOS 14 or newer (Intel Macs run on the processor only), Windows 10 22H2 or newer, Linux	macOS 14 or newer on Apple Silicon, Windows (x64 and Arm), Linux (Ubuntu 20.04 or newer)
Where models come from	The Ollama model library	Hugging Face, a large public model hub, searched inside the app
Chat with your documents	Drag a text file or PDF into the Mac or Windows app	Built in, offline
Lets other apps use the model	Yes, a local API (a door other programs use) on port 11434	Yes, a local server that copies the OpenAI API style
Optional cloud models	Yes, after you sign in	Yes, on paid plans
Best for	Home servers, scripts, and feeding other self-hosted apps	Trying models on a laptop with no terminal at all

Sources: the Ollama GitHub repo and docs, and LM Studio's system requirements and app terms.

My suggestion: if you want to click, not type, start with LM Studio. If you plan to run AI on a home server and connect other apps to it later, start with Ollama. This is a judgment on ease of use, not a tested ranking.

What you need: memory first

The single biggest limit is memory. The whole model must fit in memory while it runs, plus room for the conversation. If it does not fit, it runs very slowly or not at all.

There are two kinds of memory to know:

• RAM is your computer's main working memory.
• VRAM is the separate memory on a graphics card (GPU). Models run much faster when they fit in VRAM. Apple Silicon Macs share one pool of memory between the processor and graphics, which helps.

Here is how to size it without guesswork. Check the official guidance for the model before you download it. As of October 2026, the Ollama quickstart says the gemma4:e2b download is about 7.2 GB, and it recommends 8 GB of available graphics memory, or unified memory on a Mac. With less, Ollama can use system RAM, but answers come more slowly. Bigger models need more. Each model's page on the Ollama library lists the size of every version.

LM Studio's own system requirements recommend 16 GB of RAM or more. On Windows and Linux x64, the processor must support AVX2, a set of math instructions. Check your processor's spec page if you are not sure. A graphics card is optional. Ollama supports NVIDIA, AMD and Apple graphics, and runs on the processor alone if there is no supported card. Expect slower answers that way.

You also need disk space. Ollama's docs warn that models can take tens to hundreds of gigabytes once you collect a few. Start with one small model.

Your first local LLM with Ollama, step by step

These steps follow the Ollama quickstart as of October 2026. Commands change, so check the docs if one fails.

1. Install Ollama. On Mac or Windows, download the app from ollama.com and run it. No administrator rights are needed on Windows. On Linux, open a terminal and run the official install script. (It is good practice to open the script URL in a browser and read it first.)

curl -fsSL https://ollama.com/install.sh | sh

2. Download a small model. This is the model the Ollama quickstart uses. The download is about 7.2 GB, so it takes a while on a slow connection.

ollama pull gemma4:e2b

3. Start a chat. Type a question at the >>> prompt and press Enter.

ollama run gemma4:e2b

4. Leave the chat. Type /bye or press Ctrl+D.
5. See what is loaded. This shows running models and whether they use the processor or the graphics card.

ollama ps
ollama ls

Want a chat window instead of a terminal? As of October 2026, Ollama's Mac and Windows apps include one.

The same thing in LM Studio

Install the app for your system. Open the model search, pick a small model, and download it. Compare its file size with your free memory, as above. Load it and start chatting. That is the whole job, and you never open a terminal.

The risky bit: what to keep away from it

A local LLM is private, but it is not automatically safe. Watch for these.

• Do not open it to the internet. As of October 2026, Ollama listens only on your own computer (address 127.0.0.1, port 11434) by default. Its API has no password, so anyone who can reach that port can use it. The OLLAMA_HOST setting can change that. Never forward that port on your router. If you want to reach it from your phone, use a private VPN instead.
• Know when you are using the cloud. Both tools now offer optional cloud models. In Ollama, model names that contain "cloud" run on Ollama's servers, not yours. If privacy is your reason, pick a local model.
• Do not hand it the keys. A chat model that only talks is low risk. An AI agent that can run commands, read files or send email is a different thing. Keep agents in a virtual machine, a computer-in-a-computer you can reset. The Hermes vs OpenClaw comparison explains how.
• Do not paste secrets. Passwords, bank details and other people's private data do not belong in any chat, local or not. Chat history can be saved on disk.
• Get models from known places. Use the official Ollama library or well-known publishers on Hugging Face.
• Check its answers. It will sound confident when it is wrong.

Connect it to AI agents (when you are ready)

A local model can also be the "brain" for a self-hosted agent. Both Hermes Agent and OpenClaw can use Ollama. Hermes needs a much bigger context than Ollama's small default, so you must raise that setting. Follow the step-by-step guides to install Hermes Agent in an Ubuntu VM or run OpenClaw with Docker Compose in a VM. Expect an agent on a local model to need more memory than chat alone.

The recovery step: clean up or remove it

Local AI is easy to undo. Models are just files.

• Free memory now: ollama stop gemma4:e2b. Ollama also unloads an idle model after five minutes by default.
• Free disk space: ollama rm gemma4:e2b deletes that model.
• Find the files: models live in ~/.ollama/models on a Mac, /usr/share/ollama/.ollama/models on Linux, and C:\Users\%username%\.ollama\models on Windows.
• Remove Ollama on Linux: the Linux docs list the full steps. They start with these two commands.

sudo systemctl stop ollama
sudo systemctl disable ollama

On Mac and Windows, uninstall the app the normal way, then delete the models folder. In LM Studio, delete models from its model list. If you test on a spare machine or a virtual machine, a snapshot taken before you start is the simplest undo of all. The guide to starting a home lab shows cheap ways to get one.

Local AI is one part of a bigger move to run apps at home. The overview of self-hosting in 2026 shows where it fits.

FAQ

Is running an LLM locally free?

The software can be. As of October 2026, Ollama is open source and LM Studio is free for personal and internal business use with local models. You pay for hardware and power. Both also offer optional cloud models, which you do not need.

Can I run a local LLM without a graphics card?

Yes. Ollama runs on the processor alone when there is no supported graphics card. Answers come more slowly. Start with the smallest version of a model and see if the speed suits you.

How much RAM do I need for a local LLM?

It depends on the model. Check its official page before you download. As of October 2026, the Ollama quickstart recommends 8 GB of available graphics memory, or unified memory on a Mac, for its starter model. LM Studio recommends 16 GB of RAM or more.

Is a local LLM as good as ChatGPT?

Usually not. Models small enough for home hardware make more mistakes and know less than the large cloud models. They are good enough for many everyday writing and summary tasks, and they keep your data at home.

Is Ollama or LM Studio better?

Neither wins for everyone. LM Studio suits people who want a point-and-click app. Ollama suits people who want a background service that other apps and home servers can use. You can install both.

Is a local LLM private?

When you run a local model, your prompts stay on your computer. Privacy breaks if you choose a cloud model, expose the port to the internet, or connect it to tools that send data out. Check each of those.

Related guides

• Hermes vs OpenClaw on Virtual Machines: Which Fits Your Setup?
• Install Hermes Agent on an Ubuntu Virtual Machine
• Home Server Setup: Hardware, OS and Your First Services
• Containers vs Virtual Machines: When to Use Which
• What Is a Home Lab? How to Start One in 2026

Self-Hosting in 2026: Why People Are Bringing Their Apps Home (and What It Really Takes)

Your phone says your photo storage is full again. The streaming bill went up. A new AI tool wants a monthly fee and all your documents. Self hosting is the other option: you run the app on a computer you own, in your home, and your data stays there. In 2026 it is easier than it has ever been, but it is still a small job you take on, not a product you buy. This guide shows why people do it, which tools to know, and what it really takes.

Who this is for: you are curious, you have an old laptop, a mini PC or a NAS (a network storage box), and you want to run one or two useful apps at home safely. Who it is not for: you want zero upkeep, or nobody in your house will ever check a backup. In that case, a good paid cloud service is the better choice.

What self hosting means, in plain words

Most apps you use run on a company's computers. You rent access, and you follow their rules on price, storage and privacy.

To self host is to run your own copy of an app on your own hardware. A "server" here is just a computer that stays on and serves apps to your other devices. It can be a ten-year-old desktop under the stairs. Your phone, TV and laptop then talk to that box.

Why self hosting is growing now

No single reason is new. Together, they explain why normal people, not only IT staff, now run a box at home.

• Subscription fatigue. Many households now pay for photo storage, music, video, cloud drives and AI tools, all on monthly plans. Self hosting replaces some of them with a one-time hardware cost and your time.
• Privacy and control. Family photos and documents are personal. When they sit on your box, no company can scan them, change the terms or close your account.
• Local AI. You can now run an AI chat model on your own computer. Your questions and files do not leave the house.
• Small, quiet hardware. Modern mini PCs are small enough to sit on a shelf and quiet enough to forget about. Used office PCs are another common start.
• A push toward open source. After Broadcom bought VMware, it changed how VMware products are licensed. Some small users and home labbers moved to open source platforms such as Proxmox as a result.

The self hosting map: who does what

Sort the projects by the job they do. This is a map of well-known names, not a tested ranking. Check each project's own site before you commit.

Platforms: the base layer

The platform is the software that runs on the box itself. Your apps sit on top of it.

Platform	What it is	Cost model (as of October 2026)	Good fit if
Proxmox VE	A hypervisor (software that runs many virtual computers on one machine) with a web page to manage it	Open source (AGPLv3), free to use. Optional paid support subscription.	You want to learn, and run many separate apps safely apart
Unraid	A storage and apps system that mixes drives of different sizes	Paid license, bought once, with a set period of updates or lifetime updates by tier. Free trial.	You have a pile of mixed hard drives and want simple app installs
TrueNAS Community Edition	A storage system built on ZFS (a file system that checks your data for damage)	Free, open source	Safe storage comes first, apps second
Synology DSM	The operating system that comes on every Synology NAS	Comes with the Synology hardware	You want a ready-made box and a polished interface
umbrelOS, ZimaOS	Appliance-style home server systems with a one-click app store	umbrelOS: free for personal use under a noncommercial license. ZimaOS: a free edition with limits on disks and users, plus a paid Plus license.	You want the easiest start and accept fewer controls

Check the current terms on each vendor page. For example, Unraid lists its tiers and trial on the Unraid license page. CasaOS, from the maker of ZimaOS, is still open source, but its team said in January 2026 that it is in a maintenance phase, with new work going to ZimaOS. For more on Proxmox itself, read what Proxmox VE is and why home labs run it.

What it means for you: there is no single right platform. Pick the one that matches your first goal. Storage first points to TrueNAS, Unraid or Synology. Learning and many apps points to Proxmox. Fastest start points to an appliance system.

Apps: the reason you do this

All app details below are as of October 2026.

• Immich replaces a cloud photo service. It backs up phones and shows a timeline, faces and maps. Immich announced its first stable release, v2.0.0, on 1 October 2025 (see the Immich v2.0.0 release post).
• Jellyfin is a media server for your own movies, shows and music. The project says the server and official apps are free, with no tracking.
• Home Assistant runs your smart home locally. Its own tagline puts "local control and privacy first."
• Nextcloud gives you files, calendar and contacts in one place, like a private cloud drive.
• Vaultwarden is an unofficial server that works with Bitwarden password apps. It is not made by Bitwarden, Inc., and its own docs tell you to back up its data.

What it means for you: start with one app that solves a real annoyance. Family photos are a common first pick, because the value is clear. They are also the data you can least afford to lose, so backups come first.

Remote access: reaching home from outside

Your apps are most useful when you can reach them from your phone on the train. The old way was "port forwarding", which opens a door on your router to the whole internet. That invites break-ins.

A safer pattern is a private network between your own devices. Tailscale is the well-known option here. It is built on WireGuard (a modern, fast VPN protocol) and works across home routers without port forwarding. As of October 2026, the Tailscale pricing page lists a free Personal plan with up to 6 users and unlimited devices. Verify that before you plan around it. Plain WireGuard is the do-it-yourself version, with more setup.

What it means for you: do not open ports on your router for your first apps. Use a private network instead.

Local AI: the new arrival

Ollama is a free tool (MIT license) that downloads and runs AI language models on macOS, Windows or Linux, per its Ollama GitHub page. Open WebUI adds a chat page in your browser, much like the big AI chat sites. Note that Open WebUI changed its license in 2025 to add a branding clause. It stays free to use, but read its license page if you plan to rebrand it.

The limit is memory. Bigger models need more RAM or a graphics card with more video memory. Check the model page for its size before you download it. Some people go further and run AI agents that can use tools. If that interests you, run them in a separate virtual machine. The guide to Hermes vs OpenClaw on virtual machines shows why.

What it means for you: local AI is real and private, but a small box runs small models. Expect it to be slower and less capable than the big cloud services.

Trends to watch, and what they mean for you

• Free tiers with limits. Appliance systems now mix free and paid editions. ZimaOS caps disks and users on its free edition, and umbrelOS is free only for noncommercial use. What it means for you: read the free tier's limits before you build on it.
• Licenses that change. Unraid moved new buyers to tiers with a set period of updates in 2024. Open WebUI added a branding clause in 2025. What it means for you: check the license again before a big upgrade.
• Apps reaching "stable". Immich's stable release is one example. What it means for you: fewer breaking updates, but you still read release notes before you update.
• AI at home. What it means for you: you can keep sensitive questions private, but you take on the job of keeping the AI box updated and fenced off.
• Containers everywhere. Many self-hosted apps now ship as containers (a packaged app that shares the host's operating system core). What it means for you: installs are faster. The guide on containers vs virtual machines explains when to use each.

What self hosting really takes

Hardware

Most people start with what they own: an old laptop, an old desktop or a NAS. A small, low-power mini PC is the common next step. The box runs all day, so power use matters more than raw speed. Check the CPU supports hardware virtualization if you plan to run virtual machines, and enable virtualization in the BIOS first. The home server setup guide walks through hardware classes and a first OS.

Skill

You do not need a degree. You need to be calm with a web dashboard, a few typed commands and an install guide read line by line.

Time

Plan a weekend for the first app, including backups. After that, budget a little time each month for updates and a quick check that backups still run. If nobody will do that, the box will slowly become a risk.

Cost

Prices change, so I do not quote them. What drives cost is simple: the hardware, the electricity to run it all day, extra drives for backups, and any paid license (Unraid, for example). Some apps also have optional paid tiers. Check each vendor page as of the day you buy.

The risky bit: what can go wrong

• One disk, one copy. If your only copy of the family photos is on one drive in the box, one dead drive loses them. Immich's own README warns you to follow a 3-2-1 backup plan (3 copies, on 2 types of storage, 1 off-site) for this reason.
• Exposed apps. An app reachable from the open internet gets attacked. Keep apps private and reach them through a private network.
• Weak account security. Your remote access account is now the key to your house. Turn on two-factor login.
• Silent failures. Backup jobs can fail for months without anyone noticing. Check them.
• Too much at once. Ten apps in a weekend is ten things to update. Add one at a time.

The recovery step: before you trust it

Before you move anything important onto the box, set up a backup and prove you can restore it. A copy you have never restored is a hope, not a backup. Keep at least one copy off the box, and one copy away from the house.

Also know how to undo it. If self hosting is not for you, export your data in a standard format (photos as files, passwords as an export) and move back to a cloud service. Keep the old cloud account until the new setup has run well for a few months.

Who should not self host

• You want to host email. It is hard to keep mail from landing in spam, and most people should not try.
• You run a small business with customer data and legal duties. Get professional help, or start with the guide to virtualization for small business.

FAQ

What does self hosting mean?

Self hosting means you run an app on hardware you control, usually at home, instead of using a company's cloud service. You get control and privacy. You also take on updates, security and backups.

Is self hosting cheaper than subscriptions?

Sometimes, over a few years, but not always. You pay for hardware, electricity, backup drives and your time. Add up your current subscriptions, then compare with those costs before you decide.

Do I need a powerful computer to self host?

No. Many first apps run well on an old laptop, an old desktop or a small mini PC. Local AI is the exception, because larger models need much more memory.

Is self hosting safe?

It can be, if you keep apps off the open internet, use a private network, turn on two-factor login and keep tested backups. Most problems come from exposed apps and missing backups.

What should I self host first?

Pick one app that fixes a real annoyance, such as a photo library or a network ad blocker. Set up its backup on day one. Add the next app only when the first runs without fuss.

Can I self host AI?

Yes. Tools such as Ollama run language models on your own computer, and Open WebUI gives them a chat page. Expect smaller and slower models than the big cloud services on modest hardware.

Related guides

• What Is a Home Lab? How to Start One in 2026
• Home Server Setup: Hardware, OS and Your First Services
• Home Lab Projects: 12 Ways to Use Virtualization
• What Is Proxmox VE? Why Home Labs Run It
• Install Hermes Agent on an Ubuntu Virtual Machine

Home Server Setup: Hardware, OS and Your First Services

A home server setup has five steps. Decide what the server will do, then pick hardware that fits that job. Pick an operating system, such as Proxmox VE, Ubuntu Server, TrueNAS or Unraid. Then add a few services, set up backups and make remote access safe.

Who this is for: you want one always-on machine for files, media, backups or self-hosted apps.

What is a home server?

A home server is a computer at home that runs all the time and provides services to other devices. Your phone backs up photos to it. Your TV streams movies from it. Your laptop saves files to it.

It is close to a home lab, but the goal is different. A home lab is for learning and breaking things. A home server is for services that you and your family depend on. Many people run both on one box. If you want the lab side, read what a home lab is and how to start one.

Step 1: decide what the server will do

Write down the jobs before you buy anything. The jobs decide the hardware and the OS. Common jobs are:

• File storage. A shared drive for the household.
• Photo and phone backup. A private copy of every photo.
• Media server. Movies, TV and music streamed to every screen.
• Network services. An ad blocker or DNS filter for the whole house.
• Home automation. A local hub for lights, sensors and heating.
• Self-hosted apps. A password manager, notes, a wiki or a recipe app.
• A test space. VMs for learning, kept apart from the services above.

Then ask two questions. How much storage do you need now, and in three years? Do you need to convert video on the fly (transcoding) for phones and TVs? Large storage needs drive bays. Transcoding needs a CPU with a good built-in graphics unit, or a separate GPU.

Step 2: pick hardware for your home server setup

There are four common classes of home server hardware. Each one fits a different set of jobs.

Old laptop

Free if you have one. It is quiet and uses little power. The limits are storage (usually one internal drive) and a single network port. It suits network services and a few small apps.

Old desktop

Free or cheap. It has room for several drives and more memory. It can be loud and use more power than you expect. It suits file storage, media and a first virtualization host.

Mini PC

Small, quiet and efficient. Many recent models have a CPU with good video transcoding. The limits are drive bays and network ports. Some have memory soldered to the board, so check the upgrade path before you buy. It suits apps, media and a small Proxmox host, with bulk storage kept on a separate box.

NAS box

A network-attached storage box is built for drives. It has several bays, hot-swap trays and its own OS. The CPU is often weak, so check it can run the apps you want. It suits storage first, with a few light apps.

Whatever you pick, check three things. The CPU must support hardware virtualization (Intel VT-x or AMD-V) if you plan to run VMs. You need a wired network port. And you need at least one SSD for the OS and apps.

Step 3: pick an operating system

The OS is the biggest choice in a home server setup. Four options cover most home servers. Check each vendor's site for current features and license terms as of October 2026.

OS	What it is	Cost (verify)	Best at	Watch for
Proxmox VE	Hypervisor for VMs and LXC containers	Free, optional paid subscription	Running many isolated services and test VMs	Storage sharing needs extra setup
Ubuntu Server	General Linux server OS	Free, optional paid support	Docker hosts, simple setups, learning Linux	No web UI by default, you use the command line
TrueNAS	Storage OS built on ZFS	Free community edition, paid enterprise options	Reliable file storage with snapshots	App and VM features are secondary to storage
Unraid	Storage and app OS	Paid license	Mixing drives of different sizes, easy apps	Not open source, license cost

Here is a simple rule. If you want many separate services and test VMs, start with Proxmox VE. Read what Proxmox VE is and why home labs run it first. If storage is the main job, start with TrueNAS or Unraid. If you only want Docker containers, start with Ubuntu Server.

Official docs: Proxmox VE, Ubuntu Server, TrueNAS and Unraid.

You can also combine them. A common pattern is Proxmox VE on the host, with a TrueNAS VM for storage and LXC containers for apps. It adds steps and risk. The TrueNAS hardware guide says a virtualized TrueNAS is not safe for critical data. If you try it, pass the real disks or the whole disk controller (HBA) to the VM, never virtual disks, and keep a separate backup.

Before you add services: network basics

A home server lives on your home network. Spend twenty minutes on the network now and you avoid hours of confusion later.

• Give the server a fixed IP address. Set a DHCP reservation in your router. Or set a static address on the server; if you do, pick an address outside the range your router hands out.
• Use a wired connection. Plug the server into the router or a switch. Wi-Fi drops and slows file copies.
• Name things. Give the server a short host name. Use the same naming pattern for every VM and container.
• Know your router. Find the admin page, the DHCP range and the DNS settings. You will need them for the ad blocker.
• Keep a fallback. If you point the whole house at a DNS server on your home server, keep a second DNS server set. Then the internet still works when the server reboots.

Later, you may want separate networks (VLANs) for the lab, smart home devices and guests. That needs a router and switch that support VLANs. It is a good second project, not a first one.

Step 4: your first five services

Do not install twenty apps on day one. Start with five that give real value. Add more once these run well.

1. A DNS ad blocker. Pi-hole or AdGuard Home blocks ads and trackers for every device on the network. It is small and teaches you about DNS.
2. File sharing. An SMB share for the household. On Proxmox, a small container can serve it. On TrueNAS or Unraid, it is built in.
3. Photo backup. An app such as Immich backs up phones automatically. Treat this data as precious and back it up twice.
4. A media server. Jellyfin or Plex streams your media library to TVs and phones.
5. A dashboard or monitoring tool. Uptime Kuma or a simple dashboard shows what runs and what is down.

Many of these run as Docker containers. Read the Docker docs before you start. On Proxmox, decide if each service runs in an LXC container or in a VM. The guide to containers vs virtual machines helps you choose. For a wider view of what "virtualization" covers, see the types of virtualization.

Step 5: backups and remote access

Backups

Follow the 3-2-1 rule. Keep three copies of important data. Store them on two different types of media or devices. Keep one copy off site.

• Use RAID or ZFS mirrors to survive a dead disk. Remember that RAID is not a backup. It does not protect you from deletion, ransomware or fire.
• Back up VMs and containers on a schedule. Proxmox VE has backup jobs built in.
• Copy the most important data off site. A cloud storage bucket or a drive at a relative's house both work.
• Restore a file every month to prove the backup works.

Remote access

You will want to reach your server from outside the house. Do it safely.

• Do not open ports on your router for admin pages. Never expose the Proxmox, TrueNAS or Unraid web interface to the internet.
• Use a VPN. Tailscale, or a WireGuard server (which needs one forwarded UDP port), lets your phone and laptop join your home network securely.
• Use strong, unique passwords and turn on two-factor login where the OS supports it.
• Keep the system updated. Set a regular time each month to apply updates.

Common mistakes

• Buying before deciding the jobs. You end up with the wrong drive count or a CPU that cannot transcode.
• Mixing the lab and the family services. A test VM that eats all the memory can take down the photo backup. Set memory limits, or use separate hosts.
• Treating RAID as a backup. It is not. Keep separate backup copies.
• Exposing admin pages to the internet. Use a VPN instead.
• No backup power. A small UPS protects the drives during a power outage and gives the server time to shut down cleanly.
• No notes. Write down IP addresses, drive layouts and what each service does. Future you will thank you.

FAQ

What is the best OS for a home server?

It depends on the main job. Proxmox VE suits many separate services and VMs. TrueNAS suits storage first. Unraid suits mixed drive sizes and easy apps. Ubuntu Server suits a simple Docker host.

Can I use an old PC as a home server?

Yes. An old desktop is one of the best ways to start. Add an SSD for the OS and check the power use. If it is loud or uses a lot of power, a mini PC may cost less to run over time.

How much does a home server cost to run?

It depends on the hardware and your power price. A low-power mini PC uses far less than a full tower with many drives. Measure your server with a plug-in power meter and multiply by your rate.

Is a NAS the same as a home server?

A NAS is a home server built mainly for storage. Most modern NAS boxes can also run apps. A general home server is often stronger at apps and VMs but needs more setup for storage.

Do I need a static IP from my internet provider?

No. A VPN service such as Tailscale works without one. Inside your home, give the server a fixed local IP address so your devices can always find it.

Should I run Docker or Proxmox?

You can run both. Proxmox VE is the host. Docker runs inside a VM or an LXC container on top of it. If you only ever want Docker containers, a plain Ubuntu Server is simpler.

Related guides

• What Is a Home Lab? How to Start One in 2026
• What Is Proxmox VE? Why Home Labs Run It
• Types of Virtualization: Server, Desktop, Storage, Network and More
• Containers vs Virtual Machines: When to Use Which
• Disadvantages of Server Virtualization (and How to Avoid Them)

What Is a Home Lab? How to Start One in 2026

A home lab is a small set of computers at home that you use to learn, test and run your own services. It can be one old laptop or a short rack of servers. Most home labs run a hypervisor, so one machine can host many virtual machines and containers. You can start one for zero dollars with hardware you already own.

Who this is for: you want to learn servers, networking or self-hosting, and you want a safe place to break things.

What is a home lab, in plain terms?

A home lab is a practice space for IT skills. It is like a workshop in a garage, but for software and networks. You build things, break them and fix them. Nothing you do there can take down a production system at work.

The word "homelab" (one word) means the same thing. People use both spellings.

Most home labs share three parts:

• A host. This is the physical computer. It can be a laptop, a desktop, a mini PC or a used server.
• A virtualization layer. This is the software that splits one host into many virtual machines (VMs) or containers. Proxmox VE is a common choice. If the term is new, read what a hypervisor is and how type 1 and type 2 differ.
• Workloads. These are the things you run: a file share, a media server, an ad blocker, a test domain controller, a Kubernetes cluster.

A home lab is not the same as a home server, but the two overlap. A home server runs services you depend on every day. A home lab is where you experiment. Many people start with a lab and end up with a server that the family uses. The home server setup guide covers that side in detail.

Why people build a home lab

People build home labs for four main reasons.

To learn job skills

Certifications and job interviews ask about Linux, networking, virtualization and cloud. Reading is not enough. You learn faster when you install a system, misconfigure it and then find the fault. A lab gives you that loop at no risk.

To self-host services

Some people want to own their data. They run their own photo backup, password manager, media server or home automation. A lab lets them host these services instead of paying a monthly fee to a cloud provider. You still pay for power and hardware, so do the math for your own case.

To test before production

If you run IT for a small business, a lab is a safe copy of your setup. You can test an update, a firewall rule or a backup restore before you touch the real thing.

Because it is fun

This is a real reason. Many people enjoy building systems. A lab is a hobby with useful side effects.

What does a home lab cost?

A home lab can cost nothing or a few hundred dollars to start. Some people spend much more, but you do not need to.

• Zero dollars: use the computer you already have. Run VMs inside a desktop app.
• Low cost: reuse an old desktop or laptop that sits in a closet. Add memory or an SSD if it needs it.
• A few hundred dollars: buy a used business mini PC or a new low-power mini PC. Many of these sell refurbished.

Prices for hardware change every month. I do not quote them here. Check current listings before you buy.

Do not forget the running cost. A machine that runs all day uses power all day. A small, low-power machine often costs less to run over a year than a big, cheap used server. Measure your own draw with a plug-in power meter.

Three ways to start a home lab setup

You do not need to pick the perfect path. Pick the one that matches what you have today. You can move later.

Path 1: nested on your laptop or desktop

You install a desktop hypervisor such as Oracle VirtualBox or VMware Workstation. Then you create VMs inside it. You can even install Proxmox VE inside a VM to learn it. This is called nested virtualization. This needs an x86 (Intel or AMD) computer. On an Apple Silicon Mac, Proxmox runs only under slow x86 emulation, so use a spare PC instead.

This path is free and fast. The limits are memory and uptime. Your VMs stop when you close the laptop. A laptop with 8 GB of memory gets tight quickly. 16 GB or more is more comfortable for two or three VMs.

To compare desktop tools, read VMware vs VirtualBox vs Proxmox for a desktop lab.

Path 2: an old PC as a dedicated host

You take an old desktop or laptop and wipe it. Then you install a type 1 hypervisor, such as Proxmox VE, directly on the hardware. The machine now runs all the time and you manage it from a web browser on another device.

This path costs little. Old desktops often have room for more drives and more memory. The trade-offs are noise, size and power use. Check that the CPU supports hardware virtualization (Intel VT-x or AMD-V) and that it is turned on in the firmware settings.

Path 3: a mini PC

You buy a small, low-power PC and install Proxmox VE on it. Mini PCs are quiet, small and cheap to run. Many people now start here.

The limits are expansion and network ports. Most mini PCs take one or two drives and have one or two network ports. Check the memory limit before you buy. Some models have soldered memory that you cannot upgrade.

Compare the three starting paths

Path	Start cost	Runs 24/7?	Best for	Main limit
Nested on your laptop	Zero	No	Learning the basics, short tests	Memory, and the VMs stop when the laptop sleeps
Old PC as a host	Zero to low	Yes	A first real host, more drives	Noise, size, power use
Mini PC	Low to a few hundred dollars	Yes	A quiet, always-on lab	Few drive bays, few network ports

What software runs a home lab?

Most home labs use a type 1 hypervisor on the host. A type 1 hypervisor runs directly on the hardware, not on top of Windows or macOS.

Proxmox VE is a popular choice for home labs. As of October 2026, it is free to use with an optional subscription. Verify on the Proxmox site. It runs both VMs and containers, and you manage it from a web page. Read what Proxmox VE is and why home labs run it for the full picture. The official documentation is at pve.proxmox.com.

Other options exist:

• VMware ESXi. Long the default in business. Licensing changed after Broadcom bought VMware. Check the current terms on the VMware site. Then see Proxmox vs VMware in 2026 before you choose it.
• Microsoft Hyper-V. Built into Windows Server and some editions of Windows. Good if you want to learn a Microsoft stack. See the Hyper-V docs.
• XCP-ng. An open-source hypervisor based on Xen. It has a smaller home lab community than Proxmox.
• A plain Linux server with Docker. No hypervisor at all. See the Docker docs. Good if you only want to run containers.

Containers and VMs solve different problems. If you are not sure which you need, read containers vs virtual machines.

Your first week with a home lab

Here is a simple plan. Do one step per evening.

1. Pick your path from the three above.
2. Install the hypervisor. For Proxmox VE, write the installer image to a USB stick and boot from it.
3. Give the host a fixed IP address on your home network. Write it down.
4. Log in to the web interface from another computer.
5. Create one Linux VM. Ubuntu Server or Debian is a good first choice.
6. Create one container and run a small service in it, such as an ad blocker for your network.
7. Set up a backup job for both. Then restore one to prove the backup works.

Step 7 is the one most people skip. Do not skip it. A backup you have never restored is only a hope.

Common mistakes

• Buying too much too soon. A rack of used servers looks great. It is also loud, hot and costly to run. Start small. Grow when you hit a real limit.
• Running the lab on the family network with no plan. One bad DHCP or DNS setting can break the internet for the whole house. Test network changes when nobody depends on the connection.
• Exposing services to the internet. Do not expose admin pages to the internet. Use Tailscale, or a WireGuard server (which needs one forwarded UDP port).
• No backups. Your lab will fail. Disks die and updates go wrong. Keep backups on a different disk, ideally a different machine.
• Not writing things down. Keep a simple text file of IP addresses, passwords (in a password manager) and what each VM does. In three months you will not remember.
• Ignoring the memory limit. Memory runs out before CPU in most labs. Check how much your host can take before you buy it.

FAQ

Is a home lab worth it?

For most people who want to learn IT or self-host, yes. You learn by doing, and a lab gives you a place to do it with no risk to real systems. If you only want one service, such as a media server, a single home server may be enough.

How much RAM do I need for a home lab?

Memory is usually the first limit you hit. 16 GB is a comfortable start for a few small VMs and containers. 32 GB or more gives you room for a test cluster or heavier services. Containers use far less memory than full VMs.

What is the difference between a home lab and a home server?

A home server runs services that you and your family depend on every day. A home lab is a place to experiment and break things. Many people run both on one machine, but it is safer to keep critical services away from your test VMs.

Can I run a home lab on a laptop?

Yes. You can run VMs inside a desktop hypervisor, or install Proxmox VE on an old laptop. A laptop has a built-in battery, which acts like a small backup power supply. But a battery held at full charge for months can swell, so check it. The limits are memory, storage and network ports.

Is Proxmox good for beginners?

Yes, with some patience. The installer is simple and the web interface covers most tasks. You will need basic Linux skills for some jobs. The official docs and forum are large and active.

Do I need a server rack?

No. Most home labs sit on a shelf or a desk. A rack makes sense when you have several machines, a network switch and a backup power supply to organize.

Related guides

• What Is Proxmox VE? Why Home Labs Run It
• Home Server Setup: Hardware, OS and Your First Services
• What Is a Hypervisor? Type 1 vs Type 2 Explained
• Proxmox vs VMware in 2026: For Home Labs and Small Teams
• Containers vs Virtual Machines: When to Use Which
• Install Hermes Agent on an Ubuntu Virtual Machine: Setup, Snapshots, and Recovery

Home Lab Projects: 12 Ways to Use Virtualization

The best home lab projects are small services you will actually use, built one at a time. Start with a network ad blocker, a VPN and backups. Then move to a media server, Home Assistant, monitoring, a Kubernetes cluster or a local AI model. Each project below says if it fits a VM or a container, how hard it is, and what it needs.

Updated October 2026. This guide replaces the 2009 version with current tools and a home lab focus.

Who this is for: you have a hypervisor running (or you plan to), and you want ideas for what to put on it.

Before you start a project

Every project here assumes one host with a type 1 hypervisor. Most examples use Proxmox VE, because it runs VMs and LXC containers from one web interface. If you do not have a host yet, read what a home lab is and how to start one first. If you are new to Proxmox, read what Proxmox VE is and why home labs run it.

Three rules make every project easier:

• One project at a time. Get it working, back it up, then start the next one.
• Write it down. Note the IP address, the VM or container ID and the login method for each service.
• Pick the right box. Small Linux services suit a container. Anything with its own OS, a custom kernel or untrusted code suits a VM. The guide to containers vs virtual machines explains the trade-off.

Home lab projects at a glance

"Difficulty" assumes you can use a Linux shell. "Resources" is relative to the other projects. Check each project's docs for real needs.

#	Project	Example tools	VM or container	Difficulty	Resources
1	Network ad blocker	Pi-hole, AdGuard Home	Container	Easy	Low
2	Smart home hub	Home Assistant OS	VM	Easy to medium	Low to medium
3	Monitoring	Uptime Kuma, Grafana	Container	Easy to medium	Low
4	Windows test VM	Windows evaluation media	VM	Easy	Medium
5	Reverse proxy	Nginx Proxy Manager, Caddy	Container	Medium	Low
6	VPN for remote access	WireGuard, Tailscale	Container or VM	Easy to medium	Low
7	Media server	Jellyfin	Container or VM	Medium	Medium, more with transcoding
8	NAS in a VM	TrueNAS	VM with passthrough	Hard	High (disks, memory)
9	Kubernetes lab	k3s	VMs	Hard	Medium to high
10	Local AI models	Ollama	VM	Medium	High (memory, GPU helps)
11	AI agent sandbox	Hermes Agent, OpenClaw	VM	Medium	Medium
12	Backups	Proxmox Backup Server	Separate machine or VM	Medium	Disk space

Easy home lab ideas to start with

1. Block ads for the whole network

Pi-hole and AdGuard Home are DNS servers that refuse to resolve ad and tracking domains. You point your router's DNS at them, and every device on the network gets the filter. That includes phones, smart TVs and game consoles. Both are open source as of October 2026.

A small unprivileged LXC container is enough. Give it a fixed IP address. Do not add a public DNS server as a second entry in your router. Devices may use it at random and skip the filter. For uptime, run a second Pi-hole or AdGuard Home instance on another host, and set both in the router.

2. Run Home Assistant in a VM

Home Assistant is a local hub for lights, sensors and heating. The full Home Assistant OS includes its add-on system. The Home Assistant project publishes a ready-made disk image for KVM, which is the hypervisor Proxmox uses. See the Home Assistant alternative install page for the current image and steps.

Use a VM, not a container, for Home Assistant OS. It is a full operating system with its own updates. If you use a USB radio stick (Zigbee or Z-Wave), pass the USB device through to the VM.

3. Set up monitoring

Uptime Kuma is a self-hosted tool that checks if your services respond and alerts you when one goes down. It runs as a single Docker container. Point it at every other project on this list.

When you want graphs, add Grafana. Grafana draws dashboards from data sources such as Prometheus or InfluxDB. Proxmox can send host metrics to an external metric server, which Grafana can then chart.

4. Build a Windows test VM

A Windows VM is a safe place to test software, scripts and updates. Microsoft offers evaluation versions of Windows and Windows Server on its Evaluation Center. Check the license terms of the version you download.

On Proxmox, attach the VirtIO driver ISO when you install Windows. Windows needs those drivers for fast disk and network access. Take a snapshot right after setup, so you can roll back after each test.

Network and remote access projects

5. Put a reverse proxy in front of your apps

A reverse proxy gives each service a name, such as jellyfin.home.example, instead of an IP address and port. It also handles HTTPS. Nginx Proxy Manager gives you a web interface and built-in Let's Encrypt support, and it runs as a Docker container. Caddy is a single program with a short config file. By default, it gets and renews TLS certificates on its own.

Keep the proxy on your home network at first. Do not forward router ports to it until you understand what you expose.

6. Reach your lab from anywhere with a VPN

A VPN lets your phone and laptop join your home network from outside. It is the safe way to reach admin pages. Two common choices:

• WireGuard. A fast, simple VPN protocol that runs over UDP. You run a WireGuard server and forward one UDP port on your router.
• Tailscale. A service built on WireGuard that connects your devices with no port forwarding. As of October 2026, it has a free Personal plan. Check the current limits on the Tailscale pricing page.

Never expose the Proxmox web interface to the internet. Use the VPN. The home server setup guide covers remote access safety and the first services in more detail.

Storage and media projects

7. Run a media server

Jellyfin streams your movies, TV and music to TVs, phones and browsers. As of October 2026, the Jellyfin project says the server and official clients are free, with no paid tier.

The hard part is transcoding. When a device cannot play a file as it is, the server converts it on the fly. The CPU can do it in software, but that is heavy. A built-in graphics unit (such as Intel Quick Sync) or a passed-through GPU makes it practical. A Docker container in a VM is a common setup. Mount your media from a NAS share, not from inside the container.

8. Build a NAS in a VM

As of October 2026, TrueNAS Community Edition is a free, open-source storage OS built on ZFS. Running it in a VM on Proxmox is a popular project. It is also the one most likely to lose data if you set it up wrong.

Pass the whole disk controller (HBA) through to the TrueNAS VM. Do not pass single disks or virtual disks. ZFS needs to see the raw disks and their health data. If you cannot pass through a controller, use ZFS on the Proxmox host instead, or put TrueNAS on its own machine.

Learning projects for your career

9. Build a Kubernetes lab with k3s

k3s is a lightweight, certified Kubernetes distribution. It ships as a single binary. As of October 2026, it is a CNCF sandbox project. Its own docs list home labs as a use case.

Create three small Linux VMs on your host. Install k3s on one as the server, and join the other two as agents. You now have a multi-node cluster to break and rebuild. Use VMs, not LXC containers, because Kubernetes needs kernel features that are hard to give to a container. The k3s documentation has the install steps.

10. Run a large language model locally

Ollama runs open large language models on your own hardware. On Linux, the official install script sets it up as a systemd service. See the Ollama Linux docs for the current command. Then you pull a model and chat with it, or call it from other apps over a local API.

This project is about memory. Small models run on a CPU, slowly. Larger models need more memory, and a GPU makes them much faster. In a VM, that means GPU passthrough, which is a project of its own. Start with a small model in a CPU-only VM.

11. Sandbox an AI agent in a VM

Treat an AI agent like untrusted software with your keyboard. Give it its own VM, and keep your real files out of reach. A bad command then breaks the VM, not your host. A snapshot lets you roll back in seconds.

Two guides on this site walk through it step by step. Read how to install Hermes Agent on an Ubuntu virtual machine, or install OpenClaw in an Ubuntu VM with Docker Compose. If you cannot choose, see Hermes vs OpenClaw on virtual machines.

12. The project that protects the rest: backups

Do this project early, not last. Proxmox VE has backup jobs built in. Proxmox Backup Server adds incremental backups, deduplication, compression and encryption. As of October 2026, it is open source under the GNU AGPLv3, with optional paid subscriptions. See the Proxmox Backup Server documentation.

Put the backup server on a different machine from the host it protects. A backup on the same disks dies with them. An old PC or a small mini PC with a large drive works. Then follow the 3-2-1 rule: three copies, two kinds of media, one copy off site.

Most important: restore a VM from backup and boot it. Do this once now and once a month after. An untested backup is a hope, not a plan.

Which order should you build them in?

Here is a sensible path for a new lab:

1. Week 1: ad blocker and backups. You learn containers, DNS and restores.
2. Week 2: VPN and monitoring. You can now reach the lab safely and see when things break.
3. Week 3: reverse proxy and one "fun" service (media or Home Assistant).
4. Later: NAS, Kubernetes, local AI and agents, as your hardware and interest allow.

In my view, the order matters more than the tool choice. Backups and remote access first make every later mistake cheap.

Common mistakes

• Running the family's DNS on a test box. If the ad blocker goes down, the whole house loses the internet. Run a second instance on another host, or know how to point the router back to your provider's DNS in one step.
• Exposing admin pages. Use a VPN, not port forwarding, for Proxmox, TrueNAS and Home Assistant admin pages.
• Virtual disks under TrueNAS. ZFS on top of virtual disks hides disk health and can hide errors. Pass through the controller.
• Backups on the same host. One dead machine then takes the data and its backup.
• Ignoring memory. Memory runs out before CPU in most labs. Local AI and NAS projects use the most.

FAQ

What is the best first home lab project?

A network ad blocker such as Pi-hole or AdGuard Home. It is small, it runs in one container, and you see the result on every device. It also teaches you DNS, which every later project needs.

Do home lab projects need a powerful server?

No. Most projects on this list run on a mini PC or an old desktop. Only a NAS with many disks, a Kubernetes cluster and local AI models need more memory or a GPU.

Should each project get its own VM?

Not always. Small Linux services fit in containers and use less memory. Use a VM for a full OS such as Home Assistant OS, TrueNAS or Windows, and for anything you do not trust.

Can I run these projects without Proxmox?

Yes. Most of these tools run on any Linux host with Docker. A hypervisor such as Proxmox adds VMs, snapshots and backups in one place, which helps as the lab grows.

Is it safe to expose home lab services to the internet?

Only with care. Use a VPN such as WireGuard or Tailscale to reach admin pages. If you must publish a service, put it behind a reverse proxy with HTTPS, keep it patched and never publish the hypervisor itself.

Related guides

• What Is a Home Lab? How to Start One in 2026
• Home Server Setup: Hardware, OS and Your First Services
• What Is Proxmox VE? Why Home Labs Run It
• Install Hermes Agent on an Ubuntu Virtual Machine: Setup, Snapshots, and Recovery
• Containers vs Virtual Machines: When to Use Which