Free guide Checked November 17, 2026. About 15 minutes to read, about an hour to do. Commands are from each project’s official instructions; check their sites for updates.
What you will have at the end
- A computer in a closet or corner that runs AI models all day.
- A chat page you open from any phone, tablet or laptop on your home Wi-Fi, with a separate login for each person.
- Optionally, safe access when you are away, without opening your home to the internet.
We call this your “AI brain”: one place in your home where your AI lives. Instead of every device paying for its own subscription, they all talk to the same private brain.
How it works, in one picture
There are only three pieces. The model is the AI itself, a big file you download once. Ollama is the engine that loads the model onto your graphics card and answers questions. Open WebUI is the friendly chat page; it sends your questions to Ollama and shows the answers. Your phone talks to Open WebUI over your home Wi-Fi. Nothing has to leave your house.
What you need
- A PC with a graphics card (see Build my PC). An older gaming PC is a great start.
- About one hour, and a USB stick if you are installing Linux.
- A network cable to the router if you can; it is more reliable than Wi-Fi for a server.
- No coding. You will copy and paste a few commands, and we explain each one.
Step by step
- Pick the computer. Any PC with a graphics card works; see Build my PC. An older gaming PC is a great start. Connect it to your router with a network cable if you can.
- Install the operating system. Ubuntu Server 24.04 LTS (free) is the classic choice for a home server. Keeping Windows 11 also works if you prefer it; the steps are the same, only the installers differ.
- Install the graphics driver. On Ubuntu with NVIDIA, run
sudo ubuntu-drivers install, then restart. Typenvidia-smito confirm the card and its memory show up. On Windows, install the driver from nvidia.com or amd.com. - Install Ollama, the part that runs models. On Linux:
curl -fsSL https://ollama.com/install.sh | sh. On Windows or Mac, use the installer from ollama.com/download. Ollama then runs quietly in the background. - Download a first model. Run
ollama pull llama3.2, thenollama run llama3.2and say hello. Browse other models at ollama.com/library and pick sizes that fit your VRAM. - Install Docker. Docker runs apps in tidy boxes so they do not mess up your system. Follow the official guide for your system at docs.docker.com.
- Add the chat page (Open WebUI). Run the official command shown below. It starts a ChatGPT-style page on port 3000 that talks to Ollama on the same machine.
- Open it from your phone. On the same Wi-Fi, visit
http://YOUR-SERVER-IP:3000. The first account you create becomes the administrator. Create accounts for family members from the admin settings. - Reach it safely when you are away. Use a private network app such as Tailscale. Do not open ports on your router to the whole internet.
- Keep it healthy. Turn on automatic security updates, back up the Open WebUI data, and consider a small battery backup (UPS) so a power blip does not corrupt anything.
Choosing models for your AI brain
You can keep several models on the server and switch between them in Open WebUI. A good family setup is three: one small, fast model as the default for everyday questions, one larger model for harder writing and reasoning, and, if you like, one model made for code or for reading pictures. Each model only uses graphics memory while it is loaded, and Ollama unloads models that have not been used for a while, so keeping several on disk is fine.
Look at each model’s page in the Ollama library. It shows the sizes available (for example 3B, 8B or 70B) and the download size. Use the VRAM table in Build my PC: if the download is bigger than your card’s memory, it will be slow. Also check the licence on the model page; most open models are free for personal use, and some have extra rules for businesses.
When a new model comes out, try it side by side with your current default on a few questions your family really asks. Keep the one that answers best at a speed people are happy with. Faster usually wins at home.
The chat page command (Open WebUI)
This is the official Docker command from Open WebUI’s quick start. First make a secret key by running openssl rand -hex 32, and paste the result where it says your-secret-key. Keep that key; using the same one later stops everyone being logged out after updates.
docker run -d -p 3000:8080 --add-host=host.docker.internal:host-gateway \
-v open-webui:/app/backend/data -e WEBUI_SECRET_KEY=your-secret-key \
--name open-webui --restart always ghcr.io/open-webui/open-webui:main
What the pieces mean: -p 3000:8080 puts the page on port 3000. -v open-webui:... keeps your chats and settings safe across updates. --add-host... lets the page reach Ollama on the same computer. --restart always brings it back after a reboot.
Set it up for the family
- Create your admin account first. The first person to sign up is the administrator, so do it yourself right after installing.
- Add an account for each person. Separate accounts keep everyone’s chats private from each other.
- Pick a default model. Choose one that is fast on your card, so kids and grandparents get quick answers.
- Add it to phones’ home screens. Open the page in the phone browser and use “Add to Home Screen”. It then opens like an app.
- Write a one-line “house rule” in the system prompt, for example: “Answer simply and kindly. If you are not sure, say so.”
Keep it safe
- Never open router ports to the internet for this. Use a private network app like Tailscale when you are away.
- Update monthly. Pull the new Open WebUI image and recreate the container with the same command and the same secret key; your data stays in the volume.
- Back up before updates. The Open WebUI data volume holds every chat and account.
- Turn on only the plugins you use. Web search and other tools can send data out of your home.
Troubleshooting
| What you see | Most likely cause | What to do |
|---|---|---|
| Open WebUI loads but shows no models | Open WebUI cannot reach Ollama | Make sure Ollama is running (ollama list should work). The Docker command must include --add-host=host.docker.internal:host-gateway. |
| My phone cannot open the page | Wrong address, or the server firewall blocks port 3000 | Use the server’s local IP address (for example 192.168.1.20), not “localhost”. Allow port 3000 on the server’s firewall for your home network only. |
| Everyone gets logged out after an update | No fixed secret key | Set WEBUI_SECRET_KEY once, as in the official command, and keep using the same value. |
| Answers are very slow | Model too big for the graphics card | Run ollama ps; it should say 100% GPU. If not, use a smaller model. |
| Answers forget the start of a long chat | The context window is full | Ollama’s documentation says the default context is 4096 tokens; it can be raised with OLLAMA_CONTEXT_LENGTH if you have spare VRAM. |
| Other computers cannot reach Ollama directly | Ollama only listens on the server itself by default | Only if you need it: set OLLAMA_HOST=0.0.0.0:11434 as the Ollama FAQ describes, and keep it inside your home network. |
Questions people ask
What is the difference between Ollama and Open WebUI?
Ollama is the engine: it loads models and answers. Open WebUI is the friendly page you type into. You can use Ollama alone from a terminal, but most families prefer the page.
Can the whole family use it at the same time?
Yes. One graphics card answers one request at a time quickly, so if several people ask at once, some will wait a little longer.
Is it private?
Your chats stay on your server. Anything you connect to the internet, such as web search or plugins, can send data out, so turn on only what you need.
Does it need the internet?
Only to download the software and models. After that, chatting works on your home network with the internet unplugged.
How do I add more models?
Run ollama pull MODEL-NAME on the server, or use the model settings in Open WebUI. Delete ones you do not use with ollama rm MODEL-NAME to save disk space.
How much electricity does it use?
It depends on your parts and how much you use it. A graphics card uses far more power while answering than while idle. Check your power bill after the first month, and let the server sleep at night if nobody uses it.
Can I use it from outside my home?
Yes, safely, with a private network app like Tailscale that only your own devices can join. Opening router ports to the public internet is the risky way; we do not recommend it.
Do I need Linux?
No. Windows 11 works too. Linux is popular for servers because it runs for months without restarts and uses less memory.
Sources
- Open WebUI quick start (Docker command, WEBUI_SECRET_KEY, image variants)
- Ollama FAQ (context length, OLLAMA_HOST, ollama ps)
- Ollama install script
- Ollama model library
- Docker Engine install guide
- Tailscale
Next: give your AI brain hands, so it can do tasks, not only chat.
Build my own agent →← Build my PC