Run your own AI tools on a Noiz Cloud server

You can run open-source AI models on a server of your own instead of sending your prompts to an AI company. This guide sets up Ollama, which runs the models, and Open WebUI, a chat interface in your browser, on a self-managed Noiz Cloud server, reachable only by you through an encrypted SSH tunnel. It is for someone comfortable on a Linux command line who wants a private assistant for documents and drafts, or a model their own applications and agents can call.

Last reviewed: 19 October 2026, against the current Ollama and Open WebUI documentation and the Noiz Cloud servers as listed on our website.

Official Documentation Reference

• Ollama: Linux: installing Ollama and running it as a service.
• Ollama: FAQ: where it listens and how to change it.
• Open WebUI: Quick start: installing Open WebUI and creating the administrator account.

Before you start

• A Noiz Cloud server running Ubuntu or Debian, with your SSH key and a firewall set up: Initial server setup with Ubuntu on a Noiz VPS and How to set up a firewall with UFW on Ubuntu and Debian.
• Enough memory for the model you want. Models run on the server's processors. Small models of a few gigabytes fit on the larger servers and answer slowly but usefully. Choose the server by the model's download size plus room for the system.
• Docker on the server for Open WebUI, installed from Docker's own instructions for your distribution.

What you are taking on

A Cloud server is self-managed. We keep the hardware, the platform and the network healthy; the operating system and everything on it are yours: updates, security, backups and the software you install. Two facts from the vendors' documentation shape every step below:

• Ollama has no login. Its documentation says the local API does not require authentication. Anyone who can reach it can use it.
• Open WebUI's first account is the administrator. Whoever signs up first controls the instance.

So both stay off the public internet. You reach them through an SSH tunnel from your own computer.

Step 1: Install Ollama

1. Sign in to the server over SSH as your normal user.
2. Install Ollama with the method on Ollama's Linux page. Read the install script before you run it, or use the manual install described on the same page.
3. Check that it runs: ollama -v.
4. Download a small model to start, for example one of the models in Ollama's library of a few gigabytes: ollama pull <model-name>.
5. Test it: ollama run <model-name> "Summarise the benefits of a static website in three lines."

Ollama listens on 127.0.0.1 port 11434 by default, which means only programs on the server itself can reach it. Leave that setting as it is.

Step 2: Install Open WebUI, bound to the server only

Run Open WebUI in Docker, published on the server's own loopback address so that it is not reachable from outside:

docker run -d -p 127.0.0.1:3000:8080 \
  --add-host=host.docker.internal:host-gateway \
  -v open-webui:/app/backend/data \
  -e WEBUI_SECRET_KEY="a-long-random-value" \
  --name open-webui --restart always \
  ghcr.io/open-webui/open-webui:main

Replace the secret with a long random value from your password manager. The 127.0.0.1:3000 part publishes the interface on the server's own loopback address only, so it cannot be reached from the internet.

Step 3: Reach it through an SSH tunnel

1. On your own computer, open a tunnel:

ssh -L 3000:127.0.0.1:3000 your-user@your-server-ip

2. Leave that window open, and visit http://localhost:3000 in your browser.
3. Create your account straight away. The first account is the administrator.
4. In Open WebUI, open Settings > Admin > Connections and check that the Ollama connection points at http://host.docker.internal:11434. Your model should appear in the model list.

New sign-ups go to a pending state until the administrator approves them, which is the setting you want. Add other people under Admin Panel > Users only when you mean to.

Step 4: Use it

• Private drafting and summaries. Paste a document you would not put into a public AI service and ask for a summary. The text stays on your server.
• An API for your own tools. Applications and scripts on the same server can call Ollama at http://127.0.0.1:11434.
• Long-running agents. A script or agent that should run all the time belongs in a system service, so that it restarts after a reboot. Ollama's Linux page shows the pattern for Ollama itself.

Only install Open WebUI tools, functions and pipelines you have checked. Its documentation warns that they run arbitrary Python code on your server.

Keep it healthy

• Updates. Update the operating system, Ollama (the install method updates it) and the Open WebUI container regularly.
• Backups. Cloud servers do not include backups. Use the server's snapshots before big changes, and keep a copy of the open-webui data volume elsewhere, or see Noiz Backup on the security store page.
• Memory. If answers stop or the server slows, the model is too big for the memory. Use a smaller model or a larger server.

Troubleshooting

• The browser cannot reach localhost:3000. The SSH tunnel is closed or the container is not running. Reopen the tunnel and check docker ps.
• Open WebUI shows no models. The Ollama connection address is wrong, or Ollama is not running. Check Settings > Admin > Connections and ollama -v on the server.
• Answers are very slow. Processor-only servers are slow on large models. Try a smaller model, and shorten your prompts.
• Someone else created the first account. The interface was reachable from outside. Remove the container and its volume, check that you used 127.0.0.1:3000, and start again.

Choosing a server, and what is coming

Noiz Cloud servers are listed with their cores, memory, disk and transfer on the Cloud page. GPU servers and an OpenStack API for managing your own cloud resources are coming, with no dates yet. If you are planning an AI workload and want advice on the size it needs, get in touch.