What's installed on your Ollama server?
- Ollama 0.40, which downloads and runs the models
- Open WebUI 0.12 for browser chat and conversation history
- The gemma3:1b model, downloaded during install and tested with a real prompt
- Memory settings: one model and one request at a time, held in memory for 30 minutes after the last request
- An admin account ready to use, with sign-up closed
- HTTPS through Caddy on an sslip.io address based on your IP
- Firewall with ports 22, 80 and 443 open; Ollama's port 11434 stays closed to the outside
What size do you need for Ollama?
Ollama loads the model into memory, so its file size is the floor for your RAM, plus headroom. As a rule of thumb, start with 4 GB for models of 1 to 3 billion parameters, 8 GB for 7 to 8 billion and 16 GB for 12 to 14 billion. These aren't official figures. Plan disk space too: in Ollama's library, gemma3:1b is 806 MB, gemma3:4b is 3.4 GB and gemma3:12b is 8.2 GB. When you move to a bigger model, raise the RAM on the Size tab of your panel. The server restarts once, and your data and IP stay the same.
| Use | Size | Price | Choose |
|---|---|---|---|
| Starter | 4 GB RAM · 2 vCore · 60 GB NVMe | $6.89/mo | |
| Medium | 8 GB RAM · 4 vCore · 120 GB NVMe | $13.62/mo | |
| Powerful | 16 GB RAM · 8 vCore · 200 GB NVMe | $25.15/mo |
Who is it for?
- People who want their chats to stay on their own server
- Short summaries, classification and email drafts handled by a model you control
- Developers trying open models or wiring them into small automations
- A personal assistant for you and perhaps a few other people
What happens after you order?
- Set the size and complete your payment.
- Your server starts, then Ollama and Open WebUI install. A model is downloaded too, so this takes a little longer.
- Open your server from My servers in the panel. On the Overview tab, wait until the App row says it's ready, then click Open app.
- Log in with the details from the Email and Password rows. There's no setup screen, and gemma3:1b is already selected in the chat.
- For a larger model, scale up the RAM first, then type the model's name into the model selector and download it.
With every server
- Daily backups included in the price
- Unlimited traffic
- Full root access
- Scale RAM and CPU up or down later, and grow the disk
- Location: Istanbul · IST-1.
Good to know
- The server has no graphics card. The first request loads the model from disk, long prompts delay the first reply, answers can arrive slower than you read, and several users at once split the speed.
- It's not suited to live multi-user services, real-time replies, heavy code generation or large models. A 32 billion parameter model is very slow on a CPU.
- Small models tend to write Turkish less well than English, so test with your own text before you decide.
- Every model has its own license, and gemma3 comes under the Gemma Terms of Use. Read the license on the model page before commercial use.
- Without an enterprise license, Open WebUI's terms don't let you remove its name and logo once a deployment passes 50 users in 30 days.
- Updating Ollama, Open WebUI and your models is up to you. The address in the Email row is only a login and doesn't receive mail.
Ollama questions and answers
How much RAM does Ollama need?
A model's file size is the minimum, and conversation memory and the operating system come on top. As a rule of thumb, start with 4 GB for 1 to 3 billion parameters, 8 GB for 7 to 8 billion and 16 GB for 12 to 14 billion. Run free -h once the model has loaded to see what it really uses.
Does Ollama work without a GPU?
Yes. With no GPU, Ollama loads the model into system memory and runs it on the CPU, and ollama ps shows 100% CPU. Seturack servers have no graphics card. Small models are usable for short tasks, while large ones run but stay very slow.
Which model should I start with on an Ollama VPS?
Try the interface with the installed gemma3:1b, then move to a 3 billion parameter model and test it on your own work. If the answers fall short, go up a size and check with free -h that the RAM is enough. Ollama's library lists each model's file size before you download it.
How do I download a new model in Ollama?
Type the model's name into the model selector on the chat screen and Open WebUI offers to download it. You can also connect over SSH and run docker compose exec ollama ollama pull model-name in /opt/seturack/ollama. Check your free disk space first.
Can my app connect to the Ollama API?
Yes, though port 11434 is closed to the internet by default, because the Ollama API has no authentication out of the box. With root access you can publish the port on the local address in the compose file and connect through an SSH tunnel, or run your app on the same server.
Do my AI chats leave the server?
The model runs on your server, so your chats and models stay there. Ollama's documentation says it doesn't see your prompts or data when you run locally. In this install, Open WebUI's connection to hosted providers and its usage reporting are both turned off.