How to Install Ollama on a VPS: Your Own AI Server
With Ollama you run AI language models such as Llama, Mistral and Qwen on your own VPS. You install Ollama with one command, download a model and talk to it via the terminal, an API or a chat interface. Your data stays on your own server in Amsterdam and you do not pay per request. Ideal for privacy-sensitive applications and automations with n8n, for example.
How much memory do you need?
On a VPS without a graphics card, small models run best. Count on about 8 GB of RAM for a model of up to 7 billion parameters, and 16 GB for larger models. An NVMe VPS loads models quickly from disk.
Step 1: install Ollama
curl -fsSL https://ollama.com/install.sh | sh Step 2: download and start a model
ollama pull llama3.2
ollama run llama3.2 Choose another model on ollama.com/library if you want. Close the chat with /bye.
Step 3: add a chat interface with Open WebUI
Open WebUI gives you a chat environment in the browser. You install it with Docker:
docker run -d -p 127.0.0.1:3000:8080 --add-host=host.docker.internal:host-gateway -v open-webui:/app/backend/data --name open-webui --restart always ghcr.io/open-webui/open-webui:main Make the interface reachable via your domain with an Nginx reverse proxy and HTTPS.
Using Ollama via the API
curl http://localhost:11434/api/generate -d '{"model": "llama3.2", "prompt": "Wat is een VPS?"}' Secure access
The Ollama API has no password. So never open port 11434 to the internet. Use an SSH tunnel or a reverse proxy with login if you want to use the API from outside.
Related articles
Fast NVMe VPS for your apps
NVMe storage with high IOPS and low latency for Docker, databases, AI models and self-hosted apps, in the data centre in Amsterdam.