Skip to main content
Ollama runs open language models on your own server: the data does not go to the cloud, and you need no API keys or subscriptions. You talk to a model in the console or over a local API.

​
Hardware

The servers have no GPU: the model computes on the CPU and loads into RAM in full. If memory runs out, the OOM-killer kills the process.
  • 1B–3B (llama3.2, gemma3:1b) – CPU and a few GB of RAM, answers, but not instantly.
  • 7B–8B – CPU and lots of RAM, noticeably slower, fine for experiments.
  • 13B and up – needs a GPU, not an option on a VPS.
Pick a model that fits the server and keep memory in reserve. On a small plan a swap file helps, but it is slow: a safety net, not a replacement for RAM. Memory is short all the time – take a bigger plan: Upgrade on the server card.

​
Installation

Connect over SSH as root: Connection. The syntax and the model list change, check ollama.com and github.com/ollama/ollama.
  1. Ollama – the official script installs the engine and a background service:
  2. Model – it downloads and opens a chat right away:
    Without a tag it pulls the 3B variant, about 2 GB. On 2–4 GB of RAM (VPS-1, VPS-2) take the light one explicitly:
    The :1b and :3b tags set the model size. The first run downloads the model, from hundreds of MB to several GB, after that it loads from disk. To leave the chat: /bye.

​
Models

  • ollama pull gemma3 – download a model ahead of time. Models and their sizes: ollama.com/library.
  • ollama list – downloaded models.
  • ollama rm gemma3 – remove a model and free up disk space.

​
API

Ollama listens on a local API at http://localhost:11434. A request from your code:
Need a chat in the browser – install Open WebUI in a Docker container, it runs on top of Ollama. How to install Docker: Docker.
Do not expose port 11434 to the internet: anyone can load an open API at your server’s expense. By default the API is reachable only locally. Need outside access – close the port with the firewall and let traffic in only through a proxy with authentication.
Network or port not working – write to @lumisup_robot, software setup is on you. Next: Swap file, Docker.