How to Run Ollama on a VPS: Self-Host Llama 3 in 2026

Published

How to Run Ollama on a VPS: Self-Host Llama 3 in 2026

Running a local LLM on your own VPS gives you a private, cost-predictable AI API. No per-token pricing, no data leaving your infrastructure. The catch: you need enough RAM, and the setup requires a few extra steps compared to a cloud API.

This guide covers the full setup: installing Ollama, pulling a model, keeping it running with systemd, and exposing a secure API through Nginx.

For VPS recommendations and RAM requirements by model size, see Best VPS for Ollama in 2026.

Requirements

What Minimum Recommended
RAM 8GB 16GB
Storage 20GB free 40GB+
CPU 2 vCPU 4+ vCPU
OS Ubuntu 22.04 Ubuntu 22.04

Good options: Hetzner CX32 (€8.30/mo, 8GB RAM — see the full CX32 pricing breakdown) or Contabo Cloud VPS S ($7/mo, 8GB RAM). A 4GB VPS will fail to load most useful models.

Step 1: Install Ollama

curl -fsSL https://ollama.com/install.sh | sh

This installs the ollama binary and creates a systemd service automatically. Verify:

ollama --version
systemctl status ollama

Step 2: Pull a Model

# Llama 3.1 8B — best balance of quality and speed on CPU
ollama pull llama3.1

# Mistral 7B — fast, good for code and structured output
ollama pull mistral

# Phi-3 Mini — runs on 4GB RAM, fast but less capable
ollama pull phi3

Model download sizes:

  • llama3.1 (8B) — ~4.7GB
  • mistral (7B) — ~4.1GB
  • phi3 (3.8B) — ~2.3GB

Models are stored in ~/.ollama/models. Make sure your VPS has enough disk space before pulling.

Step 3: Test Locally

# Chat in terminal
ollama run llama3.1

# Or call the REST API directly
curl http://localhost:11434/api/generate -d '{
  "model": "llama3.1",
  "prompt": "What is the capital of France?",
  "stream": false
}'

The API is OpenAI-compatible for /api/chat calls, making it easy to swap with existing integrations.

Step 4: Keep Ollama Running (systemd)

Ollama’s installer creates a systemd service automatically, but verify it’s set to start on reboot:

sudo systemctl enable ollama
sudo systemctl start ollama
sudo systemctl status ollama

If the service isn’t created (older installs), create it manually:

sudo nano /etc/systemd/system/ollama.service
[Unit]
Description=Ollama Service
After=network-online.target

[Service]
ExecStart=/usr/local/bin/ollama serve
User=ollama
Group=ollama
Restart=always
RestartSec=3
Environment="OLLAMA_HOST=127.0.0.1:11434"

[Install]
WantedBy=default.target
sudo systemctl daemon-reload
sudo systemctl enable --now ollama

The OLLAMA_HOST=127.0.0.1 binding is important — it keeps Ollama on localhost only. Nginx handles external access in the next step.

Step 5: Nginx Reverse Proxy with API Key Auth

Never expose port 11434 directly to the internet. Use Nginx as a gatekeeper.

sudo apt install nginx -y
sudo nano /etc/nginx/sites-available/ollama
server {
    listen 443 ssl;
    server_name your-domain.com;

    ssl_certificate /etc/letsencrypt/live/your-domain.com/fullchain.pem;
    ssl_certificate_key /etc/letsencrypt/live/your-domain.com/privkey.pem;

    # Simple API key check via header
    if ($http_x_api_key != "your-secret-key-here") {
        return 401;
    }

    location / {
        proxy_pass http://127.0.0.1:11434;
        proxy_set_header Host $host;
        proxy_read_timeout 300s;   # LLM responses can be slow
        proxy_send_timeout 300s;
    }
}
sudo ln -s /etc/nginx/sites-available/ollama /etc/nginx/sites-enabled/
sudo certbot --nginx -d your-domain.com
sudo nginx -t && sudo systemctl reload nginx

Now call your API with the key:

curl https://your-domain.com/api/generate \
  -H "X-API-Key: your-secret-key-here" \
  -d '{"model": "llama3.1", "prompt": "Hello", "stream": false}'

Step 6: Use with n8n or Custom Scripts

Ollama’s API is compatible with the OpenAI SDK. In Node.js:

import OpenAI from 'openai';

const client = new OpenAI({
  baseURL: 'https://your-domain.com/v1',
  apiKey: 'your-secret-key-here',
});

const response = await client.chat.completions.create({
  model: 'llama3.1',
  messages: [{ role: 'user', content: 'Explain PM2 cluster mode' }],
});

For n8n self-hosted workflows, use the Ollama node or the HTTP Request node pointing to your Nginx endpoint.

Performance Tips

  • Preload models at startup to avoid cold-start latency:

    # Add to /etc/systemd/system/ollama.service [Service] section:
ExecStartPost=/bin/sh -c 'sleep 5 && ollama run llama3.1 ""'

  • Limit context length for faster responses on CPU:

    ollama run llama3.1 --num-ctx 2048

  • Monitor RAM usage — if the VPS starts swapping heavily, the model is too large. Switch to phi3 or upgrade to 16GB RAM.

What’s Next

Self-hosted Ollama pairs well with n8n for AI automation workflows. See our n8n on VPS setup guide for the full n8n + Ollama stack on a single VPS.

For VPS provider comparisons at the 8GB RAM tier (what Ollama needs), see Best VPS for Ollama.