Self-Host Your Own AI Coding Assistant with Ollama and Continue

What You Need

• VS Code
• Ollama running on a VPS or locally
• The Continue VS Code extension

Step 1: Install Ollama on Your VPS

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

Step 2: Pull a Code Model

Qwen 2.5 Coder is a strong, widely used local coding model:

# For 8 GB RAM
ollama pull qwen2.5-coder:7b

# For 16 GB RAM (better quality)
ollama pull qwen2.5-coder:14b

For completions (faster, smaller):

ollama pull qwen2.5-coder:1.5b

Step 3: Expose Ollama Externally

By default Ollama only listens on localhost. To access it from VS Code on your laptop:

cat > /etc/systemd/system/ollama.service.d/override.conf << EOF
[Service]
Environment="OLLAMA_HOST=0.0.0.0"
EOF
systemctl daemon-reload && systemctl restart ollama

Protect the port, only allow your IP:

ufw allow from YOUR_HOME_IP to any port 11434
ufw deny 11434

Or set up an SSH tunnel instead:

ssh -L 11434:localhost:11434 user@your-vps-ip

The tunnel keeps traffic encrypted and avoids exposing the port publicly.

Step 4: Install and Configure Continue

Install the Continue extension from the VS Code marketplace.

Open Continue settings (~/.continue/config.json):

{
  "models": [
    {
      "title": "Qwen Coder 7B",
      "provider": "ollama",
      "model": "qwen2.5-coder:7b",
      "apiBase": "http://localhost:11434"
    }
  ],
  "tabAutocompleteModel": {
    "title": "Qwen Coder 1.5B",
    "provider": "ollama",
    "model": "qwen2.5-coder:1.5b",
    "apiBase": "http://localhost:11434"
  }
}

If using SSH tunnel, localhost:11434 works as-is. If exposing directly, replace with your VPS IP.

Step 5: Use It

In VS Code, the Continue panel opens with Cmd/Ctrl+L. You can:

• Chat about your code: "Explain what this function does"
• Ask it to make changes: "Refactor this to use async/await"
• Highlight code and ask specific questions

Tab completions appear automatically as you type, same experience as Copilot, running locally.

Performance

On a VPS with 4 cores and 8 GB RAM, expect roughly (from our LLM on CPU benchmark):

• Chat responses (7B model): about 4 tokens/second on shared cores, about 8 on dedicated ones. Usable for questions, much slower than Copilot
• Tab completions (1.5B model): faster than 7B, but each completion first has to read the code around the cursor. In our tests a 7B model read input at 13–18 tokens/second, so keep the context small or completions will lag

Running this on NoctHost

Qwen 2.5 Coder 7B wants about 8 GB of RAM, so the Pro plan (4 vCPU, 8 GB) is the right fit for a responsive assistant. On NoctHost that runs hourly from a prepaid balance you top up with crypto (no KYC) or a card, and you get a dedicated IPv4 so your Continue endpoint has a stable address to reach.