Unlock Free, Local AI Code Completion with continue.dev + Ollama in VSCode

Ditch the paid subscriptions. Here’s how to get powerful, privacy-first AI code suggestions for $0.

---

Intro

If you’ve been itching to use AI-powered code completion & understanding your code but don’t want to pay for GitHub Copilot or risk sending your code to third-party servers, this post is for you.

Today, I’ll show you how to combine continue.dev’s open-source VSCode extension with a local instance of Ollama to get fast, offline-friendly code suggestions—for free.

No API keys. No subscriptions. Just your machine, a code editor, and the magic of locally running language models.

---

Why This Setup Rocks

1. Privacy: Your code never leaves your machine.

2. Cost: $0. Forever.

3. Offline Use: Code in the middle of nowhere (or on a plane).

4. Custom Models: Use CodeLlama, Mistral, DeepSeek Coder or any Ollama-supported model.

---

Step 1: Install Ollama & Load a Model

First, download and install Ollama (macOS/Linux/WSL only for now). Then, pull a coding-focused model like CodeLlama or DeepSeekCoder:

ollama pull deepseek-coder-v2:16b  # capable model, slightly slow

Pro tip #1: Use codellama:7b if you are finding deepseek to be slower for a slightly less capable model. Use codellama:32b only if you have >32GB of memory.

Pro tip #2: You can list multiple models in the config and choose in the VSCode UI which model the continue.dev extension uses.

---

Step 2: Install the continue.dev VSCode Extension

Head to the VSCode Marketplace and install Continue.dev Coder.

---

Step 3: Connect continue.dev to Ollama

continue.dev defaults to cloud-based models, but we’ll reconfigure it to use your local Ollama server.

1. Open continue.dev’s settings (JSON) and add:

"models": [
    {
        "title": "DeepSeek Coder 2 16B",
        "provider": "ollama",
        "model": "deepseek-coder-v2",
        "apiBase": "http://localhost:11434/"
    }
  ],
  "tabAutocompleteModel": {
      "title": "DeepSeek Coder 2 16B",
      "provider": "ollama",
      "model": "deepseek-coder-v2",
      "apiBase": "http://localhost:11434/"
  }, 

(Pro tip: Continue.dev’s config.json can be found at ~/.continue/config.json on macOS.)

2. Restart VSCode to apply changes.

---

Step 4: Start Coding (with Examples!)

Let’s test it out. Ollama will run locally in the background, and continue.dev will tap into it for suggestions.

Example 1: Python Function Generation
Type a comment like:

# Create a function to calculate factorial  

Watch continue.dev + Ollama suggest: (Use TAB to apply the suggested code)

Example 2: React Component
Start typing a React component, and the model will start to auto-fill :

Example 3: Ask questions and get more information about your code

Highlight some code in your existing code-base, click CMD+L to start a chat window in VSCode and then you can ask the LLM questions about the code.

---

Troubleshooting Tips

• Slow? Use smaller models (e.g., codellama:7b instead of deepseek-coder-v2:16b).

• No suggestions? Check if Ollama is running (ollama serve).

---

Limitations

• Local models (even deepseek & codellama) aren’t quite as sharp as GPT-4 or larger models like deepseek-coder-v2:236b.

• Requires decent RAM (8GB+ recommended).

---

Final Thoughts

This setup is a game-changer for developers prioritizing privacy, cost, or offline workflows. While you won’t get ChatGPT-level nuance, CodeLlama via Ollama and continue.dev delivers surprisingly solid suggestions—and it’s completely free.

Give it a spin, and let me know how it works for you!

P.S. If you found this helpful, subscribe for more dev tools deep dives. Got questions? Drop them in the comments!