Running Local AI Coding Assistants: Ollama, DeepSeek-Coder & VS Code

Cloud-hosted AI coding assistants are convenient, but sending proprietary source code or private API keys over external network connections is not always acceptable. Running open-weights Large Language Models (LLMs) locally gives you full privacy, zero latency rate limits, and complete offline availability.

With modern open models like DeepSeek-Coder and lightweight runners like Ollama, running a high-performance coding assistant on your local workstation is straightforward.

---

Why Run Local AI Coding Models?

• 100% Data Privacy: Source code and API tokens never leave your local machine or internal network.
• Zero API Metering Costs: No monthly subscription fees or token usage limits.
• Offline Availability: Works seamlessly on laptops without an active internet connection.

---

1. Installing Ollama on Linux & Windows

Ollama simplifies downloading and running GGUF-quantized LLMs with single-command setup.

Installing on Linux

Run the official installation script in your terminal:

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

Pulling the DeepSeek-Coder Model

DeepSeek-Coder is engineered specifically for software development, code autocompletion, and refactoring tasks. To pull the 6.7B parameter model:

ollama pull deepseek-coder:6.7b

For machines with high VRAM GPUs (16GB+ VRAM), pull the larger 33B model:

ollama pull deepseek-coder:33b

Verify Ollama service status and test model inference via command line:

ollama run deepseek-coder:6.7b "Write a Python script to parse JSON server logs and output error counts."

---

2. Integrating Ollama with VS Code

To turn Ollama into a seamless IDE pair programmer inside Visual Studio Code, use open-source extension integrations like Continue.dev or Twinny.

Configuring Continue.dev Extension

1. Install the Continue extension from the VS Code Marketplace.
2. Open ~/.continue/config.json.
3. Configure Ollama as your local model provider:

{
  "models": [
    {
      "title": "DeepSeek Coder 6.7B",
      "provider": "ollama",
      "model": "deepseek-coder:6.7b",
      "apiBase": "http://localhost:11434"
    }
  ],
  "tabAutocompleteModel": {
    "title": "DeepSeek Coder Autocomplete",
    "provider": "ollama",
    "model": "deepseek-coder:1.5b",
    "apiBase": "http://localhost:11434"
  }
}

---

3. Hardware Requirements & Quantization Tips

• VRAM vs RAM: Running LLMs directly on GPU VRAM (NVIDIA CUDA or Apple Silicon Unified Memory) yields fast inference speeds (30+ tokens/sec). CPU RAM fallback works but is significantly slower.
• Choosing Quantization Levels: Q4_K_M or Q5_K_M quantizations offer the best balance between model intelligence and memory footprint.

---

Conclusion

Setting up Ollama with DeepSeek-Coder inside VS Code delivers a private, fast, and capable local coding environment. You get real-time code completion and refactoring without relying on external cloud APIs.