DEV Community

niuniu
niuniu

Posted on

I Switched from GitHub Copilot to Continue.dev + Ollama for 30 Days — My Code Never Left My Laptop

On July 17th I cancelled GitHub Copilot and set a rule: 30 days, local models only, no cheating. Here's the honest report.

The setup (10 minutes) 

# 1. Install Ollama and pull a code model
curl -fsSL https://ollama.com/install.sh | sh
ollama pull qwen2.5-coder:7b

# 2. Install Continue.dev extension in VS Code, then configure:
Enter fullscreen mode Exit fullscreen mode 
// ~/.continue/config.json
{
  "models": [
    {
      "title": "Qwen Coder (local)",
      "provider": "ollama",
      "model": "qwen2.5-coder:7b"
    }
  ],
  "tabAutocompleteModel": {
    "title": "Autocomplete",
    "provider": "ollama",
    "model": "qwen2.5-coder:1.5b"
  }
}
Enter fullscreen mode Exit fullscreen mode

Two models: the 7B for chat/edit, the 1.5B for tab autocomplete (fast enough to not feel laggy).

30 days of real numbers

Metric Copilot (before) Continue + Ollama (30 days)
Monthly cost $10 $0
Autocomplete latency ~300ms ~180ms (M-series, 16GB)
Accept rate (my estimate) ~35% ~25%
Offline on a plane No Yes (tested, 2 flights)
Code privacy Sent to cloud 100% local

What broke (honesty section)

  • Day 4: 7B chat responses were noticeably dumber than Copilot chat on a gnarly regex refactor. I waited 20s for a wrong answer. Almost cheated.
  • Day 9: Battery life took a real hit — local inference on a laptop costs ~1.5h of unplugged time.
  • Day 15: The breakthrough. Once I switched autocomplete to the 1.5B model and reserved the 7B for deliberate "explain/refactor this" calls, the workflow clicked. Latency complaints vanished.

The uncomfortable conclusion

Copilot's moat in 2026 isn't the model — it's the zero-config polish. Continue.dev + Ollama is 90% of the experience for $0 and full privacy, but you pay with one evening of setup and occasional model babysitting.

For the agent-style multi-file work that local 7B models still can't do well, I use MonkeyCode — free and open-source, and it doesn't require me to hand my whole repo to a cloud API: https://ly.cyberserval.tech/iIETXiF

Would you trade 10% completion quality for $120/year and never sending your code to anyone's server? Or is cloud convenience worth the privacy cost?

Top comments (0)