DEV Community

niuniu
niuniu

Posted on

I Replaced My $240/Year Cursor Subscription with Continue.dev — Here's the Free Setup

I paid Cursor $20/month for 12 months. Then I looked at my usage: 90% of my AI coding was autocomplete and chat. The "AI-first IDE" features I actually used? Cmd+K for inline edits. That's it.

I cancelled Cursor, installed Continue.dev in VS Code, and pointed it at local Ollama models. Total cost: $0. Here's exactly what I set up and what I gave up.

The Continue.dev Setup 

# 1. Install Ollama (if you haven't)
curl -fsSL https://ollama.com/install.sh | sh

# 2. Pull a coding model (7B is the sweet spot for laptops)
ollama pull codellama:7b-code

# 3. Install Continue extension in VS Code
code --install-extension Continue.continue
Enter fullscreen mode Exit fullscreen mode 
// ~/.continue/config.json
{
  "models": [
    {
      "title": "CodeLlama 7B (Local)",
      "provider": "ollama",
      "model": "codellama:7b-code",
      "apiBase": "http://localhost:11434"
    },
    {
      "title": "Llama 3.2 3B (Fast)",
      "provider": "ollama",
      "model": "llama3.2:3b",
      "apiBase": "http://localhost:11434"
    }
  ],
  "tabAutocompleteModel": {
    "title": "Autocomplete",
    "provider": "ollama",
    "model": "codellama:7b-code"
  },
  "embeddingsProvider": {
    "provider": "ollama",
    "model": "nomic-embed-text"
  }
}
Enter fullscreen mode Exit fullscreen mode

What Works (90% of My Usage)

Feature Cursor ($240/yr) Continue + Ollama ($0)
Tab autocomplete ✅ GPT-4-level ✅ CodeLlama 7B — 200ms latency
Chat with codebase ✅ ✅ @codebase reference
Inline edit (Cmd+K) ✅ ✅ Select code → Cmd+I
Multi-file context ✅ ✅ Manual @-mentions
Privacy ❌ Cloud ✅ 100% local
Works offline ❌ ✅

What I Actually Gave Up

  1. GPT-4 reasoning. For "explain this legacy codebase" questions, Claude/GPT-4 are still better. I use the free Claude web UI for those — maybe 3 times a week.

  2. Cursor's "Composer" multi-file editing. Continue can edit multiple files, but Cursor's UI for reviewing diffs across files is genuinely nicer. I compensate with smaller, more focused prompts.

  3. Zero-config setup. Continue requires you to understand context length, model selection, and prompt engineering. Cursor hides all that. This is either a bug or a feature depending on how much you like tweaking configs.

The Performance Reality

Task Cursor (GPT-4) Continue (CodeLlama 7B)
Autocomplete latency 300-800ms 150-300ms
"Fix this bug" accuracy 85% 70%
"Explain this code" quality 9/10 7/10
Cost per month $20 $0

The 70% vs 85% accuracy gap is real, but here's the thing: I review all AI-generated code anyway. The 15% worse suggestions get caught in the same code review process. The $240/year savings is permanent.

My Hybrid Workflow 

Day-to-day coding:
  Continue.dev + Ollama (local, free, fast)
    ↓
Complex architecture questions:
  Claude.ai free tier (web, ~10 questions/week)
    ↓
Never:
  Cloud-based AI coding subscriptions
Enter fullscreen mode Exit fullscreen mode

The Tooling I Use for Setup

I built the migration scripts and config templates with MonkeyCode — the free AI coding assistant that runs entirely on your machine. It generated my Continue config from my Cursor usage logs in about 20 seconds: https://ly.cyberserval.tech/iIETXiF

What's your AI coding stack? I'm especially curious if anyone's running Continue with a local 13B+ model — does the accuracy gap close enough to justify the RAM?

Top comments (0)