I've been paying $10/month for GitHub Copilot for 8 months. Last week I got the renewal email and asked myself a simple question: am I getting $120/year of value?
The honest answer was no. Most of what I use Copilot for — autocomplete on boilerplate, docstrings, simple refactors — a free, local, open-source setup can do just as well. So I switched to Continue.dev + Ollama and canceled my subscription.
Here's what actually happened.
The Cost Comparison Nobody Talks About
| GitHub Copilot | Continue + Ollama | |
|---|---|---|
| Price | $10/month ($120/year) | $0 |
| Code leaves your machine | Yes | No |
| Works offline | No | Yes |
| Model choice | Locked to OpenAI | Any local model |
| Rate limits | Yes | None |
The privacy part matters more than people admit. Every line you type in your editor gets sent to a third party with Copilot. With Continue + a local model, nothing leaves your laptop. For anyone working on proprietary code, that alone is worth the switch.
What I Actually Set Up
I'm running Continue in VS Code with qwen2.5-coder:7b via Ollama on a 3-year-old laptop with 16GB RAM. Setup took 11 minutes:
# 1. Install Ollama
curl -fsSL https://ollama.com/install.sh | sh
# 2. Pull the coding model (4.7GB, one-time download)
ollama pull qwen2.5-coder:7b
# 3. Start the server
ollama serve
Then in VS Code: install the Continue extension, add this to config.json:
{
"models": [{
"title": "Qwen Coder Local",
"provider": "ollama",
"model": "qwen2.5-coder:7b"
}],
"tabAutocompleteModel": {
"title": "Qwen Autocomplete",
"provider": "ollama",
"model": "qwen2.5-coder:7b"
}
}
Done. No API keys, no account, no billing page.
The Honest Downsides
This isn't a paid ad, so here's the real talk after 5 days of daily use:
- Autocomplete latency is higher. Copilot suggestions appear in ~300ms. Local 7B takes 1-2 seconds. For me, that's acceptable. For someone writing code at 100 WPM, it might not be.
- Complex reasoning is weaker. Copilot (backed by GPT-4-class models) is better at "refactor this entire module" tasks. The local 7B model handles line-level and function-level completions well but struggles with multi-file reasoning.
- You need the RAM. 16GB is the realistic minimum. On an 8GB machine, this setup will swap and crawl.
Where the Free Setup Actually Wins
- Boilerplate and repetitive code. Writing CRUD endpoints, test stubs, config files — the local model is just as good and unlimited.
- Offline work. I wrote code on a 6-hour train ride with no WiFi. Copilot would have been a brick.
- Learning. I can see exactly what the model is doing, swap models freely, and tune prompts. With Copilot, it's a black box.
- Privacy. Non-negotiable for client work.
My Actual Verdict
If your work is mostly greenfield algorithm design or large refactors, keep Copilot. If your day is 70% boilerplate, tests, and standard patterns — the Continue + Ollama combo will save you $120/year with maybe a 10% quality drop on edge cases.
I'm keeping the $0 setup. The 1-2 second latency is a feature, not a bug — it makes me actually read the suggestions instead of tab-tab-tab accepting everything.
For a fully-managed free alternative with no local setup, I've also been testing MonkeyCode alongside Continue. Worth a look if you don't want to run models yourself.
Have you tried running your coding assistant locally? What was your experience — is the latency worth the privacy and cost savings?
Top comments (0)