Ollama just dropped support for GitHub’s Copilot CLI. If you care about local code completion, this matters.

What Changed

You can now use Ollama as the backend for GitHub Copilot CLI instead of relying on cloud APIs. Your code stays local.

Why It Matters

Before After
Cloud API Local model
Internet required Works offline
Data leaves your machine 100% private

Setup

# Install Copilot CLI first
brew install copilot

# Configure to use Ollama
copilot config set provider ollama
copilot config set model codellama

Which Models Work Best

  • Codellama — Already trained on code
  • DeepSeek-Coder — Great at context
  • Qwen2.5-Coder — Good balance

Speed on M4 Max

With a Q4 quantization, you’re looking at ~30-50 tokens/sec. Enough for real-time completion without cloud latency.

Alternative: Use LM Studio

If Ollama + Copilot feels clunky, LM Studio has better GUI integration for code completion with its VSCode plugin.

Verdict

This is a solid step toward fully local coding. Try it if privacy matters to you.

Guide completed 2026-04-19