code-llama-for-vscode

Why we included this project

This small Python script gives you a fully local coding assistant inside Visual Studio Code, so you can skip hosted services and keep your code off a cloud API. It works by mocking the llama.cpp API, which lets the Continue extension talk to Meta's Code Llama Instruct models running on your own machine. The setup is hands-on: you download a Code Llama model, run the provided script with torchrun, and point Continue's config at it. The result is a Copilot-style experience that never leaves your hardware and runs on Windows, Linux, and macOS. Teams that can't send code to the cloud or don't have a budget for API keys will find it a practical way to try local code completion, and since the whole thing is one file, it's easy to read and adapt. Just know it's a glue layer rather than a finished product, so you'll do the wiring yourself.