Local LLM setup for coding (pair programming style) - GPU vs MacBook Pro?

Hey everyone, I'm a programmer and I'd love to use local LLMs as a kind of "superpower" to move faster in my day-to-day work. Typical use case: I'm working on a codebase (Rust, Python, Go, or TypeScript with React/Vue), and I want the model to understand the existing project and implement new features on top of it — ideally writing code directly in my IDE, like a pair programming partner. Right now I've tried cloud models like Claude, Qwen, ChatGPT, and GLM. Results are honestly great (especi