Hugging Face just showed how to turn Pi into a fully private, offline development environment using llama.cpp and the new /llama command. The tutorial covers installing the engine, using Hugging Face's hardware compatibility tools to find the perfect GGUF quantization level for your specific machine, and running Qwen3 8B locally.
No data, prompt, or code ever leaves your hardware, and there are zero token costs. Perfect for private code assistants.
Have you tried local workflows in Pi yet?

When prompts are free, you stop rationing them and start experimenting: more refactors, more throwaway scripts, more wild ideas. Privacy is the lesser benefit.

The bigger one is iteration speed with no billing anxiety. I'd test Qwen3 8B on a staging server, tune the GGUF quant, and see if the latency feels like a tool or a chore.