Run a local coding model with pi and LM Studio

This post is for people who generally use Claude, Codex, or Gemini but have heard you can run open-source models locally for free. The goal is to get you set up in no time so you can play around with the power of local models. If you already use a coding agent, you already know how this works. A coding agent (e.g., Claude Code) talks to a model over an HTTP API: it sends your request, the model sends back tokens, and the agent uses tools (read, edit, run) to do real work. With Claude Code or Codex, that API lives in a datacenter and you reach it over the internet. Running locally changes one main thing: where the endpoint is. Instead of pointing your agent at a remote provider, you point it at a server running on your own machine. Same request in, same completion out except the model just happens to be sitting on your local computer. Here's the architecture, side by side: The three pieces on the local side: pi — the coding agent (the equivalent of Claude Code / Codex). It sends…