Pi Coding Agent with LM Studio: offline setup

This tutorial connects Pi Coding Agent to LM Studio on Windows so a local LLM can answer requests through the coding assistant. The speaker presents the setup as a way to avoid hosted API quotas and assumes Pi is already installed; installation belongs to an earlier video.

The walkthrough downloads LM Studio, selects a model and loads it in the Developer section. The speaker uses a small Gemma model described as having 2 billion parameters and a 4 GB download, while advising viewers to choose a model their hardware can run. He then starts the local server and creates an API token.

The supplied description gives the extension command as pi install npm:pi-lmstudio. Configuration connects the server address, token and model identifier to Pi. There is a filename discrepancy: the description specifies models.json, while the transcript says model.json. Viewers should account for that inconsistency when following the setup.

After selecting LM Studio in Pi, the speaker tests a greeting and lowers the thinking level to Minimum to improve response speed. He then turns off Wi-Fi and reports that the model continues answering offline. These brief chat tests demonstrate the connection described in the tutorial; they do not establish coding performance or how larger models will behave on other hardware.