Pi coding agent: setup, sessions and extensions

This tutorial introduces Pi as a minimal terminal coding assistant with file reading, editing, writing, Bash execution and sessions. The speaker describes its design as a small core that users adapt through prompts, skills and extensions.

The setup walkthrough uses the installer linked at pi.dev on WSL and covers Linux and Mac support. For Windows, the speaker says a Bash shell is required. The model connection example authenticates a ChatGPT subscription through /login, then tests it with a short prompt. This demonstration uses a provider account; it does not establish an offline workflow.

Navigation covers model selection, thinking levels and an external editor. The speaker also explains interactive mode, print and JSON output, RPC integration and an SDK for embedding Pi in applications.

The session example makes an important distinction: branching or forking changes conversation history, but leaves code edits in place. The speaker recommends Git or a checkpoint extension for file recovery. Customization then covers instruction files, system prompts, skills, saved prompt templates and packages. According to the tutorial, Pi has no built-in plan mode, MCP support, subagents or permission pop-ups; users can add capabilities through extensions.