Pi Agent setup: terminal workflows and local models

This tutorial introduces Pi Agent, an open source coding assistant that works directly in the terminal. The presenter covers installation and provider login, then demonstrates repository tasks with Docker Compose, GitLab and Terraform. He describes support for Windows, Linux and multiple terminal emulators, and uses Warp for his examples.

Provider setup includes subscription login through OAuth or API keys. For a local LLM connection, the tutorial points to custom provider configuration in models.json, including LM Studio and OpenAI-compatible endpoints. It does not walk through installing a model runtime.

The practical examples show how to attach file context with @, execute shell commands with an exclamation mark and use AGENTS.md to explain repository structure and deployment targets. One container check initially fails because Pi looks on the local machine; a more specific prompt directs it to the remote server.

Session tools cover resuming conversations, browsing a tree, forking and compacting context. Returning to an earlier conversation state does not automatically restore files. The presenter also creates a reusable Markdown prompt command and discusses extensions. He closes with a caution: Pi executes commands without permission popups in the setup described, so destructive requests can proceed without another confirmation.