Pi vs OpenCode: setup and extension tutorial

This tutorial introduces Pi as a minimal coding assistant and compares its design with OpenCode. The speaker prefers Pi for customization: its basic interface can chat and edit files, while extensions supply features such as approval prompts and undo/redo. He describes OpenCode as a more complete starting point, with MCP, LSP and other features already included.

The setup covers installation options including npm, Windows PowerShell and an AUR package for Arch Linux. The demonstration connects to OpenCode Zen Go with an API key, selects GLM 5.2 and sets the reasoning level to maximum. Although the speaker discusses open-weight models, this example uses a provider connection; it does not demonstrate how to run models locally.

The permissions extension adds approval prompts for writes, edits and command execution. The speaker configures exceptions for selected read-only operations, then tests the prompts with a Python script. Next, he installs pi-undo-redo and demonstrates removing and restoring a generated file. Its non-Git mode only restores files touched by Pi's write/edit tools, and shell-created paths are not restorable.

The final example adds pi-web-access to search for LangChain documentation and review a summary in a browser. The speaker recommends checking extension packages and repositories before installing them. His preference for Pi rests on its configurable design, rather than a measured comparison of agent performance.