Player not loading? Watch on YouTube
In this Russian-language tutorial with English captions, Pi is a terminal coding assistant that the presenter describes as a minimal harness around a model. Its core tools read, write and edit files and run shell commands. The tutorial shows how to extend that core on a React + Vite project rather than relying on a predefined agent setup.
The presenter uses Pi version 0.87 and explains installation instructions on an already installed setup, then demonstrates subscription or API-key login, model selection and thinking levels. An AGENTS.md file supplies project rules and completion criteria. The first task builds a task tracker; a queued clarification asks the agent to avoid external icon libraries. The demonstration also shows switching models without losing the session context and asking about code without requesting edits.
Customization takes up much of the guide. The presenter installs pi-mcp-adapter and pi-subagents, then creates a skill that assigns implementation and review to separate agents. Their task adds localStorage persistence and rejects blank tasks. Playwright MCP connects through .mcp.json for browser checks in Chrome, although the presenter manually starts the development server during the demonstration.
The final examples create a /check extension for build and lint, a /review prompt that requests no code changes, and session branches through /tree. This is a guide to configuring an AI agent with model providers; it does not demonstrate offline inference or running a local LLM.
