Pi (pie.dev) is a minimal, fully open-source AI agent that runs in the terminal, designed around simplicity rather than feature bloat. Unlike Claude Code or Codex CLI, it has no MCP setup, no sub-agents, and no permission pop-ups — it executes plain CLI commands directly. It supports 15+ LLM providers including local models via LM Studio or Ollama, and uses OAuth or API keys. Key features include an agents.md context file for project-specific instructions, custom slash-command prompts defined as markdown files, session management with tree view and forking, context compaction for long sessions, and shell passthrough via the ! prefix. The video walks through installation, provider setup, practical use cases like Docker inspection and Terraform/OpenTofu infrastructure management, and warns that the lack of guardrails means commands execute immediately without confirmation.
Sort: