Mario Zechner has released Pi, a minimal terminal coding harness designed for extensibility over built-in features. Available via npm as `@mariozechner/pi-coding-agent`, Pi supports 15+ providers including OpenAI, Anthropic, Google, and open source models like Ollama. The tool offers four modes: interactive TUI, print/JSON output, RPC protocol, and embeddable SDK.

Pi's architecture emphasizes user control through context engineering, with support for AGENTS.md project instructions, SYSTEM.md custom prompts, and auto-compaction to manage context windows. Users can extend functionality through TypeScript modules, skills packages, and prompt templates. The tool features tree-structured sessions for navigating conversation history and supports model switching mid-session with keyboard shortcuts.

Unlike other coding agents, Pi deliberately omits features like sub-agents, plan mode, and MCP support, instead providing primitives for developers to build these capabilities themselves. This approach keeps the core minimal while allowing users to shape the tool to their workflows. The project is open source under MIT license, with packages installable from npm or git. Developers can find community resources on GitHub, Discord, and the project's documentation.