Pi 1.0 is a hardened, minimal, extensible agent harness designed for reliable, production-ready use. It’s built around a conservative adoption philosophy: new capabilities are only added after proving durable and worth their complexity. This release formalizes that approach by adding native Codemode (including MCP and non-LLM models like Jev and image models), extension support for virtual models, deferred tool loading, cache warming for Anthropic models, transcript-aware mid-conversation system messages that alter prompts and tools, a refreshed TUI theme, and default full-screen mode. Hundreds of thousands of users already run Pi weekly, and these changes aim to keep the tool simple while making it more robust and versatile for coding agents and other agentic workflows.
Alongside Pi 1.0, an experimental Pi Durable package targets long-running agentic applications, carrying Pi’s minimalism and supermalleability into a substrate for durable tasks and multi-surface use. Examples show Codemode condensing a week of commits into a short script, and a virtual-model workflow where Claude Opus plans, Jev detects handoff points, and GPT-6 Luna implements; sessions can reload, pick routers/auto models, run finished scripts, and report per-model cost and cache use. Pi and Pi Durable are MIT licensed; install via curl or PowerShell for Pi and npm for Pi Durable, with docs at pi.dev and code on GitHub.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.