Pi 1.0
Earendil has released Pi 1.0, a minimal and extensible AI agent harness that introduces native Model Context Protocol support, non-LLM model integration, and an experimental package called Pi Durable.
summary is enough The release post outlines new features and an experimental companion package clearly, but technical readers can grasp the core updates quickly without reading the full text.
summary
Earendil has launched Pi 1.0, a stable and minimal agent harness featuring native support for the Model Context Protocol, non-LLM models like Jev, deferred tool loading, and Anthropic cache warming. Alongside the core release, the team introduced Pi Durable, an experimental package designed for long-running agentic applications outside the terminal. The release builds on months of user feedback while striving to maintain a lightweight core. Commenters discuss its practicality compared to other tools like Claude Code and oh-my-pi, debating its claims of minimalism and user numbers.
key points
- Pi 1.0 adds native Model Context Protocol support, deferred tool loading, and Anthropic cache warming.
- The new release integrates non-LLM models such as Jev and image models via Codemode.
- Pi Durable is introduced as an experimental package for building long-running, durable agentic applications outside the terminal.
who should read it
Software developers and engineers looking for customizable, minimal AI agent frameworks.
discussion positive
consensus
Commenters generally praise Pi for its speed and hackability compared to bloated alternatives, though some debate whether it truly qualifies as minimal or if its user count claims are accurate.
pushback
A minority of commenters express skepticism about TypeScript-based agent tools consuming excessive RAM, wishing for more lightweight, statically compiled Rust alternatives.
Love pi. I tried to run some local models and pi was the only one that actually worked decently because it didn’t have a gargantuan system prompt that would take minutes to prefill on my scrawny ass laptop.
Been running it almost barebones vanilla for a couple of months. Just a bunch of basic extensions and some skills.
Now, if only they could fix the very annoying bug of the history jumping back at the beginning if I am not a the end while the model is reasoning that would great.
FacelessJim shares a practical user experience explaining why Pi succeeds with local models where others fail due to smaller system prompts.
Generated by an LLM from the article and thread. Check the sources before relying on it.