Earendil Releases Pi 1.0 Minimal Terminal Coding Agent and Experimental Pi Durable
Key point
Pi 1.0 introduces native MCP support and virtual model extensions, while the new Pi Durable package targets long-running agentic applications.
Details
Earendil has released Pi 1.0, a hardened, minimal, and extensible agent harness designed for terminal-based coding. The update emphasizes stability and minimalism, incorporating features that have proven their utility over months of community feedback rather than adopting every new trend in agentic tooling.
Key Features in Pi 1.0
The release adds several core capabilities to the existing minimal framework:
- Codemode: Native support for MCP (Model Context Protocol) and non-LLM models, including Jev and image models.
- Virtual Models: Extension support allowing users to define custom model behaviors.
- Performance Enhancements: Deferred tool loading and cache warming specifically for Anthropic models.
- Dynamic Context: Mid-conversation system messages that allow for transcript-aware prompt and tool changes.
- UI Updates: A new TUI theme and full-screen mode enabled by default.
Introduction of Pi Durable
Alongside Pi 1.0, Earendil is shipping Pi Durable, an experimental package designed for long-running agentic applications. While Pi remains focused on minimalism and terminal usage, Pi Durable extends these principles to support durable, multi-surface interactions and longer-running tasks. It maintains the core philosophy of supermalleability but adapts it for applications requiring sustained intelligence steering.
Availability and Licensing
Both Pi 1.0 and Pi Durable are released under the MIT license. Pi 1.0 is available via a curl script for Unix-like systems and PowerShell for Windows. Pi Durable is distributed as an npm package (@earendil-works/pi-durable). Documentation is hosted at pi.dev, and the source code is available on GitHub.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.