Pi (AI agent)
Open-source AI coding agent
From Wikipedia, the free encyclopedia
Pi is an open-source artificial intelligence coding agent and agent harness developed by Earendil Works. It operates primarily through a terminal user interface and allows large language models to read, write, and modify source code and execute shell commands.[1] Pi is designed around a small extensible core and can be customized using TypeScript extensions, skills, prompt templates, themes, and packages.[2]
| Pi | |
|---|---|
| |
| Original author | Mario Zechner |
| Developer | Earendil Works |
| Written in | TypeScript |
| Operating system | Windows, macOS, Linux |
| Type | AI coding agent |
| License | MIT License |
| Website | pi |
| Repository | github |
History
Pi was created by software developer Mario Zechner and was initially distributed through the pi-mono repository under his badlogic account. The project was originally published through npm packages using the @mariozechner namespace.[3]
In May 2026, the project was transferred to Earendil Works. Its source repository moved to the earendil-works organization and its npm packages were renamed under the @earendil-works namespace, beginning with version 0.74.0.[4]
Features
Pi provides an interactive terminal interface through which an AI model can inspect a software project and use built-in tools for reading, writing, and editing files and executing shell commands.[5]
It supports models from multiple providers, including OpenAI, Anthropic, Google, Amazon Web Services, and OpenRouter, and allows users to switch between models during a session.[6] Pi can also connect to locally running language models through inference software including Ollama, LM Studio, vLLM, and SGLang.[7]
Pi emphasizes extensibility rather than including all agent functionality in its core application. Additional functionality can be implemented through TypeScript extensions and installable packages. Its software is divided into reusable components including a multi-provider LLM interface, an agent runtime for tool calling and state management, a terminal user interface library, and the coding-agent application.[8]
Pi can also operate non-interactively and provides a software development kit that allows its agent runtime to be embedded into other applications.[9]
Oh My Pi
Oh My Pi (OMP) is an open-source AI coding agent for the terminal developed as a fork of Pi. It is designed as a terminal coding agent with integrated IDE functionality, including Language Server Protocol integration, debugging support through the Debug Adapter Protocol, subagents, browser tools, and support for more than 60 artificial intelligence model providers.[10] The software supports model providers including OpenRouter, where it is listed as an actively used coding agent, with activity on the service dating to February 2026.[11] Oh My Pi can also be used with local language models through inference software including Ollama, LM Studio, llama.cpp, vLLM, and LiteLLM.[12][13] Oh My Pi also includes Hashline Edits, a code-editing system designed to reduce token usage by allowing models to reference hashed line anchors rather than reproduce existing code.[14]
