Oh My Pi (OMP) is a fork of Pi created by Can Bölük, described in Вам не нужен ещё один Claude Code: зачем Pi оставляет сборку агента пользователю as having a broader set of built-in development tools. It keeps Pi’s split into components for working with models, executing the agent and the terminal interface, but makes more of the decisions about how they work together inside the project itself. The overview places it closer to Claude Code, Codex CLI and OpenCode — a ready tool for everyday coding — while noting that its harness source remains available to study and change, and that extensions can add tools and commands without maintaining one’s own fork.
Capabilities
OMP provides LSP tools for diagnostics, navigation, renames and applying the fixes a language server suggests, making these semantic operations part of the agent’s normal work with code; the overview notes that results still depend on the language, the server and the project. It offers persistent Python and JavaScript sessions from which the agent’s own tools can be called, with the JavaScript side using Bun, and a debugger connected through DAP giving the agent breakpoints, stepping, the stack and variables. For delegation it has its own task tool, including parallel work and optional workspace isolation, with an Agent Hub for observing and steering the workers. It supports multiple model providers and lets different models be assigned to different roles, such as one for main work and another for auxiliary tasks.
Its divergence from Pi goes beyond the tool list: a pi-natives package brings in native code through N-API, separate Rust components handle the shell, syntax trees, file traversal and workspace isolation, and an omp-stats component provides a local dashboard of model usage statistics. Its Hashline editing mechanism anchors edits to lines and hashes of their content, changing how the model specifies where to edit and how it gets feedback when the expected content does not match.
Adoption & Ecosystem
The overview uses OMP to illustrate what it calls the pull of a fork: the more interconnected decisions of its own a fork accumulates, the more later changes must be reconciled with them, making it harder to maintain as a thin layer over the original.