oh-my-pi (omp)

oh-my-pi, commonly invoked as omp, is a terminal-first coding agent derived from Mario Zechner’s Pi project. It can run as an interactive terminal interface, a one-shot command, an RPC server, or an Agent Client Protocol integration for compatible editors. The agent reads and edits repository files, runs shell commands, and combines those workflows with language-server, debugger, code-review, subagent, browser, and git tooling. It is an actively developed developer tool rather than a hosted coding service.

Users install the client from the official site and connect a supported model provider or compatible endpoint. The tool does not make inference universally free: API usage, subscriptions, and rate limits belong to the chosen provider. The install guide documents macOS and Linux shell installation plus Windows PowerShell installation. Provider lists and authentication methods can still change. The canonical repository is public under the MIT license.

Top features

• Terminal and protocol surfaces: use the interactive TUI, scripted one-shot mode, RPC, or ACP connection according to the surrounding workflow.
• LSP-aware operations: language-server context supports symbol-level changes, rename handling, diagnostics, and repository navigation.
• Debugger adapters: documented integrations can attach to supported native, Go, or Python debugging sessions and inspect runtime state.
• Subagents and worktrees: delegate bounded investigations or implementation tasks into isolated git worktrees and collect structured results.
• Code review workflow: reviewer agents can report prioritized findings and a clear recommendation before changes are accepted.
• Browser and Electron automation: inspect web interfaces or compatible desktop applications when UI behavior is part of a coding task.
• Provider routing: configure different model backends, roles, and fallback paths rather than depending on one hosted model.

Use cases

• Trace a defect with language-server references and a real debugger, then apply a scoped code change.
• Split repository exploration, test analysis, and implementation among isolated subagents.
• Run a second review pass that ranks concrete correctness or security concerns before merge.
• Use ACP to bring the agent into a compatible editor while retaining its terminal toolset.
• Verify a web or Electron UI after a code change with explicit browser-automation permissions.