Set it up

About 20 minutes, four steps. I checked the commands against the omp README and docs at v17.0.9 (July 2026). omp releases almost daily, so if a command has drifted, trust omp --help.

1 · Install

Pick one:

# macOS / Linux
curl -fsSL https://omp.sh/install | sh

# Homebrew
brew install can1357/tap/omp

# Bun (the README's recommended path; needs bun ≥ 1.3.14)
bun install -g @oh-my-pi/pi-coding-agent

# Windows (PowerShell)
irm https://omp.sh/install.ps1 | iex

# Version-pinned via mise
mise use -g github:can1357/oh-my-pi

Run omp in a repo. The first launch runs omp setup to help you pick a default model.

2 · Connect models (you can mix these)

A subscription you already pay for

/login starts OAuth for anthropic (your Claude account), github-copilot, cursor, gitlab-duo and others. Check the plan-terms trap in the traps.

API keys

Set env vars such as ANTHROPIC_API_KEY or OPENROUTER_API_KEY. The provider list has the rest (openai, google, groq, deepseek, amazon-bedrock, openrouter…). On a headless machine, run omp auth-broker login <provider>.

Local models, free

ollama (127.0.0.1:11434), llama.cpp (:8080) and lm-studio (:1234) are found automatically on their default ports. Other OpenAI-compatible endpoints, vLLM included, go in models.yml (next page).

3 · Migrate your Claude Code setup

omp does this for you. On first run it reads rules, skills and MCP servers from .claude, .cursor, .windsurf, .gemini, .codex, .cline, .github/copilot and .vscode. MCP servers for omp alone go in .omp/mcp.json (project) or ~/.omp/agent/mcp.json (user), over stdio, http or sse.

4 · Turn off YOLO mode

omp runs in YOLO mode out of the box. It runs tool calls, shell included, without asking. Add this to ~/.omp/agent/config.yml before you point it at anything that matters:

tools:
  approvalMode: write

write asks before changes. always-ask asks for everything, tools.approval.<toolName>: allow|deny|prompt sets rules per tool, and --yolo turns it off for one run.

That’s a working setup. Routing comes next.