# Oh My OpenAgent

> Preconfigured multi-agent bundle for OpenCode and Codex CLI, driven by one keyword

- Category: AI Coding / Coding-Agent Harnesses & Runtimes
- Language: TypeScript
- Licence: SUL-1.0
- Repository: https://github.com/code-yeongyu/oh-my-openagent
- Website: https://omo.dev
- Canonical page: https://ai-tldr.dev/tools/oh-my-openagent/

## In short

Oh My OpenAgent — OmO — is a configuration layer for coding agents. It sits in the Coding-Agent Harnesses & Runtimes category of the AI Coding tools and is open-source, written in TypeScript and licensed SUL-1.0. To get started: ultimate Edition (OpenCode).

## Overview

Oh My OpenAgent — OmO — is a configuration layer for coding agents. Instead of assembling sub-agents, hooks, MCP servers and model routing yourself for every project, you install the bundle into a host CLI you already run and drive the whole arrangement with a single keyword: type `ultrawork` (or `mass ulw` for the multi-model graph mode) with your prompt and the preset takes over.

It ships as three editions of the same product. The Ultimate Edition plugs into OpenCode and is the full package: 11 agents, 54+ lifecycle hooks, four built-in MCP servers (websearch, context7, grep_app and lsp), the slash commands, Team Mode and `/goal`. The Light Edition is the portable subset that fits Codex CLI's plugin system — rules, comment-checker, git-bash, lsp, ultrawork and the loop/continuation components, plugin-scoped MCPs, and Codex agent TOMLs installed into `~/.codex/agents/`. The Senpi Edition is a standalone `omo` command with the extension built in, published on a beta channel.

The project's stated premise is provider independence: rather than committing to one vendor's agent, OmO orchestrates several models together — its own demo runs Team Mode across Kimi K3 and GPT-5.6 Sol. It is licensed under SUL-1.0 and maintained as a side project with sponsor support.

## What it does

- One-keyword entry point: `ultrawork` runs the whole preset; `mass ulw` drives the multi-model graph mode
- Ultimate Edition for OpenCode — 11 agents, 54+ lifecycle hooks, Team Mode, /goal and the full slash-command set
- Four built-in MCP servers wired in by default: websearch, context7, grep_app and lsp
- Light Edition for Codex CLI, installing agent TOMLs and plugin-scoped MCPs into ~/.codex
- Senpi Edition: a standalone `omo` binary with the OMO extension bundled (beta channel)
- Multi-model orchestration across providers rather than a single vendor's agent

## Getting started

Pick the edition that matches the host you already run. The Ultimate installer is a TUI that walks you through plugin registration, agent/model config and provider auth.

1. Ultimate Edition (OpenCode)

   Registers the plugin in opencode.json along with agent and model configuration.

```bash
bunx oh-my-openagent install
```

2. Light Edition (Codex CLI)

   Runs through Node/npm and lands the plugin cache, marketplace entries and agent TOMLs under ~/.codex, plus component CLIs in ~/.local/bin.

```bash
npx lazycodex-ai install
```

3. Install both hosts at once

   The shared toolkit CLI takes a platform flag; it defaults to opencode.

```bash
bunx oh-my-openagent install --platform=both
```

4. Senpi Edition (standalone, beta)

   A bare `npm i -g omo-ai` fails by design — the standalone edition is beta-channel only.

```bash
npm i -g omo-ai@beta
omo
```

5. Use it

   Once installed, the preset is invoked by keyword inside your host CLI rather than by a separate command.

```text
ultrawork  <your prompt>
mass ulw   <your prompt>   # multi-model graph mode
```

## When to use it

- Get a tuned multi-agent setup in OpenCode or Codex without hand-writing hooks and MCP config per project
- Run long, unattended coding tasks through a preset loop instead of babysitting each step
- Spread a task across several model providers rather than committing to one vendor's agent
- Standardize agent behaviour across a team by installing the same bundle everywhere