Sub-Agent Orchestrators: Catalog & Installation

A sub-agent orchestrator is a system where one “lead” agent manages a team of specialized agents, each with its own model and role.
This is not a framework (LangGraph, CrewAI), but a ready-to-use configuration that works “out of the box”.

← Ready-made AI Agents · AI Agent Architecture → · 🇷🇺 Русский

Contents

1. What Is a Sub-Agent Orchestrator
2. OhMyOpenAgent (Sisyphus)
3. Other Systems Approaching This
4. Comparison Table
5. Installing OhMyOpenAgent
6. What’s Next

1. What Is a Sub-Agent Orchestrator

Regular AI agent: one model → one task → one answer.

Sub-agent orchestrator: one task → planner → multiple specialized agents with different models → result synthesis.

Task
  │
  ▼
┌──────────────┐
│  Orchestrator │  ← decides who does what
└──────┬───────┘
       │
   ┌───┼───┬───┬───┐
   ▼   ▼   ▼   ▼   ▼
  A1  A2  A3  A4  A5   ← different agents, different models
  │   │   │   │   │
  └───┴───┴───┴───┘
       │
       ▼
    Result

Key principle: each agent uses the model optimal for its role. Cheap models — for grep and search, expensive — for architecture and planning.

As of writing (July 2026), the only system with this architecture “out of the box” is OhMyOpenAgent.

2. OhMyOpenAgent (Sisyphus)

GitHub → · 66K+ · TypeScript

Author: @code-yeongyu
Company: Sisyphus Labs
License: SUL-1.0
Releases: 222+ · Contributors: 300

Architecture (11 Agents)

OhMyOpenAgent contains 11 built-in agents in three layers:

Planning Layer

| Agent | Role | Models | |——-|——|——–| | Prometheus | Strategic planner. Interview, research, plan creation | claude-opus-4-8, gpt-5.6-sol, glm-5.2 | | Metis | Requirements analysis. Finds hidden intents, ambiguities, AI-slop | claude-sonnet-4-6, claude-opus-4-8 | | Momus | Plan QA review. Checks completeness, feasibility, consistency | gpt-5.6-terra, gpt-5.6-sol |

Execution Layer

| Agent | Role | Models | |——-|——|——–| | Atlas | Execution conductor. Reads plan, delegates tasks to sub-agents, accumulates knowledge | claude-sonnet-4-6, kimi-k3 | | Sisyphus | Main orchestrator. ultrawork — entry into full autonomous work mode | claude-opus-4-8, kimi-k3 |

Worker Layer

| Agent | Role | Models | |——-|——|——–| | Hephaestus | Deep executor. Receives goal → self-researches, plans, writes code | gpt-5.6-sol | | Sisyphus-Junior | Focused executor. Receives detailed prompt → executes | claude-sonnet-4-6, kimi-k3 | | Oracle | Architecture and debugging. Read-only consultant | gpt-5.6-sol, gemini-3.1-pro | | Explore | Code researcher. Contextual grep | gpt-5.4-mini-fast | | Librarian | Documentation and OSS. External reference search | gpt-5.4-mini-fast | | Multimodal-Looker | Image, PDF, media analysis | — |

How It Works in Practice

Simple path: write ulw or ultrawork in chat → Sisyphus decides which agents are needed and executes the task.

Precise path: @plan "do X" → Prometheus interviews, creates plan in .omo/plans/ → Momus + Oracle review → /start-work → Atlas reads plan and delegates tasks to sub-agents.

Deep path: switch to Hephaestus → Hephaestus self-researches, plans, and executes as an autonomous deep worker.

Work Modes

Two Editions

	Ultimate (OpenCode)	Light (Codex CLI)
Agents	11 agents with orchestration	8 components, no orchestration
Models	Own model per agent	Codex CLI model
Team Mode	Up to 8 participants
Lifecycle hooks	54+	(limited)
MCP	5 built-in + custom	Plugin-scoped MCP
Installation	bunx oh-my-openagent install	npx lazycodex-ai install
Requirements	Bun + OpenCode	Node.js + Codex CLI

Name History

The project has been renamed several times:

1. oh-my-opencode — original name (npm package still named this)
2. oh-my-openagent — current name (since March 2026)
3. lazycodex — Codex CLI version
4. omo — internal abbreviation (don’t confuse with npm package omo, that’s a different project!)

3. Other Systems Approaching This

No full analogs of OhMyOpenAgent exist, but some systems have individual multi-agent elements.

Claude Code (sub-agents)

GitHub → · 139K+

What it has: Claude Code supports sub-agents — can spawn child agents for parallel task execution. Has CLAUDE.md for instructions, MCP servers.

What it lacks: No role separation (architect vs executor vs researcher). One model for everything. No planner separate from executor.

Cline — Kanban mode

GitHub → · 65K+

What it has: Kanban board (npx kanban) for parallel launch of multiple agents. Each agent works on its task independently. MCP server support.

What it lacks: No orchestrator to distribute tasks. No different models for different roles. No planner.

Aider — Architect/Editor mode

GitHub → · 48K+

What it has: Two-model mode: one (architect) plans changes, second (editor) writes code. Ollama support.

What it lacks: Only two agents, no role specialization. No sub-agents for research/documentation.

OpenHands — Agent Canvas

GitHub → · 82K+

What it has: Web UI for launching multiple agents. ACP protocol for coordination. Different model support.

What it lacks: Each agent is a copy of the same pipeline, no role separation. No planner.

4. Comparison Table

System	Agents	Role Separation	Different Models	Planner	Editions
OhMyOpenAgent	11	Yes	Yes	Prometheus	2 (OpenCode/Codex CLI)	66K
Claude Code	sub-agents	No	No	No	1	139K
Cline	N (kanban)	No	No	No	1	65K
Aider	2	Two roles	Two models	No	1	48K
OpenHands	N (UI)	No	Different	No	1	82K
CrewAI*	N (code)	Roles	Different	Via code	Framework	25K
AutoGen*	N (code)	Roles	Different	Via code	Framework	38K

*CrewAI and AutoGen are frameworks, not ready-to-use systems. They require writing code to configure the team.

5. Installing OhMyOpenAgent

Method 1: Ultimate Edition (OpenCode) — Recommended

# Install via Bun
bunx oh-my-openagent install

Interactive TUI (Terminal UI) will guide you through:

1. Subscription selection (ChatGPT, Kimi Code, GLM Coding Plan)
2. Provider setup (Anthropic, Gemini, Copilot, Z.ai)
3. Model assignment to 11 agents
4. Team Mode configuration

Recommended subscriptions:

• ChatGPT ($20/mo) — for GPT-5.6 Sol
• Kimi Code ($19/mo) — for Kimi K3 (excellent Claude alternative)
• GLM Coding Plan ($10/mo) — for GLM-5

Method 2: Light Edition (Codex CLI)

# Install via npm (Node.js, no Bun required)
npx lazycodex-ai install

# Non-TUI mode with autonomous permissions
npx lazycodex-ai install --no-tui --codex-autonomous

Installs 8 components to ~/.codex/plugins/cache/sisyphuslabs/omo/.

Method 3: Both Editions at Once

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

Installation via AI Agent (Recommended by Authors)

Copy this prompt into Claude Code, Cursor, AmpCode, or any other agent:

Install and configure oh-my-openagent by following the instructions here:
https://raw.githubusercontent.com/code-yeongyu/oh-my-openagent/refs/heads/dev/docs/guide/installation.md

The agent will read the full guide, execute each step, and configure everything automatically.

Important Notes

1. Package name: npm package is still called oh-my-opencode (dual-published as oh-my-openagent). Command bunx oh-my-openagent install works.
2. Don’t confuse omo: package uses omo as bin alias, but DO NOT use npx omo — that’s a different npm package.
3. OpenCode vs Codex CLI: Ultimate edition requires OpenCode. Light edition works with official Codex CLI from OpenAI.
4. Telemetry: anonymous telemetry enabled by default (1 request/day). Disable: "telemetry": false in config.

After Installation

# Check status
bunx oh-my-openagent doctor

# Basic command — just write:
ulw <your task>

# Or precise path:
@plan <task description>   # Prometheus plans
/start-work               # Atlas executes

6. What’s Next

In section: architecture · evaluation · frameworks · memory · multi-agent · ollama-for-agents · orchestrators · patterns · prompting · ready-made · safety · skills
Related sections: Zero Level · Local Models · Use Cases · Resources
Navigation: ← AI Agents · ↑ Back to main · 🇷🇺 Русский