Oh My OpenAgent: The Open Source Multi-Model Agent Orchestration Revolution
Oh My OpenAgent: The Open Source Multi-Model Agent Orchestration Revolution
Introduction
"It made me cancel my Cursor subscription. Unbelievable things are happening in the open source community." — Arthur Guiot
In the AI programming tools space, one project is quietly changing how developers work. As of 2026, it has garnered 67,953 GitHub stars, 5,547 forks, making it one of the most followed open source projects globally. This is Oh My OpenAgent (OmO for short).
This article takes you through an in-depth understanding of the project's design philosophy, core features, Agent system architecture, and how to get started quickly.
I. Project Overview
1.1 What Is Oh My OpenAgent?
Oh My OpenAgent is a Multi-Model Agent Orchestration Harness that transforms a single AI programming assistant into a coordinated development team that actually ships code.
Its core characteristics:
- Not locked to any single model: Supports Claude, GPT, Kimi, GLM, and more
- Not locked to any single platform: Supports OpenCode, Codex CLI, Pi, and other runtimes
- True agent orchestration: Not simple model switching, but professional agents working together
- Open source transparency: Fully open source code, community-driven development
1.2 Project Scale & Impact
| Metric | Data |
|---|---|
| GitHub Stars | 67,953 |
| Forks | 5,547 |
| Primary Language | TypeScript |
| License | SUL-1.0 |
| Default Branch | dev |
1.3 User Reviews
"If Claude Code does in 7 days what a human does in 3 months, Sisyphus does it in 1 hour. It just works until the task is done. It is a discipline agent." — B, Quant Researcher
"Knocked out 8000 eslint warnings with Oh My Opencode, just in a day." — Jacob Ferrari
"I converted a 45k line tauri app into a SaaS web app overnight using Ohmyopencode." — James Hargis
II. Design Philosophy: Breaking Free
2.1 Core Philosophy: Reject Closed, Embrace Open
The project team describes their philosophy:
"We used to call this 'Claude Code on steroids.' That was wrong."
"This isn't about making Claude Code better. It's about breaking free from the idea that one model, one provider, one way of working is enough. Anthropic wants you locked in. OpenAI wants you locked in. Everyone wants you locked in."
"Oh My OpenAgent doesn't play that game. It orchestrates across models, picking the right brain for the right job. Opus 5 for orchestration and visual work. GPT-5.6 Sol for deep reasoning. Kimi K3 and GLM 5.2 as visual fallbacks. Kimi high-speed for quick tasks. All working together, automatically."
2.2 Why Multi-Model Orchestration?
Limitations of single models:
- Different models have different strengths on different tasks
- Some models perform better in specific domains
- Choosing the right model can significantly reduce costs with usage-based pricing
- Avoid vendor lock-in
OmO's answer:
User Request
↓
[IntentGate] — Analyzes your true intent
↓
[Sisyphus] — Main orchestrator, plans and delegates
↓
├─→ [Prometheus] — Strategic planning (interview mode)
├─→ [Atlas] — Todo orchestration and execution
├─→ [Oracle] — Architecture consultation
├─→ [Librarian] — Documentation/code search
└─→ [Explore] — Fast codebase grep
2.3 The Concept of "Discipline Agent"
The project team proposes the concept of Discipline Agent:
- Not: A passive tool that does whatever you tell it
- Is: A self-disciplined worker with goals, plans, and execution strategies
- Characteristics: Never gives up halfway, never gets distracted, never stops until goal is achieved
III. Core Features in Detail
3.1 ultrawork: One-Button Smart Workflow
Usage: Simply type ultrawork or ulw in conversation
ultrawork
Workflow:
- Explore codebase structure
- Research existing patterns and best practices
- Formulate implementation plan
- Execute code writing
- Run diagnostic verification
- Iterate continuously until task is complete
Supported services (personal recommendations):
| Service | Price | Recommendation |
|---|---|---|
| ChatGPT Subscription | $20/mo | Mature and stable |
| Kimi Code Subscription | $19/mo | Excellent Chinese support |
| GLM Coding Plan | $10/mo | High cost performance |
3.2 Discipline Agents
OmO comes with multiple specialized agents, each optimized for specific tasks:
Sisyphus — Main Orchestrator
Role: Main coordinator, responsible for planning, delegating tasks, driving completion
Recommended models:
- Claude Opus 5 (best overall experience)
- Kimi K3 (strongest Kimi model)
- Kimi K2.7 (restrained fallback)
- GLM 5.2 (via OpenCode Go)
Characteristics:
- Never gives up halfway
- Never gets distracted
- Doesn't stop until done
Hephaestus — The Legitimate Craftsman
Role: Autonomous deep worker
Origin of ironic name: Anthropic blocked OpenCode from using their API because of this project, so the team intentionally named this GPT-native autonomous agent "The Legitimate Craftsman"
Recommended model:
- GPT-5.6 Sol (via OpenAI, GitHub Copilot, Vercel, or OpenCode)
Use cases:
- When deep architectural reasoning is needed
- Complex cross-file debugging
- Cross-domain knowledge synthesis
Prometheus — Strategic Planner
Role: Strategic planner, works through interview mode
Workflow:
- Ask questions to understand user needs
- Identify scope and ambiguities
- Build detailed plan before touching any code
Activation: Press Tab key to enter Prometheus mode
Atlas — Execution Conductor
Role: Executes Prometheus's plans
Responsibilities:
- Distribute tasks to specialized subagents
- Accumulate learnings across tasks
- Independently verify completion
3.3 Agent Category System
When Sisyphus delegates to subagents, it doesn't pick a specific model — it picks a category:
| Category | Suitable Tasks | Default Model |
|---|---|---|
visual-engineering |
Frontend, UI/UX, design | Claude Opus 5 max → Kimi K3 |
ultrabrain |
Complex logic, architecture | GPT-5.6 Sol xhigh |
deep |
Autonomous research & execution | GPT-5.6 Sol medium |
artistry |
Art/creative related | Claude Fable 5 |
quick |
Quick single-file changes | Kimi high-speed |
unspecified-low |
Low priority unspecified | Grok 4.6 |
unspecified-high |
High priority unspecified | Kimi K3 |
3.4 IntentGate
Function: Analyzes user's true intent before taking action
Problems solved:
- Misunderstandings from unclear user descriptions
- Execution in the wrong direction from mechanical interpretation
- Lack of contextual understanding
3.5 Hashline: Hash-Based Editing Tool
Inspired by: oh-my-pi
Core idea: Most "Agent failures" are not the model being dumb — it's the file editing tools being terrible.
"Currently all tools fail to provide models with a stable, verifiable line positioning identifier... They all rely on the model to re-copy the text it just saw. When the model makes a mistake — and it happens often — users blame the LLM for being stupid." — Can Bölük, The Harness Problem
Hashline solution:
# When Agent reads files, each line has a hash at the end
11#VK| function hello() {
22#XJ| return "world";
33#MB| }
How it works:
- Each modification is validated through
LINE#IDcontent hash - If the file has changed, hash verification fails
- Modification is rejected before code gets polluted
- No more indentation errors, no more editing wrong lines
Result: On Grok Code Fast 1, just by changing this editing tool, modification success rate jumped from 6.7% to 68.3%.
3.6 Built-in MCP Servers
| MCP | Purpose |
|---|---|
| Exa | Web search |
| Context7 | Official docs query |
| Grep.app | GitHub code search |
IV. Team Mode: True Multi-Agent Collaboration
4.1 What Is Team Mode?
Team Mode upgrades OmO from "one agent with subagents" to a true multi-agent system.
Core features:
- Lead agent + up to 8 parallel members
- Real-time tmux visualization
- Dedicated
team_*tool family - Members communicate via mailbox mechanism
4.2 Team Configuration Example
// .opencode/oh-my-openagent.jsonc
{
"team_mode": {
"enabled": true,
"max_parallel_members": 4,
"tmux_visualization": true
}
}
4.3 Built-in Team Skills
hyperplan — Five-Fold Hostile Review
5 hostile agents tear apart your plan from orthogonal angles:
- Security perspective
- Performance perspective
- Maintainability perspective
- Business logic perspective
- Edge case perspective
security-research — Security Research Team
3 vulnerability hunters + 2 PoC engineers audit your codebase in parallel, severity calibrated by actual exploitability.
4.4 Team Lifecycle Tools
| Tool | Purpose |
|---|---|
team_create |
Create team |
team_delete |
Destroy team |
team_shutdown_request |
Request member shutdown |
team_send_message |
Point-to-point/broadcast messages |
team_task_create |
Create shared task |
team_task_update |
Update task status |
team_status |
View team status |
V. Installation Guide
5.1 Three Edition Choices
| Edition | Install Command | Use Case |
|---|---|---|
| Ultimate | bunx oh-my-openagent install |
OpenCode users |
| Light | npx lazycodex-ai install |
Codex CLI users |
| Senpi (Beta) | npm i -g omo-ai@beta |
Users who don't want to install a host |
5.2 Recommended: Let AI Install It
Strongly recommended: Let an LLM Agent install it for you. Ultimate installation involves subscription detection, model selection across 11 agents, provider authentication — humans make mistakes.
Installation prompt:
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
5.3 Manual Installation (Ultimate)
# Install
bunx oh-my-openagent install
# Run health check
bunx oh-my-openagent doctor
5.4 Manual Installation (Light — Codex CLI)
# Recommended: auto-configure autonomous mode
npx lazycodex-ai install --no-tui --codex-autonomous
# Or normal installation
npx lazycodex-ai install
5.5 Telemetry & Privacy
Enabled by default for tracking active installations (DAU/WAU/MAU).
- Maximum one event per UTC day per machine
- Uses hashed installation identifier (never raw hostname)
- No PostHog person profiles created
Disable telemetry:
# Disable main plugin telemetry
OMO_DISABLE_POSTHOG=1
# Disable Codex CLI telemetry
OMO_CODEX_DISABLE_POSTHOG=1
VI. Usage Tutorial
6.1 Quick Start
- After installation, type in OpenCode or Codex CLI:
ultrawork
- Describe your task, for example:
ultrawork
Help me migrate this React project from Create React App to Vite
- The system automatically completes all work until the task is done.
6.2 Precision Mode (Prometheus Mode)
If you want more control:
- Press Tab to enter Prometheus mode
- Prometheus interviews you like a real engineer
- Ask questions, clarify scope, build detailed plan
- Run
/start-workto start Atlas executing the plan
6.3 Team Mode Usage
- Enable Team Mode in config
- Restart OpenCode
- Use
team_createto create a team - Team automatically executes tasks in parallel
VII. Comparison with Other Tools
7.1 vs Claude Code
| Aspect | Claude Code | OmO |
|---|---|---|
| Model binding | Anthropic exclusive | Multi-model support |
| Multi-model orchestration | Not supported | Supported |
| Team Mode | Limited | Full implementation |
| Background parallel agents | Not supported | 5+ parallel |
| Open source | No | Yes |
7.2 vs Vanilla Codex CLI
| Aspect | Vanilla Codex | OmO Light |
|---|---|---|
| Multi-model orchestration | Not supported | Supported |
| Background agents | Not supported | Supported |
| Team Mode | Not supported | Supported |
| Rules injection | Limited | Full |
| Open source | Partial | Complete |
VIII. Architecture Design Analysis
8.1 Layered Architecture
OmO uses layered design for easy cross-host reuse:
┌─────────────────────────────────────┐
│ Agent Layer │
├─────────────────────────────────────┤
│ Skills Layer │
├─────────────────────────────────────┤
│ MCP Layer │
├─────────────────────────────────────┤
│ Core Layer │
├─────────────────────────────────────┤
│ Adapter Layer │
└─────────────────────────────────────┘
8.2 Why This Architecture?
Ongoing refactoring: Separating pure TypeScript core logic, MCP servers, skills, and adapter shims into distinct layers to:
- Reuse logic across harnesses without duplication
- Support OpenCode, Codex, Pi, Claude Code, and more
- Enable community contribution and maintenance
IX. Summary & Outlook
9.1 Key Insights Summary
Insight 1: Multi-Model Collaboration Is the Future
"The future isn't picking one winner; it's orchestrating them all. Models get cheaper every month. Smarter every month. No single provider will dominate."
The era of single-model dominance is fading; multi-model collaboration is the trend.
Insight 2: Tool Chain Quality Determines AI Capability Ceiling
"The LLM being dumb" is often a misunderstanding. The real problem is the quality of the tool chain (Harness).
Improvements like Hashline can increase modification success rate by 10x.
Insight 3: Discipline Agent > Passive Tool
A good AI programming assistant should not be a passive tool that follows orders blindly, but should be able to:
- Understand true intent
- Formulate execution plans
- Autonomously complete tasks
- Iterate continuously until done
Insight 4: Open Source Breaks Monopoly
"Anthropic blocked OpenCode because of us. Yes, this is true. They want you locked in. Claude Code is a nice prison, but it's still a prison."
The power of open source is breaking the closed ecosystem in AI, giving users real choice.
9.2 Applicable Scenarios
Very suitable for:
- Projects requiring deep code exploration and refactoring
- Large codebases with multi-member collaboration
- Teams sensitive to costs but needing high-quality results
- Developers wanting to avoid vendor lock-in
Not very suitable for:
- Simple single-file modifications (overkill)
- Complete beginners unfamiliar with AI programming
- Environments with restricted network access
9.3 Future Outlook
The project is undergoing Multi-Harness Agent OS Refactoring, planning to support:
- OpenCode
- Codex
- Pi
- Claude Code
- More hosts
This will make OmO a true "universal agent orchestration layer."
X. Quick Reference
Installation Command Summary
# Ultimate (OpenCode)
bunx oh-my-openagent install
# Light (Codex CLI)
npx lazycodex-ai install
# Both editions
bunx oh-my-openagent install --platform=both
# Senpi standalone
npm i -g omo-ai@beta
Common Commands
| Command | Purpose |
|---|---|
ultrawork or ulw |
One-button start all agents |
| Press Tab | Enter Prometheus planning mode |
/start-work |
Start Atlas executing plan |
/init-deep |
Generate project AGENTS.md |
Resource Links
| Resource | Link |
|---|---|
| GitHub Repo | https://github.com/code-yeongyu/oh-my-openagent |
| Official Docs | https://omo.vibetip.help/docs |
| Discord Community | https://discord.gg/PUwSMR9XNk |
| LazyCodex (Codex Edition) | https://lazycodex.ai |
Conclusion
Oh My OpenAgent is not just a programming assistant — it represents a new philosophy: reject closed, embrace open; reject single, embrace collaboration; reject passive, embrace self-discipline.
In the AI programming tools赛道, it is breaking the monopoly of tech giants with an open source attitude, providing developers with truly free choice.
If you yearn to break free from single-model limitations, if you want a truly collaborative AI development team, if you believe in the power of open source — Oh My OpenAgent is worth trying.
"Type
ultrawork. Done."
About Author
ERIC — Co-author of "Blockchain Core Technology and Applications", former Huobi Institution Division/Mining Pool Technical Lead, Bit Finance/Nxt Venture Capital Founder
Share on Social Media
📱 Share this article on X (Twitter)
🐦 Share on X.com →About the Author
ERIC
AI Technology Expert, focusing on research and application of artificial intelligence and automation tools
Contact & Platforms