Multi-AI Workflow (MAW)
Unleash the Power of AI Collaboration
__ __ _ __ __
| \/ | /_\ \ \ / /
| |\/| |/ _ \ \ \ /\ / /
| | | / ___ \ \ V V /
|_| |_/_/ \_\_\_/\_/
Multi-AI Workflow Framework
Claude + Codex + Gemini = Unstoppable
Features | Installation | Quick Start | Workflow Modes | Usage in Claude Code
What is MAW?
MAW (Multi-AI Workflow) is a revolutionary CLI framework that orchestrates multiple AI agents to work together seamlessly. Imagine having Claude's reasoning, Codex's code execution, and Gemini's multimodal capabilities all working in harmony on your projects.
┌──────────────────────────────────────────────────────────────┐
│ │
│ [CLAUDE] [CODEX] [GEMINI] │
│ Planning & Code Execution Multimodal │
│ Reasoning & Analysis Analysis │
│ │
│ \ | / │
│ \ | / │
│ \ | / │
│ v v v │
│ +-------------------------------+ │
│ | MAW ORCHESTRATOR | │
│ | Unified Session Control | │
│ +-------------------------------+ │
│ | │
│ v │
│ +-----------------+ │
│ | YOUR CODE | │
│ +-----------------+ │
│ │
└──────────────────────────────────────────────────────────────┘
Features
Multi-AI Delegation
Delegate tasks to the right AI for the job. Claude plans, Codex executes, Gemini analyzes.
7 Workflow Modes
| Level | Description |
|---|---|
| Lite | Instant execution, no planning |
| Lite-Plan | Quick planning then execute |
| Plan | Standard planning workflow |
| TDD-Plan | Test-driven development |
| Brainstorm | Multi-role ideation |
| Five-Phase | Professional 5-phase collaboration |
| Ralph Loop | Iterative AI loop until completion |
Smart Routing
Auto-detect the best AI for your task based on content analysis.
Report Generation
Generate professional reports with auto-generated diagrams.
Real-time Dashboard
Visual workflow management with WebSocket live updates.
Unified Sessions
Single session ID syncs across all AI agents.
Installation
Prerequisites
- Node.js >= 18.0.0
- Python >= 3.9
- Claude Code CLI
- Codex CLI (optional)
- Gemini CLI (optional)
Quick Install (Recommended)
# Clone and run the installer
git clone https://github.com/haoyu-haoyu/Multi-AI-Workflow.git
cd Multi-AI-Workflow
./install.sh
# Restart terminal or source your shell config
source ~/.zshrc # or ~/.bashrc
# Verify installation
maw --version
The installer will:
- Install MAW CLI to
~/.maw/ - Set up global
mawcommand - Install Claude Code slash commands
- Configure your PATH
Manual Installation
# 1. Clone the repository
git clone https://github.com/haoyu-haoyu/Multi-AI-Workflow.git
cd Multi-AI-Workflow
# 2. Install Node.js dependencies
npm install
# 3. Build the project
npm run build
# 4. Install Python bridges
pip install -e bridges/
# 5. Copy to global directory
mkdir -p ~/.maw/bin
cp -r maw ~/.maw/
# 6. Create global command
echo '#!/bin/bash
node ~/.maw/maw/bin/maw.js "$@"' > ~/.maw/bin/maw
chmod +x ~/.maw/bin/maw
# 7. Add to PATH (add to ~/.zshrc or ~/.bashrc)
echo 'export PATH="$HOME/.maw/bin:$PATH"' >> ~/.zshrc
# 8. Install slash commands for Claude Code
mkdir -p ~/.claude/commands
cp claude-commands/*.md ~/.claude/commands/
# 9. Restart terminal and verify
maw --help
Installation Structure
After installation, your ~/.maw directory will look like:
~/.maw/
├── bin/
│ └── maw # Global command
├── maw/ # MAW CLI
│ ├── bin/maw.js
│ └── dist/
└── skills/ # Installed skills
├── collaborating-with-gemini/
├── collaborating-with-codex/
└── report-generator/
Configure API Keys (Optional)
For Gemini with proxy API:
# Edit the bridge configuration
# File: ~/.maw/skills/collaborating-with-gemini/scripts/gemini_bridge.py
# Set your API key and base URL
Quick Start
# Smart routing - auto-selects best AI
maw run "Analyze this codebase structure"
# Run a simple workflow
maw workflow lite "Fix typos in README"
# Standard planning workflow
maw workflow plan "Implement user authentication"
# 5-phase professional workflow
maw workflow five-phase "Refactor the authentication module"
# Delegate to specific AI
maw delegate codex "Write unit tests" --cd .
maw delegate gemini "Analyze this architecture"
# Open the dashboard
maw view
Workflow Modes
1. Lite Mode - Instant Execution
User Task ────────► Direct Execute ────────► Result
Best for: Simple fixes, small changes, one-time tasks
maw workflow lite "Add comments to this function"
maw workflow lite "Format this file"
maw workflow lite "Change all console.log to logger.debug"
2. Lite-Plan Mode - Quick Planning
User Task ────► Quick Plan (1-2 steps) ────► Execute ────► Result
Best for: Medium complexity tasks, quick development
maw workflow lite-plan "Add error handling to API"
maw workflow lite-plan "Implement simple caching"
3. Plan Mode - Standard Planning (Default)
User Task
│
▼
┌──────────┐
│ Plan │ ← Analyze requirements, design solution
└────┬─────┘
│
▼
┌──────────┐
│ Review │ ← Check plan feasibility
└────┬─────┘
│
▼
┌──────────┐
│ Execute │ ← Implement step by step
└────┬─────┘
│
▼
┌──────────┐
│ Verify │ ← Confirm results
└────┬─────┘
│
▼
Result
Best for: Daily development, standard features
maw workflow plan "Implement user login"
maw workflow plan "Add database connection pool"
maw workflow plan "Refactor error handling"
4. TDD-Plan Mode - Test-Driven Development
User Task
│
▼
┌────────────────┐
│ Write Tests │ ← RED: Define expected behavior
│ (RED) │
└───────┬────────┘
│
▼
┌────────────────┐
│ Run Tests │ ← Confirm tests fail
│ (Confirm Fail) │
└───────┬────────┘
│
▼
┌────────────────┐
│ Implement │ ← GREEN: Write minimal code to pass
│ (GREEN) │
└───────┬────────┘
│
▼
┌────────────────┐
│ Refactor │ ← REFACTOR: Optimize with test safety
│ (REFACTOR) │
└───────┬────────┘
│
▼
Done (with tests)
Best for: Core business logic, APIs, critical features
maw workflow tdd-plan "Implement order calculation"
maw workflow tdd-plan "Add user permission validation"
maw workflow tdd-plan "Implement payment interface"
5. Brainstorm Mode - Multi-Role Ideation
User Topic
│
▼
┌─────────────────────────────────────────┐
│ Multi-Role Discussion │
│ │
│ ┌──────────┐ ┌──────────┐ ┌────────┐ │
│ │ Architect│ │Developer │ │ QA │ │
│ └────┬─────┘ └────┬─────┘ └───┬────┘ │
│ │ │ │ │
│ └──────────┬──┴────────────┘ │
│ │ │
│ ▼ │
│ Discussion & Debate │
└─────────────────────────────────────────┘
│
▼
┌──────────────┐
│ Conclusion │
│ Best Solution│
└──────────────┘
Best for: Architecture decisions, technical choices, complex analysis
maw workflow brainstorm "API design approach"
maw workflow brainstorm "Database selection"
maw workflow brainstorm "Performance optimization strategy"
Roles:
| Role | Focus |
|---|---|
| Architect | Overall design, scalability, tech debt |
| Developer | Implementation difficulty, efficiency, code quality |
| QA | Testability, edge cases, potential issues |
6. Five-Phase Mode - Professional Multi-AI Collaboration
User Task
│
▼
┌─────────────────────────────────────────────────────────┐
│ Phase 1: CONTEXT [Claude] │
│ - Analyze project structure │
│ - Understand existing code │
│ - Gather relevant information │
└─────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────┐
│ Phase 2: ANALYSIS [Codex + Gemini] │
│ │
│ ┌─────────────┐ ┌─────────────┐ │
│ │ Codex │ Parallel │ Gemini │ │
│ │ Code-level │◄────────────►│ Arch-level │ │
│ │ Analysis │ │ Analysis │ │
│ └─────────────┘ └─────────────┘ │
└─────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────┐
│ Phase 3: PROTOTYPE [Claude] │
│ - Synthesize analysis results │
│ - Design initial solution │
│ - Create code skeleton │
└─────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────┐
│ Phase 4: IMPLEMENT [Claude] │
│ - Complete code implementation │
│ - Handle edge cases │
│ - Add error handling │
└─────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────┐
│ Phase 5: AUDIT [Claude+Codex+Gemini] │
│ - Code review │
│ - Security check │
│ - Final verification │
└─────────────────────────────────────────────────────────┘
│
▼
Complete (High-quality, multi-AI verified)
Best for: Large features, system refactoring, critical modules
maw workflow five-phase "Implement complete payment system"
maw workflow five-phase "Refactor database access layer"
maw workflow five-phase "Implement OAuth2 authentication"
7. Ralph Loop Mode - Iterative AI Development
Ralph Loop is inspired by the "Ralph technique" - a continuous AI agent loop for iterative development until task completion.
┌──────────────────────────────────────┐
│ RALPH LOOP │
│ "Persistence Wins" │
└──────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────────┐
│ │
│ ┌──────────┐ │
│ │ Prompt │ ◄───────────────────────────────────┐ │
│ │ (Fixed) │ │ │
│ └────┬─────┘ │ │
│ │ │ │
│ ▼ │ │
│ ┌──────────┐ │ │
│ │ AI │ ← Claude/Codex/Gemini/Auto │ │
│ │ Execute │ │ │
│ └────┬─────┘ │ │
│ │ │ │
│ ▼ │ │
│ ┌──────────┐ No ┌───────────────────┐ │ │
│ │ Complete?├────────────►│ Next Iteration │──┘ │
│ │ │ │ (AI sees its │ │
│ └────┬─────┘ │ previous work) │ │
│ │ Yes └───────────────────┘ │
│ ▼ │
│ ┌──────────┐ │
│ │ Done │ ← Completion Promise Found │
│ └──────────┘ │
│ │
└─────────────────────────────────────────────────────────────────────┘
Best for: Tasks with clear success criteria, iterative refinement, automated verification
# Basic usage
maw workflow ralph "Build a REST API for todos. Output <promise>COMPLETE</promise> when done."
# With options
maw workflow ralph "Implement user auth with tests" \
--max-iterations 30 \
--completion-promise "COMPLETE" \
--ai auto \
--verbose
Key Concepts:
| Concept | Description |
|---|---|
| Completion Promise | A signal phrase that indicates task completion |
| Iteration | Each loop cycle where AI sees its previous work |
| Persistence | Keep trying until success (within max iterations) |
| Self-Correction | AI learns from its own failures in previous iterations |
When to Use:
- ✅ Tasks with clear, testable success criteria
- ✅ Tasks requiring iterative improvement
- ✅ Tasks with automatic verification (tests, linters)
- ❌ Tasks requiring human judgment
- ❌ Tasks with unclear completion criteria
Mode Comparison
| Mode | Complexity | Time | AI Involved | Use Case |
|---|---|---|---|---|
| Lite | ⭐ | Shortest | Claude | Simple fixes |
| Lite-Plan | ⭐⭐ | Short | Claude | Medium tasks |
| Plan | ⭐⭐⭐ | Medium | Claude | Daily development |
| TDD-Plan | ⭐⭐⭐⭐ | Longer | Claude | Test-covered features |
| Brainstorm | ⭐⭐⭐ | Medium | Claude (multi-role) | Design decisions |
| Five-Phase | ⭐⭐⭐⭐⭐ | Longest | Claude+Codex+Gemini | Complex, high-quality |
| Ralph Loop | ⭐⭐⭐⭐ | Variable | Auto-selected | Iterative tasks |
How to Choose?
What's your task?
│
┌───────────────┼───────────────┐
│ │ │
▼ ▼ ▼
Simple fix? Development? Design decision?
│ │ │
▼ │ ▼
Lite │ Brainstorm
│
┌──────────┴──────────┐
│ │
▼ ▼
Need tests? Need multi-AI?
│ │
┌────┴────┐ ┌────┴────┐
▼ ▼ ▼ ▼
TDD-Plan Plan Five-Phase Plan
Usage in Claude Code
Slash Commands
MAW provides slash commands for Claude Code:
| Command | Description |
|---|---|
/gemini <task> | Delegate to Gemini |
/codex <task> | Delegate to Codex |
/maw <command> | Execute MAW command |
/report <topic> | Generate report with diagrams |
/ralph-loop <prompt> | Start iterative AI loop |
Examples:
/gemini Analyze the performance of this code
/codex Write unit tests for the auth module
/maw workflow plan "Implement caching layer"
/ralph-loop "Build REST API with tests. Output COMPLETE when done."
/report "System Architecture Overview"
Natural Language
You can also use natural language with Claude:
| You Say | Claude Does |
|---|---|
| "Let Gemini analyze this..." | Calls Gemini bridge |
| "Delegate to Codex..." | Calls Codex bridge |
| "Use MAW to..." | Starts MAW workflow |
| "Generate a report about..." | Runs report generator |
Examples:
"Let Gemini analyze this module's architecture"
"Delegate to Codex to write tests for this function"
"Use MAW five-phase workflow to refactor authentication"
"Generate a report about our API design with diagrams"
Multi-AI Collaboration Triggers
"Use multiple AIs to collaborate on..."
"Let Claude plan and Codex execute..."
"Start five-phase collaboration workflow..."
Report Generation
MAW can generate professional reports with auto-generated diagrams:
# Using slash command
/report "Machine Learning Pipeline Architecture"
# Using Python script directly
python ~/.maw/skills/report-generator/report_generator.py \
--topic "System Design" \
--content "Your research content..." \
--output report.md
Features:
- Auto-analyzes content structure
- Identifies where diagrams would help
- Generates Mermaid diagrams (or images when available)
- Professional academic style output
Architecture
Multi-AI-Workflow/
│
├── maw/ # TypeScript CLI Core
│ ├── src/
│ │ ├── cli.ts # Main CLI entry
│ │ ├── commands/ # Command implementations
│ │ │ ├── delegate.ts # AI delegation + semantic routing
│ │ │ ├── workflow.ts # Workflow execution
│ │ │ └── session.ts # Session management
│ │ └── core/ # Core modules
│ │ ├── workflow-engine.ts # 6-level workflows
│ │ ├── session-manager.ts
│ │ └── skill-registry.ts
│ └── bin/maw.js # CLI executable
│
├── bridges/ # Python AI Bridges
│ └── src/maw_bridges/
│ ├── codex_bridge.py # Codex CLI wrapper
│ └── gemini_bridge.py # Gemini CLI/API wrapper
│
├── dashboard/ # Web Dashboard
│ ├── src/
│ │ ├── server.ts # Express + WebSocket
│ │ ├── storage.ts # SQLite database
│ │ └── maw-bridge.ts # CLI data integration
│ └── public/ # Frontend SPA
│
└── .maw/ # Configuration & Skills
├── config.json
└── skills/ # Installed skills
Configuration
MAW configuration is stored in .maw/config.json:
{
"ai": {
"claude": { "enabled": true },
"codex": { "enabled": true, "cliPath": "codex" },
"gemini": { "enabled": true, "cliPath": "gemini" }
},
"workflow": {
"defaultLevel": "plan",
"parallelExecution": true,
"maxConcurrency": 2
},
"security": {
"defaultSandbox": "read-only",
"maxExecutionTime": 300000
},
"dashboard": {
"port": 3000,
"autoOpen": true
}
}
Sandbox Levels
| Level | Description | Use Case |
|---|---|---|
read-only | Cannot modify any files | Code analysis, auditing |
workspace-write | Can modify project files | Daily development |
full-access | Complete control | System-level operations |
Commands Reference
Workflow Commands
maw workflow lite <task> # Instant execution
maw workflow lite-plan <task> # Quick plan + execute
maw workflow plan <task> # Standard planning
maw workflow tdd-plan <task> # Test-driven development
maw workflow brainstorm <topic> # Multi-role ideation
maw workflow five-phase <task> # 5-phase collaboration
maw workflow ralph <prompt> # Iterative AI loop
# Ralph Loop options
maw workflow ralph "Build feature X" \
--max-iterations 30 \
--completion-promise "COMPLETE" \
--ai auto \
--verbose
Delegation Commands
maw delegate codex <task> --cd . # Delegate to Codex
maw delegate gemini <task> # Delegate to Gemini
maw run <task> # Smart routing
Session Commands
maw session list # List all sessions
maw session new <name> # Create new session
maw session resume <id> # Resume session
maw session sync # Sync across AIs
Other Commands
maw view # Open dashboard
maw skill list # List installed skills
maw skill install <source> # Install skill
Contributing
We welcome contributions! Please see our contributing guidelines.
# Fork the repository
git checkout -b feature/amazing-feature
git commit -m 'Add amazing feature'
git push origin feature/amazing-feature
# Open a Pull Request
License
This project is licensed under the MIT License - see the LICENSE file for details.
Star us on GitHub!
If you find MAW useful, please consider giving us a star.
Built by haoyu-haoyu
Making AI collaboration accessible to everyone