BMAD-BUFF System 🚀

Revolutionary Multi-LLM Orchestration for 10x Faster AI-Driven Development

🎯 What is BMAD-BUFF?

BMAD-BUFF is an advanced orchestration system built on top of the BMAD-METHOD™ that intelligently distributes development tasks across multiple specialized LLMs (Claude, GPT, Gemini, Perplexity) to achieve unprecedented development speed and quality.

Key Features

• 🧠 Multi-LLM Orchestration: Automatically routes tasks to the most suitable LLM
• ⚡ 10x Faster Development: Parallel execution across multiple AI providers
• 💰 Cost Optimization: Uses cheaper models for simple tasks, premium for complex
• 🔄 ROMA Integration: Recursive task decomposition and orchestration
• 🎭 BUFF Routing: Intelligent load balancing across providers
• 🚀 YOLO Mode: Skip confirmations while maintaining quality
• 📊 Vector Database: Enhanced context with ChromaDB/Vectra

🏗️ Architecture

BMAD-BUFF System
├── ROMA Orchestrator (Task Decomposition)
├── BUFF Router (LLM Selection)
├── Agent Spawner (Specialized Agents)
└── Vector Database (Context Enhancement)

📦 Installation

npm install -g bmad-buff-system

Or clone and install locally:

git clone https://github.com/papuman/bmad-buff-system.git
cd bmad-buff-system
npm install
npm run setup

🔧 Configuration

1. Create a .env file in expansion-packs/bmad-buff/:

# Required API Keys
OPENAI_API_KEY=your_openai_key
GEMINI_API_KEY=your_gemini_key
PERPLEXITY_API_KEY=your_perplexity_key

# Optional
ANTHROPIC_API_KEY=your_anthropic_key  # Only if not using Claude Code

2. The system automatically detects and uses available LLMs based on configured API keys.

🚀 Usage

Basic Usage

const { BMADWorkflow } = require('bmad-buff-system');

const workflow = new BMADWorkflow({
  project: 'Build AI-powered app',
  requirements: ['auth', 'api', 'frontend', 'ml-pipeline']
});

await workflow.execute();

CLI Usage

# Start BMAD workflow
bmad start "Build a social media analytics platform"

# Enable YOLO mode (skip confirmations)
bmad start --yolo "Create REST API with auth"

# Use specific LLM strategy
bmad start --strategy multi-llm "Complex project"

🎯 LLM Specialization

BMAD-BUFF automatically assigns tasks to specialized LLMs:

| Task Type | Preferred LLM | Why | |-----------|--------------|-----| | Code Implementation | Claude Opus | Best for complex coding | | Architecture Design | Gemini Pro | Deep thinking & planning | | Strategic Planning | GPT-5 | Advanced reasoning | | Research & Docs | Perplexity | Real-time web search | | Quick Validation | Gemini Flash | Fast execution | | Frontend | Claude Sonnet | UI/UX development | | Simple Tasks | GPT-3.5 | Cost-effective |

📊 Performance Benefits

Multi-LLM vs Single-LLM

• 4x better rate limit capacity - Distribute load across providers
• 60% faster execution - Parallel processing
• 30% cost reduction - Smart model selection
• 99.9% uptime - Fallback chains ensure reliability

🔄 Workflow Example

// BMAD-BUFF automatically:
// 1. Decomposes your project into tasks
// 2. Assigns optimal LLM to each task
// 3. Spawns specialized agents
// 4. Executes in parallel
// 5. Manages dependencies

const project = {
  name: "E-commerce Platform",
  tasks: [
    "Design database schema",      // → Gemini Pro
    "Build authentication API",     // → Claude Opus
    "Create React components",      // → Claude Sonnet
    "Research payment gateways",    // → Perplexity
    "Write unit tests",            // → Claude Opus
    "Generate documentation"       // → Perplexity
  ]
};

// All handled automatically!
await bmadBuff.execute(project);

🛠️ Advanced Features

YOLO Mode

Skip confirmations while maintaining quality:

workflow.enableYolo(); // Move fast, break nothing

Vector Database Integration

Enhanced context with persistent memory:

workflow.enableVectorDB({
  type: 'chromadb',
  collections: ['patterns', 'solutions']
});

Custom Agent Templates

Define specialized agents:

workflow.registerAgent({
  name: 'SecurityExpert',
  model: 'gpt-4',
  skills: ['penetration-testing', 'code-audit']
});

📈 Monitoring & Analytics

Track your development metrics:

• Tasks completed per hour
• LLM usage distribution
• Cost optimization savings
• Performance bottlenecks

🤝 Contributing

We welcome contributions! Please see CONTRIBUTING.md for details.

📄 License

MIT License - see LICENSE for details.

🙏 Acknowledgments

• Built on top of the BMAD-METHOD™ framework
• Inspired by ROMA and Codebuff methodologies
• Built for the Claude Code community
• Powered by OpenAI, Anthropic, Google, and Perplexity

🚨 Support

• GitHub Issues
• Documentation

Built with ❤️ by Javier Carrillo (papuman)

Revolutionizing AI-driven development, one task at a time.