Getting Started

Get momo-kibidango running in your OpenClaw environment in under 5 minutes.

Prerequisites

• ✓ macOS with Apple Silicon (M1/M2/M3/M4)
• ✓ OpenClaw v1.12.0 or higher
• ✓ 16GB RAM minimum (32GB recommended)
• ✓ 50GB free disk space

Quick Install

pip install momo-kibidango

OpenClaw Integration

Add momo-kibidango to your OpenClaw configuration:

{
  "models": {
    "accelerated": {
      "provider": "momo-kibidango",
      "base_model": "sonnet-3.5",
      "draft_models": ["haiku-2", "haiku-3"],
      "device": "mps"
    }
  }
}

Verify Installation

Test that momo-kibidango is working correctly:

momo-kibidango test

Expected output:

✅ Models loaded successfully
✅ Draft model: haiku-2 (45.6 tok/s)
✅ Middle model: haiku-3 (30.5 tok/s)  
✅ Target model: sonnet-3.5 (12.5 tok/s)
✅ Acceleration: 1.97x (24.6 tok/s effective)
✅ Memory usage: 11.6GB

First Run

Use momo-kibidango with your OpenClaw subagents:

from openclaw import spawn_subagent

# Automatically uses accelerated inference
result = spawn_subagent(
    task="Write a comprehensive analysis",
    model="accelerated"  # Uses momo-kibidango
)

Monitor Performance

View real-time metrics with the built-in dashboard:

momo-kibidango monitor

This opens a web dashboard at http://localhost:9090 showing:

• • Token generation speed
• • Acceptance rates per model
• • Memory usage
• • Cache hit rates
• • Fallback statistics

Next Steps