Diy Smart Home Ai Companion

Zero-Cost (¥0): Mac mini M4 + Old Phones ⭐ Zero Investment

Hardware: Mac mini M4 (existing), 2x old smartphones (existing)

• 2 cameras (old phones via IP Webcam), motion detection
• Llama3.1 8B or Qwen2.5 7B model
• 5-10 WiFi/Zigbee devices
• Use case: Development, testing, personal use
• Pros: No new hardware, Metal GPU acceleration, portable
• Cons: Not 24/7 (Mac may sleep), phone battery wear

See Mac mini Setup Guide below.

Low (¥2,800-3,500): Raspberry Pi 5 8GB

• 2-3 cameras, 5fps AI detection
• Qwen2.5 3B model
• 20-30 Zigbee devices

Mid (¥6,000-8,000): Intel N100 + 16GB ⭐ Recommended

• 4-6 cameras, 10fps detection + face recognition
• Qwen2.5 7B/14B model
• 50-100 Zigbee devices
• 7x24 stable operation

High (¥15,000-20,000): Intel NUC i5 + 32GB + GPU

• 8-12 cameras, 25fps full analysis
• Qwen2.5 32B model
• 200+ devices
• Enterprise-grade reliability

Mac mini Zero-Cost Setup

For users with existing Mac hardware and old smartphones.

Architecture

Mac mini M4 (16GB RAM)
├── Home Assistant (Docker)
├── Ollama (Llama3.1 8B, Metal GPU)
├── Whisper (Wyoming, medium-int8)
├── Piper TTS (zh_CN-huayan-medium)
└── IP Cameras (old phones)
    └── Android/iPhone → IP Webcam App

Resource Usage (~8-9GB total)

• Home Assistant: ~500MB
• Ollama 8B: ~6GB
• Whisper: ~2GB
• Piper: ~100MB

Docker Compose (Mac)