OpenClaw + Frigate + Hailo: AI-Powered Home Security on a Raspberry Pi

Why Frigate + OpenClaw

I’ve been running Frigate NVR on a Raspberry Pi 5 for months. It’s excellent at detecting people, cars, and packages with the Hailo-8L AI HAT doing object detection at 13 TOPS. But the notification system? Basic. I wanted something smarter.

Enter OpenClaw. Instead of simple push notifications, I now have an AI agent that:

• Analyzes Frigate events and decides what’s worth alerting
• Sends contextual Discord/WhatsApp messages (“Person at front door, carrying a package, 2:34 PM”)
• Responds to natural language queries (“Show me the last detection at the back gate”)
• Learns my patterns and reduces false alerts over time

The Setup

Hardware:

• Raspberry Pi 5 (8GB) — runs Frigate
• Hailo-8L AI HAT — 13 TOPS neural network accelerator
• Raspberry Pi 5 (4GB) — runs OpenClaw gateway
• 2x PoE IP cameras (Amcrest 4K)

# On the Frigate Pi — docker-compose.yml
services:
  frigate:
    image: ghcr.io/blakeblackshear/frigate:stable
    container_name: frigate
    restart: unless-stopped
    privileged: true
    volumes:
      - ./config:/config
      - /dev/bus/usb:/dev/bus/usb
    ports:
      - "5000:5000"
      - "8554:8554"  # RTSP
    environment:
      - FRIGATE_RTSP_PASSWORD=changeme

Connecting OpenClaw to Frigate

Frigate publishes events via MQTT. I wrote a simple bridge that OpenClaw monitors:

# openclaw.yaml — webhook configuration
webhooks:
  frigate:
    url: http://frigate-pi:5000/api/events
    poll_interval: 30s
    
# Or use MQTT directly
mqtt:
  broker: mqtt://frigate-pi:1883
  topics:
    - frigate/events

The Smart Alert Logic

Instead of getting 50 notifications a day for squirrels, OpenClaw filters intelligently:

# In HEARTBEAT.md or a custom skill
# Check Frigate events every 5 minutes
# Only alert on: person, car, package
# Ignore: cat, dog, bird (unless inside the house)
# Group rapid-fire events (same object, <2 min apart)

When a person is detected at the front door, OpenClaw sends me:

🚪 Front Door — Person Detected Time: 2:34 PM | Confidence: 94% Duration: 12 seconds Note: Appears to be a delivery driver (package visible)

The Hailo Difference

Without the Hailo HAT, Frigate runs object detection on the Pi’s CPU at maybe 5 FPS with high latency. With Hailo:

Detection FPS: 30+
Latency: <50ms
CPU usage: 15% (vs 95% without Hailo)
Power draw: ~8W total

The Pi barely breaks a sweat. This is genuinely impressive hardware for $70.

Voice Queries

The best part? I can ask OpenClaw about my cameras:

• “Any activity at the front door today?” → Gets a summary from Frigate’s API
• “How many cars parked in the driveway this week?” → Queries event database
• “Show me the last 5 detections” → Returns thumbnail links

Cost Breakdown

Frigate Pi 5 (8GB):          $80
Hailo-8L AI HAT:             $70
OpenClaw Pi 5 (4GB):         $60
2x Amcrest 4K PoE cameras:  $120
PoE switch:                  $40
Total:                       $370

Compare that to a Ring or Nest subscription — $100+/year, cloud-dependent, and they can see your footage. This setup is fully local, fully private, and way smarter with OpenClaw handling the brains.