Edge Device AI Monitoring And Metrics Engine
EDAMAME is a lightweight, extensible edge AI observability and monitoring system designed for Raspberry Pi–class devices and AI accelerators (Hailo, Jetson-class, CPU-only).
It provides real-time health, performance, and AI pipeline metrics from edge devices and streams them to a centralized dashboard for monitoring, alerts, and analysis.
Edge AI deployments often fail silently:
- FPS drops due to thermal throttling
- Pipelines hang after days of uptime
- Models change but nobody knows which version broke things
- Devices overheat, reboot, or lose camera feeds
EDAMAME makes edge AI observable.
- CPU usage
- RAM usage
- Disk usage
- CPU temperature
- Throttling & under-voltage flags
- Uptime & reboot detection
- Inference FPS
- Inference latency (avg / p95)
- Dropped frames
- Pipeline crash detection
- Last successful inference timestamp
- Model version / hash tracking
- Hailo device temperature & utilization (where available)
- Accelerator presence & health checks
- Camera disconnect detection
- Frame freeze detection
- Resolution & format change detection
- HTTPS-based metric ingestion
- JSON-based schema
- Compatible with AWS serverless backends
- Alert-ready (email / webhook)
EDAMAME has been tested on real hardware, not just emulators:
| Device | OS | Accelerator |
|---|---|---|
| Raspberry Pi 5 | Bookworm | Hailo-8 / Hailo-8L |
| Raspberry Pi Zero 2 W | Bullseye, Bookworm | CPU-only |
| Radxa Rock 5T | Ubuntu | CPU / NPU / LiDAR (SSL) |
┌────────────────────┐
│ Edge Device │
│ (RPi / Rock) │
│ │
│ EDAMAME Agent │
│ - System metrics │
│ - AI metrics │
│ - Health checks │
└─────────┬──────────┘
│ HTTPS (JSON)
▼
┌────────────────────┐
│ Cloud Backend │
│ - API Gateway │
│ - Lambda │
│ - Dynamo DB / S3 │
└─────────┬──────────┘
▼
┌────────────────────┐
│ Web Dashboard │
│ - Live metrics │
│ - Alerts │
│ - History │
└────────────────────┘