Evaluating a full-product ODM project? A complete shipped case: structure, app, device APIs and cloud services — as a reference for your project.
Key capabilities and boundaries of the private cloud solution
| Item | Specification |
|---|---|
| Positioning | NAS private cloud as the storage and compute hub, IPC cameras as sensing nodes, AI engine for local monitoring and alarms |
| Data sovereignty | Video stored 100% on the customer's own NAS, never uploaded; RAID1 dual-disk redundancy |
| Subscription | 0 CNY/year — no cloud storage fee, no algorithm license, no per-channel billing |
| Camera support | ONVIF + RTSP (95%+ of cameras), compatible with Hikvision, Dahua, TP-LINK, EZVIZ, Xiaomi, Haikel, HarmonyOS |
| Software stack | Frigate NVR + Home Assistant, both open source |
| AI capabilities | Person / vehicle / pet / package detection, face recognition, zone intrusion, line crossing, loitering; advanced adds fall, smoke-fire, license plate |
| AI latency | On-device inference under 200 ms, works with the network down |
| Storage planning | 1080p / H.265 / 15fps: ~10.8 GB per channel per day, ~324 GB for 30 days |
| Event recording | Records only when something happens, saving 70–80% of storage |
| NAS specs | N100 quad-core (UGREEN DXP2800 / DXP4800) up to AMD V1500B (Synology DS1621+) |
| Hardware acceleration | Intel N100 / N5095 / AMD V1500B support Intel QSV |
| Capacity | N100 tested for 4-8 channels of 1080p AI detection (yolov8n) |
| Local LLM | Optional Ollama / LLava for natural-language video queries |
| Remote access | Frigate proxy or Tailscale zero-trust tunneling, no public IP needed |
Tell us the model or scenario and we'll send the docs and config advice
Request docs