Docker

Docker Hub rate-limits anonymous pulls to 100 per 6 hours (200 for authenticated users). In a homelab with frequent rebuilds, multiple hosts, or CI-style workflows, you’ll hit that limit fast. Even if you don’t, pulling the same nginx:alpine image ten times across ten containers wastes bandwidth and time. A local Docker registry mirror fixes both: it acts as a pull-through cache. First pull is from Docker Hub (or any upstream). Every subsequent pull from any host in your lab hits the local cache instead. ...

Frigate is an open-source NVR built for real-time object detection with local processing. No cloud subscriptions, no vendor lock-in — just cameras, a GPU, and decent detection models. This is how I set it up on my Proxmox host, with a Tapo C100 camera, Intel GPU inference via OpenVINO, MQTT eventing, and Telegram notifications. Architecture ┌──────────────┐ RTSP (stream1/stream2) ┌──────────────┐ │ Tapo C100 │ ───────────────────────────▶ │ Frigate │ │ 10.0.50.101 │ │ 10.0.20.15 │ └──────────────┘ │ │ CCTV VLAN (50) │ Web UI:5000 │ │ RTSP :8554 │ ┌──────────┐ │ WebRTC:8555 │ │ Telegram │ ◀──── MQTT events ──────── │ go2rtc:1984 │ │ alerts │ (frigate-notifier) │ API :8971 │ └──────────┘ └──────┬───────┘ │ ┌──────▼───────┐ │ Mosquitto │ │ MQTT:1883 │ └──────────────┘ The camera lives on VLAN 50 (CCTV). Frigate runs on a Debian LXC on VLAN 20 (LAB). The router’s firewall allows Frigate’s host (10.0.20.15) to reach the camera subnet — nothing else touches CCTV. ...