finds.dev← search

// the find

blakeblackshear/frigate

★ 33,686 · TypeScript · MIT · updated Jun 2026

NVR with realtime local object detection for IP cameras

Frigate is a self-hosted NVR that runs object detection locally on your IP camera feeds using TensorFlow/OpenCV, with first-class Home Assistant integration. It's for home lab enthusiasts and privacy-conscious users who want surveillance without cloud dependencies. The 33k stars reflect genuine community adoption, not hype.

Motion-gated detection is smart engineering — it only runs the expensive object detection model on regions where motion was detected, which keeps CPU/GPU load manageable on modest hardware. Hardware accelerator support is wide: Google Coral, Haiku, Rockchip NPU, TensorRT, ROCm, OpenVINO — you're not locked to one vendor. The go2rtc integration for restreaming is genuinely useful, reducing camera connection count and enabling WebRTC for low-latency live view. The review workflow with semantic search and per-object retention policies is more sophisticated than most commercial NVR software.

Configuration is entirely YAML and deep — a first-time user faces a wall of camera stream settings, ffmpeg presets, zone definitions, and detector tuning before seeing anything work. The Frigate+ paid model subscription for better accuracy creates a soft dependency: the default COCO model misses a lot in real-world conditions, so you either pay or accept mediocre detection. Running everything in a single Docker container (nginx, go2rtc, the Python backend, s6-overlay process supervisor) makes debugging failures harder and updates an all-or-nothing proposition. Storage management is DIY — there's retention configuration but no built-in alerting when disk fills, which is a real operational footgun on 24/7 recording setups.

View on GitHub → Homepage ↗

// want more like this?

We dig through GitHub every week and send a few repos picked for what you actually care about — each with an honest take like this one.

Get finds in your inbox → Search again →