blakeblackshear/frigate

Frigate is a self-hosted network video recorder, or NVR, which means it stores video footage from IP cameras on your own hardware rather than uploading it to a cloud service. What sets it apart is that it performs real-time AI-powered object detection locally, so it can distinguish between a person, a car, a pet, or other objects rather than just triggering alerts on any motion. This means far fewer false alarms compared to basic motion-sensing systems. The system works by analyzing the video streams from your cameras using OpenCV for motion detection and TensorFlow for object recognition. It is designed to be efficient: it first uses a lightweight motion detector to identify which parts of the frame are active, then focuses the more expensive AI analysis only on those regions. Object detection runs in separate processes to maximize performance. For the fastest results, Frigate supports dedicated AI accelerator hardware such as Google Coral, though it also runs on a regular CPU or GPU. Frigate integrates tightly with Home Assistant, the popular home automation platform, and communicates over MQTT, a lightweight messaging protocol common in home automation. You can configure it to record video only when specific objects are detected, set retention policies, define zones in the camera frame, and view a live dashboard with low latency through WebRTC. It also re-streams camera feeds via RTSP to reduce the load on your camera hardware. You would use Frigate when you want a privacy-respecting, locally-run security camera system with smart detection, ideally as part of a Home Assistant smart home setup. The frontend is built with TypeScript, and the backend relies on Python and OpenCV with TensorFlow for AI inference. It is licensed under MIT.