
Frigate
Self-hosted NVR that uses real-time object detection (TensorRT/Coral/CPU) with camera restreaming, recording, snapshots, a web UI, and strong Home Assistant integration.

Frigate is a self-hosted network video recorder (NVR) focused on real-time AI object detection for IP cameras. It combines efficient video processing with event-based recording and a web UI so you can review live views, detections, and recordings with minimal manual triage.
Key Features
- Real-time object detection pipeline designed for low-latency and low CPU usage
- Hardware acceleration support, including Google Coral (Edge TPU) and NVIDIA TensorRT (where available)
- Uses FFmpeg for camera ingest and decoding with configurable roles (detect/record/audio)
- Restreaming and multi-camera management with configurable streams and recording retention
- Motion-based and object-based event handling with snapshots and clip recording
- Web UI for live views, event timeline, clips, snapshots, and per-camera configuration visibility
- MQTT integration for publishing events and states (commonly used with Home Assistant)
- Config-driven setup (YAML) with per-camera zones, masks, and object filters
Use Cases
- Smart home security NVR with person/vehicle detection and automated alerts via MQTT/Home Assistant
- Reduce false alarms by filtering motion events through object detection and zones
- Local recording and review for multiple RTSP cameras with efficient retention policies
Limitations and Considerations
- Initial setup requires careful stream selection and FFmpeg tuning per camera model/codec
- Best performance typically depends on supported accelerators (e.g., Coral/TensorRT) and compatible hardware
Frigate is a strong choice for users who want an NVR that prioritizes fast, accurate detections and automation-friendly event output. Its architecture is optimized for continuous camera workloads while keeping review and alerting centered on meaningful objects rather than raw motion.




