# Frigate — Self-Hosted AI-Powered Network Video Recorder

> A real-time NVR that uses local AI object detection to monitor security cameras, with tight Home Assistant integration and minimal resource usage.

## Install

Save as a script file and run:

# Frigate — Self-Hosted AI-Powered Network Video Recorder

## Quick Use
```bash
docker run -d --name frigate 
  --shm-size=256m 
  -v /path/to/config:/config 
  -v /path/to/media:/media/frigate 
  -p 5000:5000 -p 8554:8554 
  ghcr.io/blakeblackshear/frigate:stable
```

## Introduction
Frigate is an open-source NVR built around real-time AI object detection. It processes camera feeds locally using Google Coral TPU or CPU-based inference to identify people, vehicles, and animals, recording only meaningful events instead of continuous footage. It integrates directly with Home Assistant for smart-home automation.

## What Frigate Does
- Processes RTSP camera streams with real-time object detection
- Records video clips and snapshots only when relevant objects appear
- Supports hardware acceleration via Google Coral TPU, OpenVINO, or GPU
- Provides a web UI with live views, event timeline, and recording playback
- Sends events to Home Assistant and MQTT for automations and notifications

## Architecture Overview
Frigate runs as a Docker container with FFmpeg handling video decoding and a TensorFlow Lite or OpenVINO model performing object detection per frame. Detected objects are tracked across frames to reduce duplicate alerts. Recordings are stored as continuous segments and trimmed to events. An embedded Go web server provides the API and UI, while MQTT publishes real-time detection events.

## Self-Hosting & Configuration
- Deploy via Docker or Docker Compose with access to camera RTSP streams
- Define cameras and detection zones in a YAML configuration file
- Attach a Google Coral USB or M.2 TPU for fast low-power inference
- Mount a local or NAS volume for recording storage
- Enable MQTT to connect with Home Assistant or other automation platforms

## Key Features
- Sub-second object detection with hardware acceleration support
- Zone-based filtering to ignore irrelevant motion areas
- Event-based recording saves storage compared to 24/7 capture
- Native Home Assistant add-on and integration
- Re-streaming via RTSP allows viewing in any compatible player

## Comparison with Similar Tools
- **Blue Iris** — commercial Windows NVR; Frigate is free and Linux-native with AI built in
- **Shinobi** — open-source NVR focused on recording; Frigate excels at AI detection
- **ZoneMinder** — long-standing open-source NVR but heavier and less AI-focused
- **Scrypted** — smart-home camera platform; Frigate offers deeper NVR and recording features
- **MotionEye** — lightweight motion-based recording without AI object classification

## FAQ
**Q: Do I need a Google Coral TPU?**
A: No. Frigate works with CPU inference, but a Coral TPU significantly reduces latency and CPU load for multiple cameras.

**Q: How many cameras can Frigate handle?**
A: With a Coral TPU, a modest server can run 10+ cameras. CPU-only setups depend on resolution and detection frequency.

**Q: Does Frigate record 24/7?**
A: It can be configured for continuous recording, event-only recording, or a combination of both per camera.

**Q: Can I use Frigate without Home Assistant?**
A: Yes. Frigate runs standalone with its own web UI and MQTT integration for any automation platform.

## Sources
- https://github.com/blakeblackshear/frigate
- https://docs.frigate.video

---
Source: https://tokrepo.com/en/workflows/89a15e12-3e67-11f1-9bc6-00163e2b0d79
Author: Script Depot