# Frigate

## 🎥 Overview
Frigate is an open-source network video recorder (NVR) with real-time AI-based object detection.  
It is designed to work with IP cameras and integrates tightly with Home Assistant to provide smart, event-driven automations.

## 🚀 What it does
- Records video streams from IP cameras
- Performs real-time object detection (people, animals, vehicles, etc.)
- Generates events and snapshots based on detections
- Provides a web UI for live view, recordings, and review
- Integrates with MQTT and Home Assistant

## 🌐 Website
- https://frigate.video/

## ⚙️ Features
- Real-time object detection using TensorFlow Lite
- Motion detection + object classification
- RTSP camera support
- Event-based recording (not just continuous)
- Timeline and review UI
- MQTT integration for automation
- Multi-camera support

## 🧠 AI Acceleration
Frigate works best with hardware acceleration:

- **Google Coral TPU (recommended)** → best performance, low power
- CPU-only → works but limited performance
- GPU acceleration (optional, more complex)

## 📦 Kubernetes Deployment Notes
- Requires:
  - Persistent storage (recordings + config)
  - Access to camera streams (RTSP)
- Recommended setup:
  - PVC for `/config` and `/media`
  - Service + Ingress for web UI
- Optional:
  - Device passthrough for Coral TPU:
    - `/dev/bus/usb` (USB Coral)
- Considerations:
  - High disk I/O (video recording)
  - CPU usage if no accelerator is present

## 📁 Storage Structure
- `/config` → Frigate configuration (YAML)
- `/media` → recordings, clips, snapshots

## 🔌 Integrations
- **Home Assistant** (native integration)
- MQTT broker (required for event communication)
- RTSP cameras (e.g. Reolink)

## 🎥 Camera Integration
Typical flow:

Camera (RTSP)
    ↓
Frigate (detects objects)
    ↓
MQTT (events)
    ↓
Home Assistant (automations, notifications)

## 🔐 Authentication
- Built-in authentication for UI
- Can be secured behind:
  - Reverse proxy (Ingress / NGINX)
  - SSO (e.g. Authentik via OAuth2 proxy)

## ⚠️ Things to know
- Object detection requires tuning for best results
- CPU-only setups may struggle with multiple cameras
- Storage grows quickly → plan retention policies
- Requires stable RTSP streams from cameras

## 🧠 Why use it
- Local, privacy-friendly camera processing
- Powerful AI detection without cloud dependency
- Deep integration with Home Assistant
- Efficient event-based recording

## 🔧 Typical Stack in Undercloud
- Frigate → NVR + AI detection
- Home Assistant → automation + UI
- MQTT (Mosquitto) → event bus
- Reolink cameras → video source
- Coral TPU (optional) → AI acceleration

## 📝 Example Config Snippet
```yaml
mqtt:
  host: mqtt.apps.undercloud.dev

cameras:
  living_room:
    ffmpeg:
      inputs:
        - path: rtsp://user:password@camera-ip/stream
          roles:
            - detect
            - record
    detect:
      enabled: true
    objects:
      track:
        - person
        - dog
        - cat