Frigate NVR

AI-Powered Network Video Recorder

Service Overview

Property	Value
Service Name	frigate
Category	Security / Surveillance
Docker Image	`ghcr.io/blakeblackshear/frigate:stable`
Web UI Port	5000
RTSP Restream Port	8554
WebRTC Port	8555
Status	Tested on Seattle (2026-03-27), removed after validation

Purpose

Frigate is a self-hosted NVR with real-time AI object detection. Instead of 24/7 recording, it detects people, cars, animals, etc. from RTSP camera streams and only records clips when objects are detected. Integrates with Home Assistant.

Tested Configuration

Successfully tested on Seattle (16 vCPU, 62GB RAM) with a Tapo camera on the Concord NUC subnet.

Camera

Model: Tapo camera with RTSP
IP: 192.168.68.67 (GL-MT3000 subnet, 192.168.68.0/22)
RTSP streams: rtsp://USER:PASS@192.168.68.67:554/stream1 (high), stream2 (low) # pragma: allowlist secret
RTSP credentials: Set via Tapo app -> Camera Settings -> Advanced -> Camera Account

Network Path

The camera is on the Concord NUC's LAN (192.168.68.0/22). For other Tailscale nodes to reach it:

NUC advertises 192.168.68.0/22 via Tailscale (already configured + approved in Headscale)
The Frigate host must have --accept-routes=true in Tailscale (tailscale set --accept-routes=true)

Compose File (reference)

services:
  frigate:
    image: ghcr.io/blakeblackshear/frigate:stable
    container_name: frigate
    restart: unless-stopped
    shm_size: 256mb
    security_opt:
      - no-new-privileges:true
    environment:
      TZ: America/Los_Angeles
    ports:
      - "5000:5000"
      - "8554:8554"
      - "8555:8555/tcp"
      - "8555:8555/udp"
    volumes:
      - ./config:/config
      - ./storage:/media/frigate
      - type: tmpfs
        target: /tmp/cache
        tmpfs:
          size: 1000000000

Config File (reference)

mqtt:
  enabled: false

detectors:
  cpu:
    type: cpu
    num_threads: 4

objects:
  track:
    - person
    - car
    - cat
    - dog
  filters:
    person:
      min_score: 0.5
      threshold: 0.7

record:
  enabled: true
  retain:
    days: 7
    mode: motion
  alerts:
    retain:
      days: 14
  detections:
    retain:
      days: 14

snapshots:
  enabled: true
  retain:
    default: 14

detect:
  enabled: true
  width: 1280
  height: 720
  fps: 5

go2rtc:
  streams:
    tapo_cam:
      - rtsp://USER:PASS@192.168.68.67:554/stream1  # pragma: allowlist secret
    tapo_cam_sub:
      - rtsp://USER:PASS@192.168.68.67:554/stream2  # pragma: allowlist secret

cameras:
  tapo_cam:
    enabled: true
    ffmpeg:
      inputs:
        - path: rtsp://127.0.0.1:8554/tapo_cam
          input_args: preset-rtsp-restream
          roles:
            - record
        - path: rtsp://127.0.0.1:8554/tapo_cam_sub
          input_args: preset-rtsp-restream
          roles:
            - detect
    detect:
      width: 640
      height: 480
      fps: 5
    objects:
      track:
        - person
        - car
        - cat
        - dog

version: 0.14

Deployment Notes

CPU detection works for 1-2 cameras but is not recommended for production. Consider a Google Coral USB TPU for hardware acceleration.
go2rtc handles RTSP restreaming — camera credentials only need to be in go2rtc streams, not in ffmpeg inputs.
Use stream2 (sub-stream, lower resolution) for detection to save CPU.
Use stream1 (main stream, full resolution) for recording.
Default credentials on first start: admin / auto-generated password (check docker logs frigate).
Config validation errors: ui -> live_mode is not valid in v0.14+. Don't add extra fields not in the docs.

Future Deployment

Best host options for permanent deployment:

Concord NUC: Same LAN as camera, no Tailscale routing needed. Has Home Assistant running.
Homelab VM: Central infrastructure host, plenty of resources.
Atlantis: Has the most storage for recordings.

All require tailscale set --accept-routes=true unless on the same LAN as the camera.

4.1 KiB Raw Blame History