Frigate NVR on Home Assistant: beginner's setup guide (2026)

Frigate is what happens when a Home Assistant, the open-source smart home automation platform, power-user gets tired of paying Ring or Nest a fiver a month to do basic object detection on their own doorbell footage. It is an open-source NVR (Network Video Recorder) that runs locally on your machine, talks to whichever cheap IP cameras you already own, applies real AI object detection to every frame, and surfaces the results back into Home Assistant as actionable events.

The pitch is genuinely good. The setup is not always genuinely simple. This guide walks you through what you need, what you can skip, and the order to install everything in if you have never touched a Coral USB accelerator before.

What does Frigate actually do?

At its core Frigate watches a video stream and tells you, in close to real time, what it sees: a person walked across the driveway at 14:23, a car arrived at 14:25, a delivery van was at the door for 47 seconds at 14:26. That sounds simple. The hard part is doing it well enough on the hundreds of frames per second a typical camera produces, without grinding your CPU into the carpet.

Frigate solves that with two clever choices. First, it uses motion detection to decide which frames are even worth running expensive object detection on (a still empty driveway is just throwing computation away). Second, it offloads the actual AI inference to a Coral USB or M.2 accelerator wherever possible - a tiny ~£60 USB stick that runs the TensorFlow Lite models Frigate ships with at roughly 100x the speed of doing it on the CPU.

The detection events then flow back into Home Assistant as MQTT messages and proper HA entities. Suddenly you have a 'person detected at front door' trigger you can use anywhere - to light up the porch, to send a notification, to start recording, to play 'Eye of the Tiger' through the smart speakers.

What hardware do you actually need?

The honest answer: less than you think for one camera, more than you think for six.

The host machine

Frigate runs as an add-on inside Home Assistant. Anything that can run Home Assistant OS comfortably can run Frigate for one or two cameras:

• Raspberry Pi 5 (8 GB): fine for one camera at 1080p, OK for two, struggles at three or more. Roughly £75 + accessories. Use a USB SSD, not the SD card.
• An old Intel mini PC or NUC (8th gen+): the sweet spot. £100-£200 used. Plenty of headroom for four to six cameras, especially with the integrated GPU.
• A bigger server: anything beats the Pi for more than two cameras. Even an N100 mini PC is dramatically better.

The AI accelerator

This is the optional-but-very-recommended bit. Without it, Frigate can still run, but it will hammer your CPU constantly and you will hit walls fast.

• Coral USB Accelerator (~£60): the easy answer. Plug it in, point Frigate at it, done. Runs detection on hundreds of frames per second at very low power.
• Coral M.2 (~£40): same chip in an M.2 form factor. Use if your host has a spare M.2 slot - usually slightly more reliable than USB.
• Intel iGPU (free): recent Intel integrated graphics (UHD 730 / Iris Xe) can do detection via OpenVINO almost as well as a Coral. Worth checking before buying a Coral.

The cameras

The single most important spec: RTSP support. Pretty much every serious IP camera supports it; most pure consumer cameras (Arlo, Nest, Ring) do not, and they will not work with Frigate without painful workarounds.

Good UK-friendly options:

• Reolink (£40-£150): the consumer favourite. RLC-820A and RLC-823A are particularly popular Frigate cameras. PoE versions save the power-supply faff.
• Amcrest (£60-£200): rebadged Dahua under the hood, well-documented Frigate setup, sub-stream support out of the box.
• Unifi Protect (£100-£300): works via RTSPS once you turn it on in the Protect controller. Good if you already live in the Unifi world.
• Tapo C320WS / C520WS (£40-£60): the budget pick, but ONLY the C320/C520 family. Most other Tapo / TP-Link consumer cameras have RTSP blocked.

How do you install Frigate?

There are three install routes; the right one depends on how you run Home Assistant.

Option A: Home Assistant OS / Supervised (the easy path)

If you run HA OS or Supervised, the official Frigate add-on installs it for you. Open Home Assistant → Settings → Add-ons → Add-on Store. Add the official Frigate repository (the URL is in the project README) and install. The add-on takes care of the Docker container, mounts the config and storage paths, and gives you a 'Start' button.

This is the route 95% of beginners should take. It is also the route that the rest of this guide assumes.

Option B: Home Assistant Container / Docker (intermediate)

If you run HA in Docker yourself, you also run Frigate in Docker yourself. The Frigate docs ship a working docker-compose.yml. Network it correctly so HA can reach Frigate via MQTT, point it at your cameras, mount a storage volume large enough for your retention window. This is more work but more flexible.

Option C: Home Assistant Core (the hard path)

If you run HA Core (a Python venv, no Supervisor), you still install Frigate in Docker - you just have to wire the MQTT broker yourself. Best skipped if this is your first Frigate setup.

What does the YAML config actually look like?

Frigate is configured via a single YAML file. Here is a minimal working config for one Reolink camera plus a Coral USB accelerator. Copy this, edit the obvious bits, save as your Frigate config.yml in the Home Assistant config directory:

(in the live published post, an actual YAML code block goes here covering: mqtt host, detectors with coral USB and CPU fallback, cameras section with main + sub stream ffmpeg inputs, motion detection masks, snapshots and clips retention, objects to track.)

The shape you want to know about:

• mqtt: tell Frigate where your Home Assistant MQTT broker is. With the official add-on this is the built-in Mosquitto add-on; HA usually pre-fills the address.
• detectors: tell Frigate what AI hardware to use. Coral USB is type 'edgetpu' with device 'usb'. Intel iGPU is type 'openvino'.
• cameras: the most important section. Each camera needs an 'ffmpeg inputs' block - one stream for detection (the sub-stream), one for recording (the main stream).
• record: how long to keep clips. The default 'retain everything for 10 days' fills disk fast at 4K; tune to your storage.
• objects: which categories Frigate flags. By default 'person' is enough; you can add 'car', 'dog', 'cat', 'package' when you need them.

The Frigate docs ship a very good example config. Use it as your starting point - copying it verbatim and editing your IP addresses and credentials in is the fastest path to a working setup.

How do you wire it back into Home Assistant?

Once Frigate is running and detecting things, you want those detections to be real Home Assistant entities. Two integrations make this happen:

1. The Frigate HACS integration: install via HACS, point it at your Frigate URL, restart HA. You now have binary_sensor entities for each tracked object on each camera (binary_sensor.frigate_front_door_person, binary_sensor.frigate_driveway_car, etc).
2. The MQTT subscription: Frigate also publishes raw events to MQTT topics, which lets you build slightly cleverer automations than the binary_sensor flow allows. Optional for beginners.

From there, normal Home Assistant automations apply. 'When binary_sensor.frigate_front_door_person turns on, send a notification with the clip URL.' 'When a car arrives in the driveway between 17:00 and 19:00, switch on the porch lights.' The detection layer is now just another sensor for HA to reason about.

What about notifications with snapshots?

The killer feature most people install Frigate for: a phone notification that says 'a person was at the front door' AND includes the actual photo of who it was. Setting this up takes about ten minutes once Frigate is running.

The Home Assistant mobile app supports actionable notifications with attached images. The image you want to attach is Frigate's snapshot URL, which the integration exposes as an attribute on each binary_sensor. Build a notification automation that fires when the binary_sensor turns on, attaches the snapshot URL as the notification image, and sends it via mobile_app.your_phone. Done.

The notification arrives within a second or two of the detection event. It is the moment most people decide Frigate has paid for itself.

What are the things that will actually go wrong?

Frigate's reputation for being fiddly comes from the fact that the same five or six issues catch most new users.

• Wrong network mode in the add-on. If Frigate cannot find your MQTT broker, it usually means the add-on is on a Docker bridge network and your MQTT broker is on the host network. Fix by setting the add-on's network mode to 'host' in its Configuration tab.
• Sub-stream not actually a sub-stream. If your CPU is pegged at 100%, you probably have Frigate doing detection on the 4K main stream. Open the camera's web UI, find the sub-stream URL, paste it into the Frigate config's detect input.
• RTSP credentials in the URL. Most cameras need username:password embedded in the RTSP URL (rtsp://user:pass@ip/stream). The Frigate logs will tell you if auth is failing.
• Coral not detected. Unplug, replug, restart the add-on. If it still does not show up, the udev rules sometimes need a host reboot.
• Cameras with shared substream paths. Some cheap cameras serve the same RTSP path for main and sub stream and just rely on the bitrate setting. These cause endless tuning headaches. Replace the camera.
• Storage filling up. Recording 4K from four cameras for 14 days fills 1 TB in roughly that long. Set sensible retention windows in the record: config from the start.

What about Nvr.gateway and the other AI options?

Frigate is not the only local-AI NVR for Home Assistant in 2026. The honest alternative landscape:

• Synology Surveillance Station: excellent if you already own a Synology NAS, but locked to the NAS hardware and licence-limited per camera. Less HA-native than Frigate.
• Unifi Protect: a brilliant product if you live in the Unifi world. Native Unifi-camera-only, but the latest controllers ship reasonable local AI detection without a separate accelerator. HA integration is decent.
• Scrypted: the technical competitor. More plugin-driven than Frigate. Some users swear by it for HomeKit Secure Video integration.

For most Home Assistant users in 2026, Frigate is the right answer: most cameras supported, biggest community, best HA integration, free.