What Frigate Is: A Local NVR With AI Object Detection
Frigate is a complete and local network video recorder designed for Home Assistant, with AI object detection built in. It has 37,562 stars on GitHub and is licensed under MIT. It uses OpenCV and TensorFlow to perform realtime object detection locally for IP cameras, which means your cameras get person, car, and pet detection without a monthly subscription and without sending video to a cloud service. The detection runs on your own hardware, in your own home, and your footage stays yours. If you already have IP cameras and a Home Assistant setup, Frigate is the piece that turns dumb cameras into a smart security system.
How Detection Works: Motion First, AI Only Where It Matters
Frigate is designed to minimize resource use and maximize performance by only looking for objects when and where it is necessary. It starts with very low overhead motion detection to find the regions of a frame where something is actually moving, and it only runs object detection on those regions. Object detection with TensorFlow runs in separate processes for maximum frames per second, and the whole system leans heavily on multiprocessing with an emphasis on realtime performance over processing every single frame. The result is a system that can watch many cameras at once on modest hardware, because it spends its effort on the parts of the picture that matter and ignores the rest.
The GPU Question: Why an AI Accelerator Changes Everything
Frigate works on a CPU, but it works best with a GPU or AI accelerator, and the difference is bigger than most people expect. The project is direct about it: use of a GPU or AI accelerator is highly recommended, because AI accelerators will outperform even the best CPUs with very little overhead. A Google Coral TPU, an Intel integrated GPU, or an Nvidia GPU takes the detection workload off the CPU entirely, which means higher frame rates, more cameras per machine, and a system that stays responsive instead of falling behind. If you are planning a Frigate build, budget for the accelerator first and the CPU second. The Frigate docs list the supported object detectors, so check that page before buying hardware.
Recordings, Retention, and Review: Keep What Matters
Frigate records video with retention settings based on detected objects, so your disk does not fill up with hours of empty driveway footage. It records 24/7 if you want it to, but it can keep recordings of real events far longer than uneventful stretches. The web interface gives you a live dashboard for every camera, a streamlined review workflow for going through the day's events, and multi-camera scrubbing so you can follow something across cameras. There is a built-in mask and zone editor, so you can tell Frigate to ignore the street and only watch your yard, or to alert on the front door but not the sidewalk. Frigate also saves snapshots of detected events, and those snapshots are still images measured in pixels. If you pull one into a report or print it, a Pixels to Inches check tells you how large it will come out on paper at full quality, so a sharp snapshot does not turn into a blurry thumbnail. It is the difference between a recorder that saves everything and a system that shows you what happened.
Home Assistant, MQTT, and Re-streaming
Frigate was built for Home Assistant, and the integration is tight: a custom Home Assistant component exposes Frigate's detections as sensors and events, so a person detected at the front door can turn on a light, send a notification with a snapshot, or trigger any automation you already have. Under the hood, Frigate communicates over MQTT, which makes it easy to integrate with other systems too. It re-streams cameras over RTSP, which reduces the number of direct connections to each camera, and it supports WebRTC and MSE for low-latency live view in the browser. Your cameras stay local, your automations stay local, and nothing in the chain needs a cloud account.
Setting Up Frigate: Docker, Config, and Your First Camera
Frigate runs in Docker, and the setup is one config file plus a container. You write a frigate.yml file describing your cameras, their RTSP stream URLs, and which objects to detect, then start the container and open the web dashboard. When you upgrade Frigate and something changes in your config, a quick Text Diff between your old config and the new example shows exactly what moved, so you can update your file instead of rewriting it. Start with one camera, get detection working, then add the rest.



