How to build a Frigate server on a budget

Frigate is a local network video recorder with real-time object detection, and it pairs tightly with Home Assistant. Built on a budget it’s one of the best-value upgrades a local-first home can make — no subscription, no cloud, footage stays on your hardware. The trick to keeping it cheap without making it miserable is matching the hardware to how detection actually works, and feeding it the kind of camera streams it actually wants.

What Frigate Actually Needs

Two jobs dominate Frigate’s workload: decoding video streams and running object detection on the frames. Decoding is best handled by hardware video acceleration — Intel QuickSync on Intel iGPUs is the common, cheap option. Detection is best offloaded to an accelerator, and the guidance here has shifted. Frigate’s documentation now states that the Google Coral is no longer recommended for new installations except where power draw is the overriding constraint; a recent Intel iGPU running OpenVINO, a discrete GPU, or a Hailo-8/8L module is the current starting point. A Coral still works and is still supported, so an existing one is not wasted. You can run detection on the CPU alone, but it limits how many cameras and how high a resolution you can sustain.

Frigate runs as software on a Linux host, normally in Docker. That rules out appliances that aren’t general-purpose Linux machines — an Android TV box or a streaming stick can’t run it. You need a real x86 mini PC, an SBC, or a server.

Picking the Hardware

A used or new Intel mini PC with QuickSync is the sweet spot, because the same iGPU that decodes the streams can also run detection through OpenVINO. The Intel NUC 12 Pro is a strong choice: it ships as a barebones kit built around the 12-core Core i5-1240P (4 performance + 8 efficiency cores) with Intel Iris Xe graphics for QuickSync decoding, and you add your own DDR4 RAM and an NVMe SSD. Budget for at least 16GB of RAM and an SSD when you price it out — the kit doesn’t include them. With the iGPU running OpenVINO detection, a machine like this comfortably runs many cameras.

If the NUC is over budget, an Intel N100 or N150 mini PC is excellent value: it has QuickSync, sips power, and its iGPU handles OpenVINO detection for a handful of cameras. If you want more detection headroom later, a free M.2 slot can take a Hailo-8/8L module. If you’d rather reuse a machine you already own, see how to run Frigate on an old PC. A Raspberry Pi 5 with an M.2 AI accelerator (the Raspberry Pi AI HAT+ pairs the board with a Hailo-8L at 13 TOPS or a Hailo-8 at 26 TOPS, from $70) can work for a couple of low-resolution feeds, but it lacks the decode horsepower of an Intel iGPU and will struggle as you scale up.

For storage, recordings are the space hog. A 2TB NVMe or SATA SSD is a sane starting point; size it to your retention. Roughly, a single 1080p H.264 stream recording continuously runs on the order of tens of gigabytes per day, so multiply by camera count and retention days and add headroom. A USB 3.0 external drive can hold recordings on a tight budget, but keep Frigate’s database and cache on the internal SSD for responsiveness.

Feeding Frigate the Right Streams

Frigate ingests RTSP streams (and works with ONVIF-capable cameras for discovery). It does not take a phone plugged in over USB or a random USB webcam as a “camera” the way an NVR appliance might — give it network cameras that expose an RTSP URL. Cheap, well-supported options include Amcrest and Reolink models; the Amcrest IP8M-2496E is a solid higher-resolution pick when budget allows.

The standard pattern is to point Frigate’s detection at a camera’s low-resolution substream (to keep CPU/decode load down) and record from the high-resolution main stream. Most IP cameras expose both. Configure cameras, streams, and detection in Frigate’s config.yaml (the project moved configuration to a YAML config and an in-app editor); a Docker compose.yaml defines the container, volumes, and device passthrough for the iGPU or whichever detector you use.

Integrating with Home Assistant

Install the official Frigate integration in Home Assistant (it’s available through HACS) to pull in camera entities, detection sensors, and event clips. From there you automate normally — for example, an Amcrest person-detection event turning on a Philips Hue White A19 at night via a Philips Hue Smart Plug or directly through the Hue integration. Running Home Assistant on something like a Home Assistant Green is fine; Frigate itself should live on the Intel box with QuickSync, not on the lightweight HA appliance.

A Note on HomeKit and Hubitat

Frigate has no native HomeKit or Hubitat integration. If you want Frigate cameras to appear in Apple Home, the usual route is Home Assistant’s HomeKit bridge re-exposing the camera entities; HomeKit Secure Video is a separate Apple feature and not what Frigate provides. Hubitat has no real Frigate support — most people who run both let Home Assistant own the camera/NVR side and use Hubitat only for whatever automations already live there.

Quick Verdict

The budget Frigate build that doesn’t disappoint is an Intel mini PC with QuickSync, using the iGPU for both decoding and OpenVINO detection — an Intel NUC 12 Pro kit (add RAM and an SSD) or an N100/N150 box if money is tighter — feeding off RTSP IP cameras with a low-res detect stream and high-res record stream, recordings on a 2TB SSD sized to your retention. Skip the Android boxes and USB-webcam ideas; give Frigate a real Linux host and proper network cameras and it will run reliably for years with no subscription.