Can Frigate use the Apple Neural Engine when running in Docker on a Mac?

No. Docker Desktop on macOS runs a Linux virtual machine, and that VM has no access to any of Apple Silicon's dedicated hardware — not the Neural Engine, not the GPU, not the Media Engine (VideoToolbox). Every operation falls through to the CPU. Object detection alone runs at roughly 40–80 ms per frame on the CPU path, limiting you to one or two cameras in real time. Fregata runs natively: detection on the ANE at 1–4 ms per frame, frame processing on the GPU, decode and encode on the Media Engine — the CPU is nearly free.

What changes when I switch from Frigate-in-Docker to Fregata?

Very little from a configuration standpoint. Fregata uses the same config.yml schema — the main change is the detector: block (swap cpu or coral for coreml) and hwaccel_args for VideoToolbox decode. Your recordings, event clips, and all integrations (Home Assistant, MQTT, Frigate+) carry over unchanged. See the migration guide for the full walkthrough.

Does Fregata work on Intel Macs?

No. Intel Macs don't have a Neural Engine, which is what Fregata's object detection runs on. Fregata requires Apple Silicon (M1 or newer). If you're on an Intel Mac, the upstream Frigate Docker image is your best option.

Does Fregata support the same Home Assistant integration as Frigate?

Yes. Fregata exposes the same HTTP API and MQTT topics as upstream Frigate, so the official Frigate HACS integration works unchanged — point it at your Mac's IP and port 8971. See the Home Assistant guide for setup details.

Is there a native Frigate app for Mac?

Yes — Fregata is a native macOS app built on the open-source Frigate NVR. It runs as a signed, notarized .app (drag to Applications), with no Docker, no Linux VM, and no terminal setup. It keeps Frigate's config, web UI, MQTT topics, and Home Assistant integration, and adds object detection on the Apple Neural Engine plus VideoToolbox hardware video decode/encode.

Does Frigate run on Apple Silicon (M1, M2, M3, M4)?

Frigate's Docker image runs on Apple Silicon only inside a Linux VM, which can't reach the Apple Neural Engine or VideoToolbox, so detection falls back to the CPU (~40–80 ms/frame). Fregata is the native Apple Silicon build: it runs object detection on the Neural Engine (~1–4 ms/frame for YOLOv9-tiny at 320 px) and uses VideoToolbox for H.264/HEVC decode and encode. Fregata requires Apple Silicon (M1 or newer); Intel Macs are not supported.

How do I run Frigate on a Mac without Docker?

Use Fregata, the native macOS port of Frigate. Instead of Docker Desktop and a Linux VM, you install a single signed .app and the bundled ffmpeg, nginx, and go2rtc come with it. The Frigate engine, config.yml schema, and integrations are unchanged; the detector is swapped for a CoreML detector that runs on the Apple Neural Engine. See https://fregata.app/frigate-on-mac for the full walkthrough.

What's the best NVR for an Apple Silicon Mac mini?

Fregata — a native macOS NVR built on Frigate. A passively-priced Apple Silicon Mac mini runs real-time object detection across many cameras on the Neural Engine, with VideoToolbox handling video decode/encode, no separate GPU, Coral TPU, or low-resolution detection sub-stream required, and no per-camera license fee ($10 one-time, 30-day free trial).

How to Run Frigate on a Mac (Apple Silicon)

Short version: don't run it in Docker. On an Apple Silicon Mac, Fregata is the native way to run the Frigate NVR — same engine, but with every operation fully hardware-accelerated across the Apple Neural Engine, GPU, and Media Engine. Docker on Mac gives you none of them.

Why not just run Frigate in Docker on the Mac?

You can, and it starts up fine — but Docker Desktop on macOS runs everything inside a Linux virtual machine, and that VM has no access to Apple Silicon's dedicated hardware. Every operation falls through to the CPU:

The Apple Neural Engine. Frigate's object detector can't reach it from inside the VM, so detection falls back to the CPU at roughly 40–80 ms per frame — too slow to watch more than a camera or two in real time.
The GPU. Frame processing — scaling, format conversion, filter pipelines — runs on the CPU instead of being offloaded to the GPU, where it belongs.
The Media Engine (VideoToolbox). Hardware H.264/HEVC decode and encode aren't available inside the VM either, so ffmpeg decodes every stream on the CPU and your cores stay pegged.

The result is a setup that technically works but runs hot, loud, and entirely CPU-bound — three dedicated hardware units sitting completely idle while the CPU does everything alone.

The native way: Fregata

Fregata is a native macOS port of Frigate. It's the same Frigate engine — same config.yml, same web UI, same MQTT topics, same Home Assistant integration, same Frigate+ support — repackaged as a signed, notarized .app you drag to Applications. No Docker, no Linux VM, no terminal.

What changes is the part that was Linux-shaped to begin with. Every stage of the pipeline is offloaded to dedicated hardware — nothing runs on the CPU that doesn't have to:

Object detection runs on the Apple Neural Engine via a CoreML detector — about 1–4 ms per frame for the bundled YOLOv9-tiny model at 320 px, versus 40–80 ms on the CPU path.
The GPU handles frame processing — scaling, format conversion, and filter operations run via Metal shaders, keeping decoded frames in GPU-accessible memory the whole time.
The Media Engine (VideoToolbox) handles decode and encode on dedicated fixed-function silicon, so H.264 and HEVC streams don't touch the CPU.
Adaptive Transcoding uses the Media Engine, not the CPU. VideoToolbox APIs automatically transcode recorded clips and live streams to lower qualities — so you can always see what's happening on your cameras even on a weak mobile connection without enough bandwidth for the full-quality stream, all with zero CPU usage. (Exclusive Fregata Feature)
No low-resolution detection sub-stream required — the ANE is fast enough to detect on the full-resolution stream, so you keep accuracy on small and distant objects.
Frigate AI Enrichments have full GPU access — face recognition, license plate reading, and other enrichment workloads can use the Mac's GPU directly, instead of being bottlenecked to the CPU inside a VM.

In practice a base Apple Silicon Mac mini handles real-time detection across 8 or more 4K cameras while the CPU sits nearly idle — because it genuinely isn't doing the work.

How to get started

Download Fregata-latest.dmg and drag Fregata.app to Applications.
Launch it and complete the welcome wizard — grant local-network access, pick where recordings live, and start the 30-day free trial (no payment required).
Add a camera by pasting its RTSP URL in the web UI. Detection runs automatically; the menu-bar tray shows the per-frame inference time so you can confirm it's on the Neural Engine.

Full walkthrough in the docs: Installation, Add your first camera, and — if you're moving an existing Docker setup — Migrating from Frigate to Fregata (your config and recordings carry over).

Requirements

Apple Silicon Mac (M1 or newer). Intel Macs aren't supported — they have no Neural Engine.
macOS 13 (Ventura) or later.

Pricing

$10 one-time for a year of updates (the version you have keeps working forever), with a 30-day free trial and no per-camera fees. See the pricing and how Fregata compares to Frigate-in-Docker, Scrypted, Blue Iris, and Synology.

Prefer to stay on Docker?

That's completely fine — Frigate is free, open source, and excellent. If you're on Linux or have a Coral / NVIDIA box, run upstream Frigate. Fregata exists specifically for people who want that same engine to run natively, and fast, on a Mac.

How to run Frigate on a Mac