Fregata vs Frigate

Fregata is built on the open-source Frigate NVR. The short pitch: Frigate is the Python core; Fregata is a native macOS shell around it that swaps in Apple-native acceleration.

In one sentence

Fregata is Frigate plus an Apple-Neural-Engine detector, plus VideoToolbox decode/encode, plus a signed Mac app, packaged as a drag-to-Applications product with a $10 license.

What Fregata adds

A CoreML detector that runs on the ANE

Frigate detects on the CPU, an EdgeTPU, an NVIDIA GPU (tensorrt), an Intel iGPU (openvino), or a couple of more exotic accelerators. None of those help on Apple Silicon.

Fregata adds a coreml detector — a Rust dylib that owns ONNX Runtime’s CoreML execution provider, configured to prefer the ANE. Bundled YOLOv9-tiny inference is in the 1–4 ms range per frame on the Neural Engine, versus 40–80 ms on the CPU path.

The detector type and how to swap models is documented at Detection tuning.

VideoToolbox decode and encode by default

Frigate-on-Linux uses CPU ffmpeg by default and lets you opt into hardware accel. Fregata-on-macOS flips that around: every input is -hwaccel videotoolbox, recording defaults to h264_videotoolbox, the RTSP re-stream uses h264_videotoolbox. Hardware H.264 / HEVC decode and encode runs on the dedicated media engine, leaving the CPU free.

You can override the presets exactly as on Linux Frigate; the mechanism is identical. The defaults just match the platform.

Frigate ships as a Docker image. Fregata ships as a .app bundle:

Menu-bar tray for live status, log access, the license screen, settings.
Supervisor that launches the Python core, watches it, restarts on crash with exponential backoff, captures early stderr to a survivable place when something goes wrong before the log redirect kicks in.
Welcome wizard for permission grants and folder choices.
Native Login Item registration via SMAppService.
Custom updater (signed Ed25519 manifest, no Sparkle).

None of this is Frigate; it’s the macOS-native experience layer.

Licensing

Fregata is paid software ($10/year for the update window). The licensing system is local-first — the activation token lives in the Keychain, the Mac runs offline indefinitely while the app is up (a 7-day window applies only to restarting with a stale token), and an expired paid license just stops new updates. Existing versions keep running forever.

The whole thing is described in Activation and Updates & renewals.

Bundled binaries

Fregata’s .app bundle ships its own ffmpeg, ffprobe, nginx, and go2rtc, signed with our Developer ID. You don’t install Homebrew, don’t brew install ffmpeg, don’t manage a Python virtualenv. The app self-contains.

What Fregata removes

The detectors and acceleration paths that don’t apply on macOS:

EdgeTPU — Coral USB / M.2 module. Not what Apple Silicon users would buy.
TensorRT — NVIDIA GPU acceleration. Macs don’t have NVIDIA GPUs.
OpenVINO — Intel iGPU / NPU. Not on Apple Silicon.
ROCm — AMD GPU. Not on Apple Silicon.
Hailo-8L / RKNN / Synaptics / Teflon — Linux- only NPU integrations.
DeepStack / DeGirum / ZMQ-IPC detectors — external inference servers, not the macOS-native shape.

If you need any of those, you want Frigate-on-Linux, not Fregata-on-macOS.

What stays the same

A non-trivial chunk of Frigate is identical:

The web UI — same React app, same tabs, same recording scrubber. We’ve added or trimmed nothing.
config.yml schema — every key Frigate understands works in Fregata, with the exception of the removed detector types.
MQTT topics, HTTP API, WebSocket events — the wire surface Home Assistant and other integrations target is identical.
The Frigate+ workflow — your account, your training data, your trained models. Fregata reads them via the same plus://... identifier.
Recordings format — segmented MP4s, same naming convention, same SQLite events table. You can rsync recordings between a Frigate-on-Linux install and a Fregata-on-macOS install and the web UI will show them.

Where to read which docs

When the docs cover macOS-specific ground — the menu-bar app, licensing, the CoreML detector, VideoToolbox presets, system requirements, supported Mac models, the Welcome wizard — read Fregata’s docs (you’re here).

When the docs cover the upstream Frigate engine — every non-detector config key, MQTT topic taxonomy, HTTP API, zone geometry, GenAI integration, semantic search, face recognition, LPR, PTZ tracking — read the Frigate docs. We don’t mirror them; doing so would mean duplicating maintenance forever, and the upstream docs are excellent.

This page is the only one that tries to keep both halves fresh — if Frigate adds a detector type and we add support, the “What Fregata removes” list shortens.