Fregata 0.17.1 — 2026-05-30
First public release. Fregata is a macOS-native port of Frigate NVR for Apple Silicon, distributed as a signed, notarized, self-contained .app. Built on Frigate 0.17.1.
Added — Apple Silicon acceleration
- CoreML object detector running on the Apple Neural Engine (~1–2 ms/frame vs. ~40–80 ms on a CPU path). ANE by default; GPU opt-in via
inference_backend. - NEON-accelerated frame transform and a Rust YOLO NMS post-process.
- Non-detector models (face, LPR, etc.) routed to the GPU so the ANE stays reserved for detection.
Added — Native macOS app
- Signed, notarized
.appwith bundled ffmpeg, ffprobe, go2rtc, nginx, and Python. No Docker, Homebrew, or virtualenv. - Menu-bar app: live status, performance metrics, log access, settings, license screen.
- First-run welcome wizard for folder/permission setup and dashboard sign-in.
- Launch-at-login item; process supervisor with automatic crash recovery; in-app "Check for Updates".
- Installer DMG with eject-on-first-launch prompt.
Added — Detection & models
- Support for arbitrary ONNX detector models, including Frigate+ models.
- Audio detection (YAMNet), Semantic Search, face recognition, and license-plate recognition working on macOS.
- On-device custom-model training.
Added — Recording & storage
- Selectable recording-cache backing: in-memory (reduces SSD wear) or disk.
- Recording-cache forensic diagnostics on low-disk conditions.
Added — Licensing
- $10 one-year update window. Releases published during the window remain usable indefinitely after it lapses; the installed version keeps running.
- 30-day trial, no card required. Runs offline once activated.
Changed (defaults differ from upstream Frigate)
- VideoToolbox hardware decode/encode is the default for camera input, recording, the go2rtc re-stream, and timelapse export.
- Detector defaults to CoreML/ANE.
- Web dashboard serves over HTTPS by default.
- Detection runs on full-resolution streams; a separate detection sub-stream is not required.
- Storage and disk-usage accounting uses macOS-native calls (correct on large and network/NFS volumes).
Removed (not applicable on macOS)
- Non-Apple detector backends (Coral/EdgeTPU, NVIDIA/TensorRT, Intel/OpenVINO, and other PC-only accelerators) — superseded by the CoreML/ANE detector.
- Docker-based deployment — superseded by the native app.
Privacy
- All video and detection processing stays on-device.
- Anonymous usage statistics are opt-out and never include video, images, or personal data.
Requirements
- macOS 13 (Ventura) or later; Apple Silicon (M-series).
Credits
Built on Frigate NVR by Blake Blackshear and contributors. Fregata and 3rd Bit Labs LLC are not affiliated with, endorsed by, or associated with Frigate, Inc.
Download
SHA-256: 35725f886737f5b9368347650113f1833d8759bb4794e15657125a6391878ac9
Requires macOS 13 (Ventura) or later on Apple Silicon.