Should I run Scrypted NVR and Frigate on one N100 mini PC?

Only if you budget separate decode and inference capacity. Stacking Scrypted transcoding, Scrypted NVR Smart Detections, and Frigate Coral inference on one N100 without QuickSync tuning commonly produces frame skips. Split hosts or run Frigate plugin inside Scrypted instead of duplicating full NVR stacks.

Which stack keeps detection fully offline after setup?

Frigate on direct RTSP feeds is the cleanest offline path—no plugin billing checks. Scrypted NVR paid plugin may require periodic license validation unless you stage WAN carefully; free Scrypted plus Frigate on a blocked VLAN avoids that subscription layer.

What hardware do I need for six cameras with local detection?

Budget an Intel N100 or Core Ultra box with QuickSync for decode, one Coral USB or Hailo-8 for Frigate-scale inference, and 16 GB RAM if Scrypted rebroadcasts cloud cameras. Scrypted NVR-only on N100 OpenVINO can work for six 1080p streams when motion is tuned; add Coral or Hailo when alerts must survive glare bursts.

Comparisons

Scrypted NVR vs Frigate: Local Object Detection

Q: Is Scrypted NVR faster than Frigate for object detection?

On identical Coral hardware, Frigate's published per-frame inference (~10 ms MobileDet on Coral) is lower than Scrypted NVR's bundled yolov9t batch benchmarks on an Intel N100 (~25 s for 2,000 concurrent detections in Scrypted's August 2026 script). Scrypted NVR wins when OpenVINO on Intel iGPU or Apple CoreML replaces Coral entirely—not when both run Coral on the same undersized box.

Q: Can Frigate publish HomeKit Secure Video without Scrypted?

Frigate does not natively publish HomeKit Secure Video accessories. You need Scrypted's HomeKit plugin upstream, Home Assistant's HomeKit Bridge with Frigate camera entities, or a dedicated bridge—each adds latency and a separate trust boundary compared to Scrypted NVR's integrated HKSV path.

Q: Does Scrypted NVR use the same Coral models as Frigate?

No. Scrypted NVR's Tensorflow-Lite plugin defaults to yolov8n quantized models on Coral, while Frigate defaults to MobileDet-class graphs on Edge TPU. Latency and night-IR recall are not interchangeable without re-tuning both pipelines.

Scrypted NVR vs Frigate for local object detection: inference latency on Coral, OpenVINO, and Hailo, CPU/GPU decode overhead, and HomeKit Secure Video paths.

Privacy Smart Home Research Desk Aug 09, 2026

Keywords: scrypted vs frigate, Scrypted NVR object detection, Frigate local AI latency, HomeKit Secure Video local NVR, Coral OpenVINO detection benchmark, Scrypted NVR vs Frigate detection

Scrypted vs Frigate for local object detection splits along two different detection products: Scrypted NVR runs Smart Detections inside a paid plugin with OpenVINO, CoreML, ONNX, or Coral Tensorflow-Lite backends, while Frigate runs motion-gated inference natively on Coral, Hailo, OpenVINO, or GPU with published per-frame latency tables. As of August 2026, Frigate’s Coral MobileDet path reports ~10 ms inference per detection frame¹; Scrypted’s public object-detection benchmark script reports ~25 seconds to complete 2,000 batched detections (yolov9t, eight concurrent) on an Intel N100 using OpenVINO GPU²—roughly 80 effective detections per second aggregate, not a direct apples-to-apples millisecond duel, but enough to plan hardware.

Quick answer: Scrypted NVR or Frigate for local object detection?

Choose Frigate when you need open-source zone detection, Coral/Hailo latency tables, and Home Assistant MQTT events on RTSP cameras. Choose Scrypted NVR when Smart Detections, face/plate search, and HomeKit Secure Video publishing are primary—and you will run OpenVINO or CoreML instead of aging Coral sticks. Many stacks run Scrypted for HKSV and Frigate for automations via RTSP rebroadcast.

Source: Frigate hardware docs and Scrypted object-detection benchmark

~10 ms — Frigate’s documented Coral MobileDet inference per detection frame versus Scrypted NVR’s ~80 det/s aggregate throughput on N100 OpenVINO (yolov9t batch script, August 2026).

Executive summary

Homelab builders searching scrypted vs frigate in August 2026 usually already run one stack and wonder whether the other duplicates work. Scrypted NVR is a commercial plugin layer: timeline Smart Detections, rich push notifications, optional AI summaries, and native HomeKit publishing when paired with Scrypted’s HomeKit plugin⁴. Frigate is the open-source detector-and-retention engine Home Assistant users treat as ground truth for zones, masks, and MQTT automations¹.

Methodology: We normalized detection performance by cross-walking Frigate’s Recommended hardware detector table (accessed 9 August 2026)¹, Scrypted’s object-detection-benchmark script results (GitHub mirror, 9 August 2026)², Scrypted Server Hardware buyer guidance (9 August 2026)³, and Scrypted NVR Features documentation for Smart Detections backends⁴. CPU/GPU overhead classes are editorial scores (1 = light, 5 = heavy) derived from ffmpeg/QSV documentation and our existing six-camera reference builds in the Ugreen Frigate NVR guide.

Verdict: For Priya, a Home Assistant owner with six Reolink PoE cameras and no Apple TVs, Frigate with Coral or Hailo is the right detection layer—she gets inspectable YAML zones and ~10 ms-class inference without Scrypted NVR subscription checks. For James, who watches doorbell tiles on HomePods and wants HKSV clips with Scrypted’s timeline UI, Scrypted NVR with OpenVINO on an N100 or Core Ultra is the coherent product—even if he later rebroadcasts RTSP to Frigate for HA automations. Where I’m less sure—whether Scrypted NVR’s yolov9t night recall beats Frigate MobileDet on the same IR camera—the data is thin; run a rainy-night clip through both before you firewall WAN.

What each detection stack actually runs

Buyers conflate Scrypted the platform with Scrypted NVR the plugin. Detection behavior depends on which path you enable.

Layer	Scrypted NVR (plugin)	Frigate (native)
Detection trigger	Smart Detections on motion + AI plugins	Motion-gated object inference
Default accelerators	OpenVINO (Intel GPU/NPU), CoreML, ONNX, Coral TFLite	Coral, Hailo, OpenVINO, ONNX, GPU
Typical models	yolov8n / yolov9t (plugin-dependent)	MobileDet, YOLOv6n, Frigate+ models
Output consumers	Scrypted NVR apps, HKSV, rich notifications	Frigate UI, MQTT, HA integration
License	Paid NVR plugin + open Scrypted core	MIT open source

Smart Detections in Scrypted NVR populate timeline and event views with people, vehicles, packages, and animals—backed by the same object-detection plugin family documented on Scrypted’s detection page⁵. Frigate’s object detector runs only on cropped motion regions, which is why a single Coral stick can sustain six substreams when masks are tuned¹.

Original research: detection latency and throughput matrix (August 2026)

This is the citable dataset for the article. Cells label Source class so you know what we measured versus what we inferred.

Stack	Hardware	Model (default class)	Metric	Value	Six-cam planning class	Source class
Frigate	Coral USB	MobileDet	Per-frame inference	~10 ms	Comfortable @ 5 fps detect	Frigate doc¹
Frigate	Hailo-8 M.2	YOLOv6n	Per-frame inference	~7 ms	Comfortable @ 5–10 fps	Frigate doc¹
Frigate	CPU OpenVINO (N100)	MobileNet	Per-frame inference	~15 ms	Testing only	Frigate doc¹
Scrypted NVR	Intel N100 iGPU	yolov9t	Batch benchmark (2,000 det)	~25 s (~80 det/s)	Moderate; tune motion	Scrypted script²
Scrypted NVR	Coral USB (TFLite)	yolov9t	Batch benchmark	~20–21 s	Moderate; Coral deprecated path	Scrypted script²
Scrypted NVR	Intel 125H iGPU	yolov9t	Batch benchmark	~10 s	Strong	Scrypted script²
Scrypted NVR	Apple M1 Ultra (CoreML)	yolov9t	Batch benchmark	~5 s	Excellent	Scrypted script²
Dual stack	N100 + Coral	Frigate detect + Scrypted rebroadcast	Editorial CPU load	High (4/5)	Split hosts if stutter	Editorial

How to read the table: Frigate publishes single-frame inference milliseconds—the number you see in the Frigate UI stats panel. Scrypted’s benchmark script runs 250 iterations × 8 concurrent detections (2,000 total) and includes image upload overhead to the detector worker². Comparing 10 ms Frigate to 25 s Scrypted batch is misleading without context; use aggregate det/s for capacity planning and per-frame ms for single-stream alert latency.

Detection auditability — Scrypted NVR vs Frigate (August 2026)

Product	Cloud required	Local storage	Mandatory account	Offline control	Score / 10
Frigate native detector	No	Full YAML + clip control	None	Excellent on RTSP	9.4
Scrypted NVR Smart Detections	No for LAN inference	NVR plugin retention	NVR license + vendor cam logins	Good; license checks may need WAN	7.6

CPU, GPU, and accelerator overhead

Neither stack magically removes ffmpeg work. Decode (H.264/H.265 → frames) and inference (frames → labels) are separate budgets.

Decode overhead (QuickSync / VAAPI)

Frigate documentation treats Intel QuickSync and preset-intel-qsv-h264 as the default path for six 1080p substreams on N100-class hosts¹. Scrypted’s server buyer guide ranks GPU decode above CPU for motion detection workloads and explicitly warns that TPU/NPU does not decode video—only inference³.

Workload	Frigate-only (6×1080p sub)	Scrypted NVR-only	Scrypted rebroadcast + Frigate detect
QSV decode load (editorial 1–5)	2	2–3	4
Inference load (editorial 1–5)	2 with Coral	2–3 OpenVINO	3–4
Combined CPU peg risk	Low with QSV+Coral	Moderate	High on one N100

Anecdotally, users who pass /dev/dri only into Frigate while Scrypted software-decodes upstream report double CPU spikes during afternoon glare—your mileage will vary depending on whether substreams are shared via rebroadcast instead of duplicate camera pulls.

Inference overhead (Coral vs OpenVINO vs Hailo)

Scrypted’s August 2026 hardware guide calls USB Coral “high latency and crash prone” and “Google Abandonware” for new purchases, recommending Intel iGPU or Core Ultra NPU instead³. Frigate still documents Coral end-to-end and adds Hailo-8 at ~7 ms as the greenfield accelerator path¹.

Steel-man for Scrypted NVR on OpenVINO: A Scrypted advocate would argue Frigate’s Coral obsession is legacy—an N100 iGPU running OpenVINO hits ~25 s on the standardized yolov9t batch versus ~20 s on Coral USB in the same script, while also powering HKSV transcoding and timeline UI in one product. You buy one license, one app, and Apple’s Neural Engine or Intel’s NPU on Core Ultra if you scale beyond eight cameras.

Rebuttal: Frigate’s motion gating and zone masks reduce inference invocations Frigate+ users can measure in MQTT logs. Scrypted NVR Smart Detections on untuned motion can flood the detector during sun glare—batch throughput matters less than false-positive rate. For Home Assistant-centric automations, Frigate’s ~10 ms Coral path with proven night MobileDet behavior (see our Coral vs Hailo benchmarks) still wins when HKSV is not required.

HomeKit Secure Video integration

Frigate does not publish HKSV accessories. Apple live tiles and encrypted iCloud clip upload require an upstream bridge.

HKSV path	Detection owner	Typical live-tile latency	Clip storage	Privacy tradeoff
Scrypted HomeKit + Scrypted NVR	Scrypted NVR plugins	Lowest in Scrypted stacks	HKSV (iCloud E2E)	Apple cloud for clips; local LAN for live
Scrypted rebroadcast → Frigate → HA HomeKit Bridge	Frigate	Higher; multiple hops	HA snapshot / Frigate clips	More moving parts
Frigate-only (no Apple)	Frigate	N/A	Local disks only	Maximum local control

Scrypted documents Adaptive Bitrate as a platform feature that also improves HomeKit, Google Home, and Home Assistant streaming—not only Scrypted NVR playback⁴. That transcode layer is orthogonal to detection but competes for the same GPU blocks you need for OpenVINO inference.

For encryption and retention tradeoffs, cross-read Apple HomeKit Secure Video vs local NVR. For Ring/Arlo bridges before detection, see Arlo and Ring to Frigate via Scrypted.

Technical comparison diagram of Scrypted NVR versus Frigate local object detection stacks showing inference latency paths on Intel QuickSync decode, Google Coral TPU, OpenVINO GPU, and HomeKit Secure Video integration for privacy-focused smart home surveillance in 2026. — Detection silicon is often shared; HKSV and rebroadcast overhead is what splits Scrypted NVR from Frigate-only builds.

Worked examples

Example 1 — James, townhome, HomeKit Secure Video primary

James owns a Reolink Video Doorbell PoE and four Arlo Pro 5 cameras. He watches feeds on a kitchen HomePod and wants person alerts without Arlo Secure. Stack: Scrypted with Arlo and Reolink plugins, Scrypted NVR with OpenVINO on a Beelink N100 (16 GB RAM), HomeKit plugin for HKSV, optional Frigate ingesting Scrypted RTSP for HA automations. Detection expectation (August 2026): OpenVINO yolov9t batch class ~25 s / 2,000 det on N100²—adequate for five cameras at 5 fps detect after motion tuning. Cost check: ~$180 N100 + Scrypted NVR license (check Scrypted store, 9 August 2026)—no Arlo subscription. Risk: vendor OAuth; James keeps WAN staged for token refresh only.

Example 2 — Priya, six PoE bullets, Home Assistant only

Priya installed six Reolink RLC-810A on VLAN 40 with WAN denied. No Apple hardware. Stack: Frigate only on Proxmox with QuickSync + Coral USB, HA Frigate integration for person-at-driveway automations. Detection expectation: ~10 ms inference, ~30 fps aggregate detect load at 5 fps × six cameras¹. Scrypted adds no detection value. Cost check (July 2026): ~$250 compute + $65 Coral on secondary markets. I haven’t tested this exact VLAN layout with Scrypted NVR license checks offline—verify if you duplicate Scrypted on the same host.

Decision matrix

Your priority	Choose	Why
Lowest published per-frame inference on Coral/Hailo	Frigate	~7–10 ms tables¹
Best Intel iGPU / Apple CoreML without Coral	Scrypted NVR	OpenVINO/CoreML benchmark leaders²³
HomeKit Secure Video + Smart Detections timeline	Scrypted NVR	Native HKSV + NVR UI⁴
Home Assistant zones, masks, MQTT	Frigate	First-class integration
Cloud Ring/Arlo inventory	Scrypted → Frigate	Bridge then detect
Maximum open-source auditability	Frigate	MIT + YAML

Checklist

List cameras: RTSP-native vs cloud-only.
Pick detection owner: Scrypted NVR plugin vs Frigate native (not both unless split).
Enable QuickSync/VAAPI on the host before buying accelerators.
Run Scrypted benchmark script OR Frigate file-mode on one night clip per problem camera.
If HKSV required, plan Scrypted HomeKit plugin before Frigate HA bridge.
Snapshot configs before plugin or Frigate major upgrades.

FAQ

Frequently Asked Questions

Is Scrypted NVR faster than Frigate for object detection?

On Coral, Frigate’s per-frame ~10 ms inference is the tighter latency metric. Scrypted NVR on OpenVINO iGPU wins aggregate throughput on Intel N100 and Core Ultra when you accept yolov9t models and NVR licensing—not a universal speed crown.

Can Frigate publish HomeKit Secure Video without Scrypted?

Not natively. Use Scrypted’s HomeKit plugin, Home Assistant’s HomeKit Bridge, or another bridge—each adds hops compared to Scrypted NVR’s integrated path.

Should I run Scrypted NVR and Frigate on one N100?

Only with careful QSV tuning and realistic camera counts. Prefer Scrypted rebroadcast into Frigate on one host, or split hosts when cloud cameras need transcoding plus Coral inference.

Does Scrypted NVR use the same Coral models as Frigate?

No. Tensorflow-Lite on Coral uses yolov8n-class graphs in Scrypted docs; Frigate uses MobileDet on Edge TPU by default. Night performance diverges.

Which stack keeps detection fully offline?

Frigate on RTSP with WAN-blocked cameras is simplest. Scrypted NVR may need license validation unless you plan offline windows per our Scrypted offline guide patterns.

What hardware for six cameras with local detection?

Intel N100 or Core Ultra with QuickSync, 16 GB RAM, one Coral or Hailo for Frigate-scale loads, or OpenVINO-only Scrypted NVR if HKSV is mandatory and masks are tuned.

Primary sources

#	Source	URL
1	Frigate — Recommended hardware (detector ms)	https://docs.frigate.video/frigate/hardware
2	Scrypted — object-detection-benchmark script	https://github.com/scryptedapp/scripts.scrypted.app/blob/main/object-detection-benchmark.md
3	Scrypted — Server hardware buyer guide	https://docs.scrypted.app/buyers-guide/servers.html
4	Scrypted NVR — Features (Smart Detections, HKSV)	https://docs.scrypted.app/scrypted-nvr/features.html
5	Scrypted — Object detection plugins	https://docs.scrypted.app/detection/object-detection.html
6	Privacy Smart Home — Hailo vs Coral benchmarks	/guides/hailo-8-vs-google-coral-frigate-benchmarks-2026/

Verdict

Scrypted NVR vs Frigate for local object detection is a product-boundary choice, not a CUDA benchmark bake-off. Choose Frigate when per-frame Coral or Hailo latency, motion zones, and Home Assistant MQTT are the contract—especially on PoE RTSP installs like Priya’s. Choose Scrypted NVR when Smart Detections timeline, rich notifications, and HomeKit Secure Video publishing define success, and you will run OpenVINO or CoreML on modern Intel or Apple silicon instead of chasing Coral on eBay.

Run both only with a deliberate split: Scrypted rebroadcasts normalized RTSP; Frigate owns zones and automations—or accept James’s dual-license complexity. Start by measuring one rainy-night clip on your worst camera; forum milliseconds mean nothing if your north-facing IR sensor misses cars at 02:00.

Dataset (JSON-LD)

Frigate Recommended hardware — detector inference table, accessed 9 August 2026. https://docs.frigate.video/frigate/hardware ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹ ↩¹⁰ ↩¹¹ ↩¹²
Scrypted object-detection-benchmark.md — batch results by server class, accessed 9 August 2026. https://github.com/scryptedapp/scripts.scrypted.app/blob/main/object-detection-benchmark.md ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹
Scrypted Server hardware buyer guide — GPU vs TPU roles, Coral guidance, accessed 9 August 2026. https://docs.scrypted.app/buyers-guide/servers.html ↩ ↩² ↩³ ↩⁴ ↩⁵
Scrypted NVR Features — Smart Detections and hardware-accelerated AI, accessed 9 August 2026. https://docs.scrypted.app/scrypted-nvr/features.html ↩ ↩² ↩³ ↩⁴
Scrypted Object Detection plugins — CoreML, OpenVINO, ONNX, TFLite, accessed 9 August 2026. https://docs.scrypted.app/detection/object-detection.html ↩

Quick answer: Scrypted NVR or Frigate for local object detection?

Executive summary

What each detection stack actually runs

Original research: detection latency and throughput matrix (August 2026)

Detection auditability — Scrypted NVR vs Frigate (August 2026)

CPU, GPU, and accelerator overhead

Decode overhead (QuickSync / VAAPI)

Inference overhead (Coral vs OpenVINO vs Hailo)

HomeKit Secure Video integration

Worked examples

Example 1 — James, townhome, HomeKit Secure Video primary

Example 2 — Priya, six PoE bullets, Home Assistant only

Decision matrix

Checklist

FAQ

Frequently Asked Questions

Primary sources

Verdict

Dataset (JSON-LD)

Footnotes

Related guides