Fregata.app: A port of Frigate to macOS

[Looksee]

In the Frigate Web UI: Settings > Global Configuration > Camera Management > Delete Camera button

u figure anything out?

Sorry for my delay in replying.

No, I didn't. I just dumped everything, started again from scratch, and loaded the cameras I wanted. At that time I just wanted to get some cameras connected.

I still don't see any settings labeled as Global Configuration, Camera Management or Delete Camera.

I can go to Settings > Cameras > Management

There, I can add cameras and disable or enable them, but only temporarily.

I think things look interesting and I will get back to playing with Fregata/Frigate, but I had an argument with Myself. "Why are you wasting time on this? You have a new NVR that you haven't even set up yet?!!!"

Myself and I can be such sticks in the mud sometimes.

 

[modplan]

Lunar lake isn’t your old Intel chip. On chip lpddr5x, NPU4- 8 bit native (48TOPS) with full 16 bit support (24T half FLOPS). VVC hardware decode. Big iGPU with BF16 support. Getting mini pc form factor brings PL1 to 30w and nearly doubles steady throughout. Yolo graphs especially well to Intel NPU. Yolo26L at FP16 precision and 320x320 input can easily hit 200 images per second or more on NPU alone after no NMS or e2e. That also keeps CPU about 85% less utilized compared to GPU inference.

I ended up buying poorly supported hardware due to specs before. Jetson AGX with dual NVENC and dual NVDEC can transcode 30 HD videos at once but is not worth the hassle.

Both Intel and Apple can handle janky LLM ports with predictable tensors just fine but neither are powerful enough to do anything especially useful. Both graph YOLO reasonably well if tensor shapes remain predictable. Nobody knows if that’ll be the case in 2028. Both are complete garbage at other SOTA AI tools using cross attention and built on custom CUDA kernels.

You can’t future proof with specs, you can only hope that the community builds for your specific hardware.

Fregata on an M6 will blow any $800 intel mini PC Frigate docker install out of the water in terms of object detection speed and AI enrichment GPU performance.

If you want to run other docker services on the same machine I might agree with you tho. Docker performance, even with OrbStack, is lacking on Mac.

 

[modplan]

In the Frigate Web UI: Settings > Global Configuration > Camera Management > Delete Camera button


Sorry for my delay in replying.

No, I didn't. I just dumped everything, started again from scratch, and loaded the cameras I wanted. At that time I just wanted to get some cameras connected.

I still don't see any settings labeled as Global Configuration, Camera Management or Delete Camera.

I can go to Settings > Cameras > Management

There, I can add cameras and disable or enable them, but only temporarily.

I think things look interesting and I will get back to playing with Fregata/Frigate, but I had an argument with Myself. "Why are you wasting time on this? You have a new NVR that you haven't even set up yet?!!!"

Myself and I can be such sticks in the mud sometimes.

View attachment 250568View attachment 250569

It looks like you might be on an older version? Those look like the v0.17 settings. v0.18 has way more in the UI.

 

[Looksee]

Yes, 17.x is what I was using. I have downloaded the latest and wiil try it later.

 

[OD1]

Fregata on an M6 will blow any $800 intel mini PC Frigate docker install out of the water in terms of object detection speed and AI enrichment GPU performance.

If you want to run other docker services on the same machine I might agree with you tho. Docker performance, even with OrbStack, is lacking on Mac.

16GB LNL minis are about $500 and when you buy them, you own them. You don’t have to ask permission from Apple periodically to keep using it. You can even fetch updates without checking in first.