Raspberry Pi AI Kit: M.2 HAT+, Hailo AI Module, 16mm Header, Spacers, Screws - Bluetooth Connectivity - Single Board Computer

Kits came very nicely packed with all required hardware. Fits like a glove on the pi 5. Work seamless with any pi camera. Works well with multiple versions of yolo.This is not designed for LLM/chat bots. Primarily different types of image recognition. The documentation is good, however one of the example yolo projects doesn’t work. I would buy again

Easy as Pi. Added to my Pi 5, running Mistral 7B locally/offline without issue. Not as slow as I expected.

I wish that they made it more obvious that this only works with visual AI and doesn't do much for LLM's. Still a solid product, and it's easy to install and operate.

Can add extensions, just be careful.

Extremely poorly designed system. All the benchmarks you're seeing on Hailo model zoo, divide those FPS by 2.5, that is the real speed of the unit. You can not program with their system like you would with a normal system like NVIDIA Orin, or on a laptop using GPU... you HAVE TO USE THEIR APPs, or nothing works. Which means you have very little control of your programs, you are constrained by their apps, and the whole system becomes brittle. You can not just install their libraries and use it, you have to download their entire file structure, and then use their own venv environment to get anything to work, therefore you will have to migrate your entire project into their setup rather than the other way around.AVOID at all cost. Better to spend a few extra dollars and buy a different system. There is a reason why you don't see this being used in the industry. I am in the process of returning mine, and purchasing Orin Nano. This is just not worth the hassle, between the price of RPI5 + heatsinks/fan + Hailo, you might as well throw in a few extra dollars and get the Orin.

Standoffs are too tall to use if you also want to use the GPIO pins

All I'm trying to do is run a custom yolov8 model. Easy right? Nope, took forever to compile to their proprietary format. Okay, so it should be easy to run that on their hardware? Nope. Getting a nonsense error.Funny enough I was able to get the model working for Google Coral AI Accelerator (but they've stopped/will stop support for these chips soon) and Rockchip (but the whole system was too unstable), but for some reason, this platform takes the cake with how terrible it is to work with.