Hoping this is the right place to ask this and that someone can help. I have support from when I purchased my start9 server but couldn’t find anything there. I’ve been looking at switching my smart home devices and cameras to Home Assistant and use Frigate NVR for the cameras and doorbell and get away from my cloud based subscription cameras. AI tells me I need to get a mini-pc for this as it can’t be done on my Start9 Pure server. I’m still on version 0.3.5.1, but it looks like HA is now available in the Start9 registry for 0.4.0. If so, is there any chance of Frigate being added, or an equivalent for IP cameras and local storage that integrates with Home Assistant. I’d prefer to take advantage of the Start9 server and not have to get another server for this. Thank in advance!
From what I can tell looking at Frigate NVR, this could definitely be packaged for StartOS v0.4.0 – the only things that really can’t is hardware that needs to be plugged into the server and detected. That doesn’t seem to be the case here, thanks to the nature of the IP cameras.
If you look me up on the support server and send me a DM, I’ll explain how we could arrange getting Frigate NVR set up (it’ll be you doing the testing)
I think I sent you a DM on the support server, but it just kept scrolling and not sure it completed. Please let me know if you got it or not. Thanks!
I don’t think I have anything from you.
Hi, I am interested as well on this as I am using Home Assistant OS with the Frigate addon ( GitHub - blakeblackshear/frigate-hass-addons: Frigate hass.io addons · GitHub ) with the google coral usb tpu for inference and i wanted to understand whether a similar solution could be done fully on startos.
So Frigate could be installed inside HA as in my HAOS setup (not sure if startos is designed to allow that level of integration) or as a separate startos service, but the main point for me is being able to use the coral usb device with Frigate, @StuPleb would you say something like this could be done and is just a matter of writing the correct startos package, or could there be substantial limitations by the startos itself?
There’s no feature in StartOS (yet) that allows passthrough of USB devices plugged into the server to connect to services running on the server. That will come in time.
What will be possible is to package a service that connects to Home Assistant via the network, which then in turn connects to cool stuff (i.e. cameras) also via the network.
Ok, and what about a server with a GPU instead of the usb coral device, would it be possible to make a startos Frigate service use the gpu? I see the marketplace has AI stuff like llama.cpp, ollama and vLLM so I guess PCI devices are usable?
Also, what is the state of vulkan support in startos services? There are a lot of AMD GPUs not usable with ROCm, so Vulkan is the only way to use them (upstream llama.cpp does work without problems with Vulkan on AMD GPUs). I see on the startos llama.cpp github page (GitHub - Start9Labs/llama-cpp-startos: llama.cpp llama-server packaged for StartOS · GitHub) that non-Intel Vulkan devices fallback to generic (CPU and RAM), but is that because of startos limitations/configuration, or could it be solved with the correct/custom llama.cpp container image/package?
If this is off topic for the original question, I’ll create a new thread, just let me know.
Immich is a good example of a non-AI focused service that leans heavily on StartOS’s ability to provide access to GPU features. Take a look at that one to see how it works.
Ok I have checked and that’s not really different from the llama.cpp package, still no AMD Vulkan option, BUT it was useful as it seems vulkan compatibility is not a matter of the startos system itself, but only of the service package, as Vulkan should be all userspace and so it only depends on the actual service image being used (startos already has amdgpu stuff and already passes /dev/dri and similar paths to the services if needed, right?), llama.cpp has a vulkan version, while immich does not.
All this to say that it should be possible to modify/update the llama.cpp package in order to support a more general vulkan option which is compatible also with AMD cards, and as a consequence a Frigate image that uses vulkan with AMD gpu (or even frigate installed directly inside home assistant maybe).
As a side note: I still don’t understand why projects never supports vulkan before rocm for discrete graphics cards, with rocm you have less compatibility and it should be more bloated/difficult to setup and manage, but maybe it is just me 
.
Anyway, thank for your answers, if I try something I will let you know.