I can recommend BrulTech. I put in a system (3 GEM’s and a Dashbox logger) to cover 3 full panels (plus water, gas, and temps) about 2 years ago and I’ve been very happy with it.

Quick review: It’s all local and is well documented and open at the integration level - with HA there’s already an integration built in for it out of the box (I recommend getting the latest from github as they are actively developing it). Firmware is proprietary but low level (read sensors, send via serial->ethernet convertor, sleep 5 secs, repeat; configuration UI). Support is technical and high quality, install/setup docs are excellent and give good examples setting up and and configuring the CT’s.

My setup: I have the dashbox set up to proxy the raw GEM data to HA’s Brultech integration which then feeds all the HA power/water/gas/temp features. The GEM’s I have read 32 CT channels each, plus have 8 ‘counter’ inputs (gas, water) and several 1wire bus I/O’s to talk to temperature sensors. They send a data snapshot of all readings every 5 seconds to the logger and to HA (greatly filtered).

The dashbox and 3 GEM’s are on wired ethernet, they also have integrated WIFI as an option. They talk on my local net between each other and the dashbox pushes messages to HA - nothing talks to anything on the internet or needs any connectivity to set up.

The HA integration reads GEM-packets, the dashbox repeats the packets received from the 3 GEM’s to the HA integration in my setup - but you can also skip the dashbox and just point the GEM straight at HA.

Some examples from the data logger dashboards:

I have a subscription in the US and have been watching Picard and Lower Decks – the few ads I get are their up front spam promoting other shows (I hit reload and the ‘forced’ up front ad switches to the CBS logo and plays in a couple of secs). Another show I watch has a few ads, I think the are trying to see how much they can get away with before people cancel. I bought a year at 1/2 price and I won’t be reupping, its a pretty low value and quality streaming service on all fronts (ads, content, video quality, app/site navigation and player, absolutely no useful technical support if their player or service is glitching, cable company class customer avoidance policies and systems).

The video quality on easynews is also much better should I want to watch more star trek in the future.

The ‘platform’ for the FEIT ones at costco are really well built. They make the Insteon units look like they were a one off hobby project. That’s the part that has the power circuitry, screw terminals, etc. The wifi module used to be ESP based but they switched about a year ago to a newer BK723 chip, these can be reflashed with a ‘OpenBeken’, which is a newer project that has similar functionality to Tasmota but for these newer chips.

The only downside I’ve seen with these is OpenBeken is not as mature and the units ‘hang’ now and then requiring a power cycle (like 2-4 weeks). I have some of the older ESP based units that are otherwise identical and they never hang, this has been getting better with upgrades so likely will eventually be fixed.

If you are OK with soldering you can buy the ESP controller modules on Amazon for about $1 each and just replace them and run stock Tasmota (you have to unsolder the module to reflash it either way, so its not a big deal to just put a different one on). The module pinout and electrical interface are identical.

Its not super hard to manually set up with docker or podman but you have to deal with integrating and updating the add-ons yourself. I ran out of CPU on a pi4 (due to a buggy websocket client in the end) and moved to a small form factor x86_64 server under Rocky. I ran manually using just containers (podman in this case) and it worked fine but integrating and updating the equivalent of add-ons was a lot of manual plumbing work that I don't find much fun anymore.

I switched back to hassos, but under KVM. This for me is the best of both worlds: I get the fully managed/integrated work of Frenck and friends for HA and can still access and manage the machine normally (and use it for other services).

There's nothing remotely realtime about the python code in the core HA, it works well in a reasonbly provisioned VM (4 cores, 8G ram) backed by a good SSD. There is some religion in the community about not using VM's: it is a layer of complexity and I understand why folks on discord don't want to help people with it, but technically it works well for this class of app.

I just got it last night too, we'll see how it goes!

I'm hoping they will be able to iterate faster now that they have their Dojo setup running, they're certainly getting TONS of training data from the fleet now.