homegui VS openai-kiss

FWIW, I had similar experiences twice:

First with IKEA when they changed the lights on-power-up behavior on upgrade. This resulted in a hub becoming a 25 euro cc2531 stick on raspberry pi + Zigbee2mqtt and HomeBridge for integration with HomeKit.

And the past February Apple decided my Apple TV 3 was not good enough for being a home hub, and botched the entire home in the process, while I was away.

So now the whole HomeKit nonsense is out and instead there is a lightweight Rust app with a simple text file config doing both the orchestration and providing the light html UI.

https://github.com/ayourtch/homegui - in case anybody finds it useful.

As a bonus the users in the household praise the new system being much more responsive..

I am missing “turn all the lights off when last person leaves”, and full editing the colors via GUI, but not enough to bother to implement it :-)

I started with a similar path a few years ago - IKEA smart home gear. The controller went away fairly quickly. I stayed for a long time on raspberry pi + Zigbee stick from DeConz + Apple home integration using homebridge.io.

Deconz felt a bit bloated/buggy, so I switched to using a cheap USB stick from aliexpress, and using zigbee2mqtt, which turned out to be a pretty good combination.

Unfortunately Apple decided (while I was on a business trip) that Apple TV 3 was no longer good enough as a home hub, so all of my soft switches stopped working...

As a result in a couple of days I wrote a quick MVP replacement in Rust, which has been working very nice for me, and the "user" (significant other) feedback was that it is now much more reliable and responsive.

Putting it here in case anyone finds it useful.

https://github.com/ayourtch/homegui

While we are at bloated things reimplemented in a few lines of shell: https://github.com/mlang/openai-kiss is my CLI client for ChatGPT. No Python, just curl and jq.

The "framework effect" seems to be a thing these days, not only with SPAs using the newest JS framework.

When I stumbled across an HN article describing a CLI app for the OpenAI API, I thought that is just what I was waiting for. Unfortunately, when I tried it, it didn't work for me. After a debugging session, I found a bug in the openai python package which led to my key not being used if I have ~/.netrc exists. Pretty much a demonstration of the "framework effect". The simplest things, read CLI tools, become buggy in subtle ways because you inherit every bug deep down in the dependency chain. I ended up implementing my own OpenAI API CLI client in just a few lines of shell. Much easier to review, and almost impossible to have subtle hidden bugs.

https://github.com/mlang/openai-kiss