Good read!

"...but I eventually found the brilliant llm-intents integration..."

immediately followed by this:

"...the results were mediocre."

made me chuckle though. I'll probably jump in this pool once we achieve better performace/response times - both the silence and the idle dialog during that 2-5 seconds are maddening to me. I do not need loading screens irl.

This is something I'm going for too. Currently just running Ollama integration with HA for voice control but I'd be moving STT/TTS locally too. The new Qwen 3.5 35B loads fairly quickly and responds quickly enough to be usable for conversation on my system. The 9B version is extremely fast and adequate for voice control.

Maybe I'm dumb but I tried to do this a few months ago and it was horrible. Couldn't do simple things unless I said the exactly perfect name of entities. Even then took forever to do anything