Inconsistencies in finding things

[swift holly]

I can ask the LLM to turn on a helper group of switches and it will turn them on and then ask it to turn off the same helper group and it cannot find it. These.. inconsistencies, can change with every HA update or change when I do a restart of HA. The inconsistencies are different whether the entire HA server has been restarted or just HA.

Also, consider HAs general acceptance will never happen as long as the administration of HA takes more time than saved by having home automation. I think you realize this and are trying to improve that. But with as much as needs to be done in this area, it's going to be quite a while before you get there. Decade+

[whole idol]

Thing is that behavior is usually highly dependent in what llm you are using, and there are countless ones and new ones every day. Also depends on the server running the llm, context window size, quantization, etc. I've personally not seen these kinds of issues, though I've experimented and tweaked my llm server settings and quantizations after a lot of testing and trial and error. 🙂

[crisp mesa]

could also be that you have "Prefer handling commands locally" enabled, which means some well defined commands will not be sent to LLM at all and then subsequent follow up command, which is less defined goes to LLM, but it doesn't have the context from a previous command, so it doesn't know what you are referring to. I disabled this setting so all commands go to LLM and it helped.