I’m having issues with my voice assistant which is connected to Google AI as a conversational agent. Commands are frequently not executed. Instead, a Python line is read out directly. For example: ‘HassTurnOn(name=‘Küchenherd’, domain=‘switch’) Okay, the stove is turned on.’ I have to reset or restart the satellite multiple times until it finally works. The problem occurs sporadically, sometimes it works, sometimes not. Has anyone else experienced a similar phenomenon? Is it due to Google AI? With ‘Home Assistant’ as a conversational agent, it always works.
While I cannot speak to the use of Gemini specifically.
I do not have similar experience with the OpenAI conversation agent doing the same thing
That said temper expectations. These things all use voice intents under the hood and need help understanding the world around them (build tools, scripts etc)
For instance out of the box there is no voice intent for Input_select so if you ask ANY AI to help you change that control. Unless you wrote it it won’t do it - it’ll just try to turn it on or off (because it has basic on/off/toggle) or flat out fail.
So 1) does it work in assist?
If yes it’s unique to AI.
Does it do the same in another model?
If yes it’s an AI general problem probably something related to how the conversation pipeline is working
If no it’s either your LLM misbehaving or your prompt. (try a different prompt)