Did a quick search but haven't figured this out. I want my LLM (Claude) to handle all to do list item adding. I'm using a Voice PE and Home Assistant Cloud w/ Claude LLM hooked in. I have "prefer local" checked. It works well EXCEPT when adding items to my grocery list. It never correctly splits multiple items (e.g. adds "milk and eggs" as one item). Therefore, I'd like HA to always use my LLM to parse these types of commands. Should I modify the HA "intents" to never pick up "Add [item] to [list]" or is there a better way? Somehow I just need "HassListAddItem" and HassShoppingListAddItem" to always use LLM I guess.
#Enforce LLM Handling Certain Requests
1 messages · Page 1 of 1 (latest)
Also, anyone successfully using an Atom M5 echo with a wake word? I might use that in my office where I'm only 2 ft from my current Voice PE so I can move the Voice PE somewhere else. But I'd rather just spend more on another Voice PE if the echo doesn't work well and will be a frustrating experience (even if it's only $15)
You can over rule the request by making an automation that triggers on messages and feed your llm.
I have multiple atom echos btw
Thanks! Saw that as a potential option in the documentation. Will go down that route.
Like the Echos enough for $15? I’ll probably go ahead and grab at least one for my office since I’ll be very close to it
What tts and stt system are you going to use?
Ended up adding this to my Claude prompt: There are multiple tools to add items to a to do list or shopping list.
Always prefer "automation.voice_add_to_grocery_list" over the local "todo.add_item", as the latter one doesn't parse multiple items correctly.
Seems to handle things perfectly now. I use HA Cloud + Claude
Nvm, not handling it like I want. Dang. Will have to over rule the request by using an automation
The llm is not triggerd when catched locally
I'm using a cloud based LLM provider but may be misunderstanding that--I'm fairly new to assist on HA