When using the Assist function I keep running into the same problem.
The AI assistant can’t help; because the relevant entities aren’t exposed.
I keep needing to expose more.
But there’s a problem.
Some integrations can have a hundred entities (even weather).
And exposing them one by one is tedious.
So as far as I can see you need to:
Go to: Settings->Voice assistants->Expose (tab)->Expose Entities
Search for what you want.
Individually select each one.
Click “Expose”.
Perhaps I’m being a little melodramatic, but surely I’m not the only one with this problem?
Is there a better way?
A filter option?
A “Select All” button I’m not seeing?
I don’t have a solution for what you actually ask but I do have a remark.
It is not a good idea to just expose everything. It is just the other way around. Expose nothing and expose only the things you want to control or use. If you use a LLM you don’t even have to expose entities that you do not want to control. (the prompt does that for you). Keep the number of exposed entities low will make the response much faster.
In my experience it is more often then not that the sentence definition is to tight and the request can’t be done. For example your sentence has ‘newsletter’ in it but what assist recognise ir ‘news letter’. In this case the request will faill
Firstly I want to say thanks for all your good advice, and the excellent troubleshooting article that you linked to.
I tried all sorts of messing with the sentences and prompts in response.
Unfortunately it didn’t really improve the results.
I ended up installing the HACS plugin " Extended OpenAI Conversation", and connected it the latest gpt-4o backend.
This model is smart enough the phrasing makes no difference, and fast enough that there’s no noticeable difference between a few hundred and a few thousand entities.
(Though it does suffer minor delays sending the HTTP request, or larger ones running scripts.)
It was still necessary to manually check a few thousand boxes so it could actually see the entities, but the mobile app at least makes this process a little quicker and easier to do in the background.
Still hoping there’s a fix for mass selection here, but right now it’s working like magic!
