#Personally I can t wait till the
frosty fable
How would you expect such logic to work ? Just if HA can't understand it, we forward it ?
Or would you expect it to be "smart" in deciding if OpenAi is a good fallback, and if so, how ?
tawdry lily
I've written and scrapped a small custom integration that did conversation agent fallback, i.e. the user defined an ordered list of conversation agents and the integration created a new conversation engine to use in a voice assistant. Then I tried them in order until there was one which didn't fail #devs_voice-archived message
it worked well apart from the fact that there are no means of listening for when a conversation engine gets loaded so that I could create the pipelining agent only after all dependencies were loaded, so i scrapped it
but probably the holy grail would be multi-language fallback, i.e. I say something and it tries all pipelines in an order I define
for everyday purposes, I guess ordering some conversation agents is good enough IMHO
spiral topaz
It's a very interesting question but perhaps what might make it easier is if you have to include some sort of keyword(s)for example if your house's name is "House" or "Betsy" perhaps the logic could be if the word house or Betsy is mentioned with an entity name then it's an ha thing
The name house or Betsy can be user defined
Else goes to open AI
tawdry lily
@spiral topaz what's wrong with trying HA first and if that fails, forward to the next conversation agent? you don't have to say Betsy anymore and it's not like you want to ask OpenAI if there's smoke detected in your home, right?
spiral topaz
That's also a very good point because it already fails if you ask it something that it cannot do
Where it says sorry I couldn't understand that that would be the channel to open AI