Please make it possible for a user to make selections of what entities are visible for a voice assistant. So that a per voice assistant selection of entities are used.
I have the assist pipeline connected to asterisk to screen incoming calls. (when its late or im busy)
Now im creating a AI call out system. But when someone speaks its name it checks the script for checking people allowed to be granted access when someone reaches the call screening. And if the name isnt found the system hangs up the call.
I have a hard time including excluding/including this in the prompt.
Functional code that sets the entities per voice agent will fix the issue.
This feature will also be very usable if you have a LLM locally in combination with a big cloud LLM. Or a slow 24/7 LLM and a big LLM system that can handle a bigger prompt.
This would be near impossible to execute. Would require a separate registry of entities per llm.
Explain why you think you need the separate entities for the separate agent and im sure we can help you work around it. I’m driving per instance prompting on Friday’d main conversation agent.
I don’t have a voice agent endpoint in the guest room for a reason… But it’s not voice.
Hate to break it to you HA currently has ZERO security boundaries inside the app. You need that fixed long before you worry about something that will require fundamental re-archetecture of the system. (read:hold your breath, you turn blue first)
There is a way to identify the active voice endpoint and bail on scripts if the endpoint is active. But it requires you identifying the active endpoint and the intercepting every tool that you want to filter. (I do something similar with a ‘MPAA rating’ in my house and completely disallow certain tools from running.)
They are adding verbal confirm for sensitive entities in the next version bit it won’t be a security feature. Just an accidental oops preventer. Don’t expect per agent exposure for a LONG time.