Hi, is posibble to tell assistant to
# general-english
k
Hi, is posibble to tell assistant to talk to user, somethink like "wait a moment, a search data" ? We use to talk with our system, with RAG from vector and sql db, and sometimes this take a moment, and when I ask about somethink by phone, i must wait without info, what is doing now.
p
Are you using custom LLM? There's this
<flush />
tag.
k
Yes, custom LLM, i will check, thanks
p
So turntaking is Assistant > User > Assistant... You cant have 2 assistant messages in a row. But.. You can yield chunk with
"one moment, please...<flush />"
This forces this message to be spoken earlier, then you send you LLM response chunks and finally the
[DONE]
stop_reason="stop"
chunk
I believe thats the best VAPI can do at the moment
k
i try find this in documentaction, we have fast api backend, user call by twilio, ask about something, this goes to our endpoint, and we response at the end, in what place i need use this flush?
p
I dont think there are docs. So whenever you start long operation, you can instantly send chunk with message and tag at the end then proceed untill you get LLM output
So basically when user stops speaking, you get vapi payload, you know user probably stopped speaking
You can send this chunk right there straight away
a
You could also is the messages delay option of what to speak while a calls to a tool is happening
2 Views