#Hello @halfalive , want to instruct

1 messages · Page 1 of 1 (latest)

acoustic pendant
#

There is no straightforward to do it currently. Can you describe the use case some more? I could think about some workarounds

warm rover
#

@acoustic pendant , I see. My use case is when Agent talks a certain topic, it fires my client tool call, opens a Form, then says something like it's now available, please fill in the fields.

acoustic pendant
#

Maybe you can describe what the agent has to say in the tool description/prompt? And reinforce it in the main prompt

warm rover
#

@acoustic pendant , would you mind sharing a quick example? Appreciate it.

acoustic pendant
warm rover
#

Will try it. Thanks. You've been a huge help

acoustic pendant
#

You can also try prompting it like "Tool call response will contain a message that you need to say" or something like this, and include said message in the tool call response

#

And in both cases you can include smth like "After receiving a response to openForm tool, make sure to notify the user..." to the main prompt to make the behavior more consistent

warm rover
#

Will encorporate this also. Thanks again!

warm rover
#

@acoustic pendant , it works! Thanks! So Tool Call Prompts also respects when & what to say

warm rover
#

Hi @acoustic pendant , is it possible to respond to Agent via text instead of saying it?

warm rover
acoustic pendant
#

Text-based communication should be coming soon, according to 11labs devs. As of now, it is not possible

warm rover
#

@acoustic pendant , i see, thanks for confirming. Also, do you happen to know the estimated deadline they’re aiming for?

acoustic pendant
#

Couple weeks maybe, not sure

warm rover
#

I see. Do you have a suggested workaround?

acoustic pendant
#

A workaround to what?

warm rover
#

Sending another way to respond to the agent.

Aside from text, possibly, an audio file right away or something like playing an audio file that would be recognized as an input.

acoustic pendant
#

I would try to add some basic and cheap TTS to turn user messages into audio. It doesn't have to be realistic (probably), just good enough for the agent's ASR to transcribe/understand

#

I don't have any exact options for this though

#

And it might as well not work because if this low-latency and cheap solution exists - it might not be a good fit since it will probably sound robotic and this might not be the type of input the ASR is optimized for

#

Also added complexity, the time to research and implement... Would definitely wait for an enhancement from 11labs, it's possible that it will take less time than implementing this massive workaround

warm rover
#

Yeah, this is what I'm thinking also. Hopefully it'll be available as soon as possible since we really need it, so we won't switch service