Hello. I've created an dynamic assistant using make.com and Vapi using the exact same prompt. When using the assistant directly from VAPI dashboard (linking the assistant to the phone number) it has the expected behavior, but when I use make.com to retrieve the assistant (Exactly same prompt) the behavior is quite different, most of the time it gets stuck in a step of the customer data collection (asking the customer name), o just makes up all the requested customer information itself and finishes the call (does not hang up, just sais goodbye) it also doe not stops talking when customer interrupts him. Quite confusing. Any recommendatios, guidance? #1211483291191083018

hey please share the call id with timestamp.

These are two of the las calls, both done today: - This is the call ID for the one answered by the dynamic Assistant (made up information such as customer name, and Address) : 9c3c7a00-67f4-45a6-b49d-fb95917c7a4a --> 2024-07-16T18:38:12.355Z This is supposedly the information the customer provided (Customer never mentioned name and address) First Name: John Phone Number: XXXXXXXXX Customer Address: 123 Main St, Springfield, IL Call time: 2024-07-16T18:38:12.355Z Requested appointment date and time: Tomorrow Service type: Appointment - and this for the one answered by the static assistant - Lina - (worked just as required): 34f303b0-0d30-481d-86f6-2872f6323f81 --> 2024-07-16T18:43:54...

Hey try using gpt-4o, gpt-3.5 is sometimes uncertain.

thanks a lot for the answer Shubham, but is there any explanation for this behavior, i did run several tests today and the result is the same: when using a static assistant, built directly on Vapi dashboard it worked just fine, but when i use a dynamic one, it is really messy and unreliable: it does not follow the prompt most of the time, makes up a lot of information out of nothing. What can be happening if the parameters are exactly the same? I'm considering using Vapi's static assistants as long as application does not require a dynamic one

share the call id and please write it as above you have done, it will help me debug this.

Thanks again for answers!, the issue is pretty much the same I share at the beggining (timestamps of second post), so, those would work for any debugging purposes. I already answer your email. Thanks again!

with gpt-4o as well?

4o works just fine