Hi team, I am using the vapi.start method and passing assistantOptions in it. The assistant successfully creates a ServiceNow ticket and receives the response, but it does not verbalize the ticket number. Instead, it stops after saying, "ticket created successfully, and the ticket number is". Here’s the response we receive from the server: { "results": [ { "toolCallId": "call_fFsby95jHyXm0tPuWG51oeIr", "result": "ticket created successfully ticket number : INC0029065" } ] } According to the documentation, this response seems correct. Could you help us understand why the assistant is not verbalizing the ticket number? Looking forward to your support.

Hey! To help track down this issue, could you share: - The call ID This would really help us figure out what went wrong!

Hi Shubham this is a call id

dc9e5b6d-5b70-46ea-9701-2944c2d682ab

where i see the ticket number in the response but the agent did not respond anything. Let me know if you need anything else

here are two key issues occurring: - **Tool Response Format Issue**: The response from the tool is a double-stringified JSON, which can be confusing for the model to parse correctly. - **Assistant Not Processing the Response**: After receiving the tool result at timestamp 1741165704534, the assistant never generated a follow-up response to communicate the ticket number back to the user. (less likely) Solution You need to make two changes: - **Format your server response properly**: Ensure your webhook returns a clean JSON response without nested stringification. - **Update your tool configuration**: Add an explicit message for the successful completion case:. (IMPORTANT)

thanks Shubham, this worked but there are some small issues like 1) It is spelling ticket number INC0001 as "Ink-0- 0-0-0-1" I want that it must say I, N, C seperately and not as 'Ink'. I adjusted the system prompt but still it says the same way 2) and also it says the ticket number one time only the second time if we ask it to repeat it doesn't says the same. You can refer this call id "f6aeb338-1497-44a9-92dc-8430870cb0d1"

Hey man, it appears that that is just the way your voice model pronounces things like this. There's two things we can do to take care of this: 1. In your system prompt for that call I don't see explicit instructions for the way it should output text relating to invoices (remember you're prompting the LLM not the voice model). Something along the lines of "When reading invoice numbers, add dashes between each letter/number" and then give it the example you have there. 2. This issue is likely caused by the prompt, specifically in the memory restriction section "I will not use any prior memory for context-free responses." This can be ambiguous and cause the LLM to not say the invoice number. This should solve both issues.