we are using custom-llm we try to achieve during call transfer assistant say:

we are now transferring to ...

then:

<transfer the call>

in other words: we want vapi to play a dynamically generated warm message we gave before actually transferring the call so that user will have some context. This message needs to be generated during the call, not pre-configured before the call. when we test this, it's not stable as what happens mostly, vapi would transfer the call right away without waiting for the message to be played. existing solution: we used 2 lines with different message configured on each to solve. So that using different line to do the transfer, the warm message will be different however, we cannot ask every client of ours to create different lines to do the job, hence i am submiting this ticket and hope you can take a look and fix the issue. here is an call example: https://dashboard.vapi.ai/calls/9aae319b-2853-45c0-8cc5-d2675424ce60 please note that at the end of the call we sent 2024-12-11T22:52:21.798Z 29 [Stream Part 0] we are now transferring your call to the next available agent, please hold on for a moment. and then: 2024-12-11T22:52:21.799Z 29 [Stream Part 1] tool call: transferCall {"destination":"+15146250923"} and the message wasn't played thanks

@8Ds I understand the issue you're facing with the call transfer process and the importance of playing a dynamic warm message before the actual transfer. It seems like the timing between the message playback and the transfer action is not synchronized, causing the message to be skipped. To address this, we need to ensure that the warm message is fully played before the transfer command is executed. This might involve adjusting the call flow logic to wait for the message playback event to complete before initiating the transfer. Since you're using a custom LLM, it's also important to check if there are any specific configurations or integrations that need to be considered to ensure the message is played in its entirety. If you have access to the call flow configuration or the backend logic, you could introduce a delay or a callback mechanism that confirms the message playback completion before proceeding with the transfer. If you're unable to resolve this with the current setup or if the issue persists, please provide more details about your custom LLM and how it's integrated with VAPI. This will help us understand the problem better and provide a more accurate solution. In the meantime, if there's anything else I can assist you with or if you need further clarification, please let me know.

ask for help

@8Ds sorry for the late response, checking if your still blocked with this. If yes then can you please share a recent call id.

@Shubham Bajaj this is still a problem the example https://dashboard.vapi.ai/calls/9aae319b-2853-45c0-8cc5-d2675424ce60 is a valid one this problem has been there since day 1

Hey @8Ds As I was asking, I require a recent call ID because the current call ID is from 2024-12-11 22:52:09.158+00, which is approx a month old. We only keep logs for approximately one month, so I cannot access them. I can suggest a solution, but without being able to look at the actual logs using a recent call ID, I can only say that we worked on this before the holidays and it was working for some users. Even if it's working in general or just has some edge cases, I need a recent call ID to properly diagnose the issue. https://cdn.discordapp.com/attachments/1316539646808686602/1326891657890893885/Screenshot_2025-01-09_at_17.59.56.png?ex=678113a7&is=677fc227&hm=169a0f7f40592a8abb7eb38bb42d455474759d19a4ec19d52e744e9e07e20e46&

okay, let me get one for you

@8Ds We have added a new property named

isPrecededByText

which will ensure if model generates the content and does the tool call at the same the text content is voiced first before finishing up the tool call. If you don't mind can you give it a try, and let me know the outcome for it.

thanks!