We are encountering an issue with our call routing logic that was previously stable. Starting a few days ago, squad members began to pronounce a "transferring the call" text . Additionally, the assistant's first message did not generate first message and there was an awkward pause at the start of the interaction with new squad member. Could you confirm if there have been any recent changes to the squad logic that might be causing this behavior? Thank you.

@Taras Please ask support for more info on this.

Hey @Taras To help track down this issue, could you share: - The call ID - When exactly this happened (the timestamp) - What response you expected to get - What response you actually got instead This would really help me figure out what went wrong!

- b34284ff-16a8-4156-a69b-8483bb488a0f - 1:48-1:51 - Silent transfer - It says "Thank you for that information. I will now proceed to transfer your call to the appropriate assistant. " when transferring the call to a different squad member

@Taras This issue is most likely related to prompting. Could you please follow this guide on Silent Squad Transfers and let me know the results? https://vapi.notion.site/Silent-Squad-Transfers-14e008c8393f80abbcd6d1a5c10532b8?pvs=4

@Shubham Bajaj we have already seen this guide and tried to do everything as described there, but still every 10-15 calls it can be reproduced, that is, it is not 100% stable

@Taras I understand that the method does not guarantee a 100% success rate; however, it does provide the opportunity to achieve the maximum possible success. We have set the mode to automatic, allowing the language model to determine the appropriate actions with the tools. Additionally, I have taken note of your suggestion regarding the simultaneous execution of content generation and tool usage.

@Shubham Bajaj thanks! Can you add this as a possible bug and add it to the backlog? I saw that not only do we have this problem but the idea proposed above should solve it

@Taras This is not a bug; rather, it reflects the requirements of the users. I will investigate the possibility of providing an option to toggle this feature and will update you once I receive confirmation. In the meantime, I am available to assist you with prompt adjustments to further enhance your squad transfers' silent chances.

@Shubham Bajaj > As mentioned when the model should produce both message.content and message.tool_calls in a single step it's a feature not a bug Yes, it's expected, but since there is no way to configure this "feature", it becomes a bug, because it often says “Transferring your call now”, etc. if we had the flexibility here and could configure that if the openai response has both message.content and message.tool_calls, then “cut off” message.content, then this problem would be solved for us

@Taras I appreciate your perspective regarding the issue you are experiencing. However, in the case of the shared call ID model, content was not generated simultaneously with the tool; rather, these occurred in two separate requests. The content was produced due to the destination assistant's initial message mode, which takes into account the message history. Therefore, there is certainly an opportunity for further experimentation. Please let me know how I may assist you in this regard. Additionally, I would like to inform you that as we advance towards implementing workflows, this matter is expected to be resolved by the end of Q1, ensuring you are adequately supported.

@Shubham Bajaj so basically the source prompt was appropriately(silently) triggered transferTool, and the problem was with the destination (other squad member) prompt?