Hello Vapi Support Team, We’ve observed an issue where, instead of performing the call transfer, the assistant starts reading out the tool configuration text — for example: destination: +81xxxxxxx call id: 019a6b1b-4ef5-7ddf-b5ca-665e36e82239 Could you please investigate this behavior? Thank you for your assistance. Best regards, Yasuhiro Nose https://cdn.discordapp.com/attachments/1437492089956073594/1437492409008259154/2025-11-11_2.21.04.png?ex=691370a1&is=69121f21&hm=5814f26694395133ffab7060e5b6422c86082c93fd2b71f9913ba9e5f646dda5&

This issue usually occurs when the assistant is not instructed to call the transfer tool, but instead reads out the configuration. To fix this, ensure your system prompt explicitly tells the assistant to call the transfer tool (using its

function.name

) rather than verbalizing the configuration. For example:

json
{
  "type": "transferCall",
  "function": { "name": "salesTransferAssistant" },
  "destinations": [
    {
      "type": "number",
      "number": "+81xxxxxxx",
      "transferPlan": {
        "mode": "warm-transfer-experimental",
        "transferAssistant": {
          "firstMessage": "Hello, I have a customer on the line. Are you available to take this call?",
          "firstMessageMode": "assistant-speaks-first",
          "maxDurationSeconds": 120
        }
      }
    }
  ]
}

Update your system prompt to instruct the assistant to call the transfer tool, not to read out its configuration. See the [assistant-based warm transfer documentation](https://docs.vapi.ai/calls/assistant-based-warm-transfer) for more details. Source: - [Assistant-based warm transfer](https://docs.vapi.ai/calls/assistant-based-warm-transfer)

Hi Yasu, If you need guidance on setting up call transfers with Vapi, you can use the

transferCall

function to forward calls to predefined numbers or use dynamic routing with an empty

destinations

array. For dynamic routing, you can choose to have an assistant supply a destination or respond with a destination from a server. To configure call transfers for different departments or using different modes such as warm transfer with summary or experimental, you can set up your configurations using the [call forwarding guide](https://docs.vapi.ai/call-forwarding). Additionally, the new

transferCompleteAudioUrl

property in the

TransferPlan

allows you to specify a custom audio file URL to play after a transfer is complete, which is especially useful in the

warm-transfer-experimental

mode. For detailed examples and configuration options, refer to the documentation provided. If you're facing specific issues with the tool, ensure that your destinations and transfer plans are correctly defined, and check the logs for possible errors.

Thank you very much. As shown in the attached screenshot, the tool is configured with the transferCall type. Also, as shown in the attached log, I found that the model output included configuration details such as the “destination.” I’ve checked the call logs, and no errors were found. Also, it seems that this issue doesn’t occur frequently with other calls. I would appreciate any advice you could provide regarding the configuration. https://cdn.discordapp.com/attachments/1437492089956073594/1437946910328750200/2025-11-12_8.22.33.png?ex=691517eb&is=6913c66b&hm=ac1fa6a203c90e4a1142cb2f0bcf508c847ef79363ee8e942eab50e22aa2f15c& https://cdn.discordapp.com/attachments/1437492089956073594/1437946910882140232/2025-11-12_8.21.24.png?ex=691517eb&is=6913c66b&hm=a67943db24a0869e25a6598750b3030eea54dba515a522d154aab6774c570059&

Hi Yasu, Thanks for the update and the detailed screenshots — that’s really helpful. From what you described, it looks like the model may be misinterpreting the tool configuration text as part of the conversation output. Could you please try switching the assistant’s model to a stronger one such as GPT-4o or GPT-4.1? These models handle structured tool calls and JSON outputs much more reliably and should prevent the configuration details from being read aloud. Let us know once you’ve tested this change or if the issue persists afterward — we’ll be happy to dig deeper. Best, Kyle Vapi Support