We've gotten a few calls where what I gather is the internal transfer logic gets shared with the caller: Transferring the call to our post completions team right now. Please hold for a moment while I transfer you. Multi 2 loose parallel slash 2 looses, 2 arch, recipient name, functions. Transfer logic, parameters, which number, plus 4, I'm transferring the call to our team right now. Please hold for a moment while I transfer you. Using gpt-4o as the model.

Hi there! I understand you're experiencing an issue where internal transfer logic is being shared with callers when using Vapi with GPT-4o. This sounds like a prompt leakage problem where the system instructions are being verbalized to the caller. Could you provide some additional details to help me better understand the situation? 1. Are you using a specific transfer function or tool in your Vapi implementation? 2. Is this happening consistently or only in certain scenarios? 3. Have you made any recent changes to your prompt templates or function definitions? 4. Do you have any logs or recordings of these calls that show the exact sequence of events? This information will help me identify what might be causing the internal transfer logic to leak into the conversation.

1) Yes using a transferCall logic. 2) In certain scenarios 3) The whole thing is a new-ish feature 4) Yes

Hey! 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 us figure out what went wrong!

Hey Shubham, this has happened a few times: call id: ba12a736-aca1-4d64-83a0-ba3aaab54b0f timestamp: 13:58:08 i expected the assistant to transfer the call, instead it shared with the caller: "I'm transferring the call to right now. Please hold on a moment. Calling the MPowered transfer logic function with plus 4 4 8 4 6 4 8 4 2 4 9 4 ." (i've changed digits here for privacy)

looking into it.