rare-sapphire•2d ago

GPT Realtime model failing to respond back to user with tool call response

When using the GPT realtime model connected to tools, I consistently found that the agent would execute a tool successfully and receive a response, but would just "hang" after responding with some variation of "Let me look that up for you". See attached image for transcript. I'm new to Vapi, so I'm wondering if this is a known bug? Or is there something extra I should be doing?

3 Replies

Vapi Support Bot•2d ago

The most common cause for your issue is an incorrect tool response format. Make sure your tool returns a JSON response like this (with no line breaks in the result):

{
  "results": [
    {
      "toolCallId": "call_123",
      "result": "No busy times found. Free time available from 2025-12-04T00:00:00.000Z to 2025-12-04T23:59:59.000Z."
    }
  ]
}

{
  "results": [
    {
      "toolCallId": "call_123",
      "result": "No busy times found. Free time available from 2025-12-04T00:00:00.000Z to 2025-12-04T23:59:59.000Z."
    }
  ]
}

Double-check that you return HTTP 200 and use a single-line string for the result. This should resolve the "hang" after tool execution. Source: - Custom Tools Troubleshooting

⤵️ Mention @Vapi Support Bot for follow-ups 🐥

rare-sapphireOP•2d ago

I made sure the response format looks exactly like that. oh looks like someone else also just posted the same issue today: https://discord.com/channels/1211482211119796234/1445727355753594931

other-emerald•2d ago

I’ve seen this same issue with the GPT Realtime model—tools return successfully, but the agent hangs right after. It usually comes from how the assistant message is sent after the tool result. I can help you pinpoint exactly where the flow is breaking. Quick question: are you sending a final assistant message after the tool call, or letting the model auto-continue? @ripes

GPT Realtime model failing to respond back to user with tool call response

Did you find this page helpful?