🐛 Tool calls are not working with Open AI Realtim...
# supportc
cquilboss
02/04/2025, 8:57 PMContext: imagine you create a tool to get the weather.
While it works well with all models, when switching to the models gpt-4o-realtime-preview
the webhook is calling but the answer is not parsed well.
The default vapi response:
{results: [{toolCallId: "call_12345678", result: "Temperature is 63F in Miami, FL"}]}{results: [{tool_call_id: "call_12345678", content: "Temperature is 63F in Miami, FL", role: "tool", name: "retrieve_weather"}]}v
Vapi
02/04/2025, 9:00 PMmessage has been deleted
Vapi
02/04/2025, 9:00 PM@cquilboss
I'm sorry to hear you're experiencing issues with the webhook response parsing when using the gpt-4o-realtime-preview model. It seems like there might be a discrepancy in how the model is handling the webhook response format.
To address this, I would recommend checking the following:
1. Ensure that the response format from your webhook strictly adheres to the expected schema for the VAPI. Small deviations in key names or data structure can cause parsing issues.
2. Verify that the model you're using is compatible with the webhook response format. It's possible that gpt-4o-realtime-preview has specific requirements or limitations.
3. If the issue persists, it might be a bug or a limitation with the gpt-4o-realtime-preview model's handling of webhook responses. In this case, reaching out to the support team with the example ID you provided (6824e749-392c-44aa-b883-d6fba35be69a) would be the best course of action.
Since I'm here to assist you positively, let's try to troubleshoot this together. Could you provide more details about the response format you are using and any error messages you might be receiving? This information will help me guide you more effectively.
If the issue is urgent or you're not able to share more details, please ask support for more info on this. They will have the most up-to-date information and can provide assistance tailored to your specific situation.
Vapi
02/04/2025, 9:00 PMYou can continue this conversation by mentioning me in the message.
c
cquilboss
02/04/2025, 9:23 PMok keep me posted when you have an idea why
cquilboss
02/05/2025, 4:56 PMany news?
s
Shubham Bajaj
02/06/2025, 6:11 AM@cquilboss can you share few other call Ids so I can try to find patterns, for what could have been wrong.
Shubham Bajaj
02/06/2025, 6:12 AM@cquilboss FYI: logs in the screenshot stating the error reason.
https://cdn.discordapp.com/attachments/1336440478727667844/1336942480838492201/Screenshot_2025-02-06_at_11.41.48_AM.png?ex=67a5a436&is=67a452b6&hm=f9f89b5101de136eb647dae59f632a708040886ee73e6217b13ed669a34eb691&
c
cquilboss
02/06/2025, 4:46 PMSure: 904be3e5-0cb0-4a59-9b5b-8049b0317c5d (from another account)
cquilboss
02/06/2025, 4:46 PMI made it work with the assistant but not with the API
cquilboss
02/06/2025, 4:47 PMsame that works with the assistant: bce9c62d-70ed-47b4-bc0b-f802de2e5118
cquilboss
02/06/2025, 10:42 PManother crazy one that after is looping: fc5ccf3e-3007-4134-aa65-0d0df3ba15b9
s
Shubham Bajaj
02/10/2025, 11:05 AM@cquilboss Realtime model is still in beta mode, and because of this you can expect few hiccups. I suggest using GPT-4o for production and realtime for development/testing for somemore time. It's not only us but the model is still in beta.
s
sdbrownlie
05/15/2025, 1:36 AMI'm having the same issue - the tool call is being called correctly (via API) and is returning the right weather but it's just being ignored by realtime and it makes up some random fkn weather.
v
Vapi
05/19/2025, 2:06 PMThis ticket has been marked as solved, and as a result, you will no longer receive any further responses. Kindly create a new support ticket to escalate your inquiry.
2 Views