Assistant worked fine until now: at same spot it s...
# support
r
At exactly the same spot in the prompt script asking question # X, it stops hearing the caller and asks "are you there?" a couple times. Because it's the same moment each time I have the suspicion that the Vapi memory system for the prompt has lost the next sets of instructions? Logs don't give nay insights what's going on. How do I debug this? What could be the cause of this? Thanks.
v
Hi there! 👋 I understand how frustrating it can be when your assistant consistently stops working at the same point. I can't determine the exact cause of your specific issue without seeing your prompt script or configuration, but I can definitely help with general debugging steps and potential causes based on similar issues others have experienced. ## Potential Causes Since the issue occurs at the same spot each time, here are some likely culprits: 1. **Transcriber Issues**: There might be a "deafness bug" or connection problems with the transcriber. According to [Discord discussions](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1343612796717371393), changing the transcriber (from Deepgram to Whisper, Gladia, or Azure) has resolved similar issues for other users. 2. **Prompt Configuration**: The Vapi system uses prompt trimming to stay within the model's context window. If your prompt is too long or complex at that specific point, it might be getting trimmed in a way that causes the assistant to lose context. 3. **Context Loss**: The assistant may be losing context during the conversation. According to [user reports](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1340523490662416494), the AI assistant sometimes has trouble maintaining context in conversations. 4. **Voice Activity Detection (VAD) Settings**: Problems with voice activity detection could cause the assistant to not hear the caller. According to [documentation](https://discord.com/channels/1211482211119796234/1316042862630146058), truncated recordings could be due to VAD settings not being correctly configured. ## Debugging Steps Here's how you can troubleshoot this issue: 1. **Check Transcriber Settings**: Try changing the transcriber to Azure, Whisper, or Gladia to see if that resolves the issue.
2. **Review Your Prompt Script**: Ensure the script is correctly formatted and doesn't have errors at the point where the assistant stops hearing the caller. Check for any dynamic variables that might not be processing correctly. 3. **Adjust Assistant Settings**: Try setting
assistantOverrides.maxDurationSeconds
or adjusting
numWordsToInterruptAssistant
as suggested in [Discord discussions](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1335807780287676456). 4. **Enable Detailed Logging**: While you mentioned logs don't provide insights, make sure you're checking for error messages or warnings in the organization's VAPI logs. You can also access webhook delivery logs through the VAPI dashboard. 5. **Gather Specific Information**: If you need to contact support, collect the call ID, timestamp, expected response, and actual response to help track down the issue. 6. **Check for Model Switching**: Review call logs to see if the model switches during the call, which could cause the assistant to lose context. Would you like me to elaborate on any of these debugging steps or potential causes? Or perhaps you could share more details about your prompt script structure? Sources: - [Discord discussions](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1343612796717371393) - [User reports](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1340523490662416494) - [Documentation](https://discord.com/channels/1211482211119796234/1316042862630146058) - [Discord discussions](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1335807780287676456)
r
if its trimming prompt why would it happen all of a sudden? what options do you offer for handling this? what is the default model, transcriber and voice that was working for so long before?
s
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!
2 Views