We have been running multiple agents on your platform, however in the past two to three days we have noticed increased latency. Please note we have not changed any of the infrastructure parameters such as agent id - b5bcac35-xxxx-xxxx-xxxx-70c9a30eeddf system prompt length the model used - gpt-4o-mini knowledge base attached - via Trieve provider - OpenAi transcriber - deepgram-en-IN-Nova 2 voice - 11 Labs Call ID - 8577dc70-0d65-4fe8-83cd-adc6623cd7eb latency - 8-10 seconds. expected latency - <1 second everything remains unchanged, however the response time has greatly increased which is resulting in poor performance.

Check the status of your third-party services as they may be experiencing issues: * Vapi: https://status.vapi.ai/ * Deepgram: https://status.deepgram.com * OpenAI: No direct status page found * 11 Labs: Check their status page If all services show normal operations, consider adjusting your Deepgram endpointing settings by increasing

transcriber.endpointing

to 300 and

startSpeakingPlan.waitSeconds

to 0.8 to potentially improve response time. Sources: - [Discord Channel](https://discord.com/channels/1211482211119796234/1296852804631265392) - [Discord Channel](https://discord.com/channels/1211482211119796234/1336875140990111796)

Hey, it's because of the smart endpoint plan. Can you make the following changes to your transcriber and start speaking plan so that you never run into this issue again? Let me know how your call goes after making the following changes.

{
  "transcriber": {
    "model": "nova-2",
    "language": "en",
    "provider": "deepgram",
    "endpointing": 300
  },
  "startSpeakingPlan": {
    "waitSeconds": 0.4,
    "smartEndpointingPlan": {
      "provider": "livekit",
      "waitFunction": "20 + 500 * sqrt(x) + 2500 * x^3"
    }
  }
}

where do i update the endpointing "300" for deepgram in the app UI?

“x” can be used as is. It stands for the probability number that the livekit endpointing model returns, and is always somewhere between 0-1. (0 means confidant the user has stopped speaking, 1 means confidant the user is still speaking, etc.)

The Deepgram's endpointing value can only be changed using API. Here is the Curl request for it.

curl --location --request PATCH 'https://api.vapi.ai/assistant/b5bcac35-xxxx-xxxx-xxxx-70c9a30eeddf' \
--header 'Content-Type: application/json' \
--header 'Authorization: Bearer YOUR_API_KEY' \
--data-raw '{
  "transcriber": {
    "model": "nova-2",
    "language": "en",
    "provider": "deepgram",
    "endpointing": 300
  }
}'