The brakdown would ideally be Component / Avg Latency STT / x ms LLM / y ms TTS / z ms

Hi there! 👋 I wasn't able to find information on how to get the exact breakdown in the tabular format you requested with specific millisecond measurements. However, I can help you with several methods to access latency data for your Vapi calls! Here are the ways you can potentially get latency information for your calls: ### 1. Contact Vapi Support with Call IDs According to [Discord discussions](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1336875140990111796), reaching out to Vapi support with specific call IDs can get you in-depth investigation. The support team can check logs and provide specific information regarding your calls, which likely includes latency data. ### 2. Use the "GET Get Call" API Endpoint The [Vapi API](https://docs.vapi.ai/api-reference/phone-numbers/list) includes a

GET Get Call

endpoint (

/api-reference/calls/get

) that might return detailed information about specific calls, potentially including latency data for the different components. ### 3. Capture Call Events with the Server SDK If you're using the server SDK, you can [implement real-time analytics](https://discord.com/channels/1211482211119796234/1211483291191083018/threads/1331285157088333955) by capturing call events. The SDK can trigger events when calls are initiated or completed, allowing you to track metrics including latency information. ### 4. Integrate with Langfuse for Telemetry Monitoring Vapi [integrates with Langfuse](https://docs.vapi.ai/providers/observability/langfuse), an open-source LLM engineering platform, for enhanced telemetry monitoring. This integration allows you to track, analyze, and visualize traces from AI interactions, which could help you monitor latency across components. ### 5. Review Vapi Configuration Settings

Hey Simon, as you already listed out how you want to break down the latency using the Turn latency and other logs available to you for a call. You can calculate it pretty simply by adding up the individual latencies. It will take some of your manual effort going through logs and identifying the right logs snippets.