### Answer It seems like the behavior you're experiencing with your agent on phone calls compared to the web client might be related to differences in the handling of transcriptions, latency, or audio processing between the two environments. A common issue mentioned in user discussions is that agents might act differently or experience issues when deployed on phone systems like Twilio compared to web-based tests [(1)]() . This could be due to the transcriber's performance varying between environments, especially if the language model or transcriber settings are not optimized for the nuances of phone audio. For instance, certain transcribers may perform well on web platforms but encounter difficulties with the audio quality or format present in phone calls. This can lead to incomplete or inaccurate transcriptions, which would affect the agent's responses. Additionally, specific configurations or features might behave differently when interfacing with third-party services like Twilio. To address such issues, you may need to adjust the transcriber settings, test different transcribers that are fine-tuned for phone audio, or review the call setup on Twilio to ensure it's optimized for your use case. If you're using Deepgram, it's noted that while it performs well in English for phone calls, there might be variations in performance for other languages or specific settings might need adjustment for optimal results. If these adjustments do not solve the issue, it would be worthwhile to inspect the specific errors or behaviors noted in the call logs and compare the configurations between your web and Twilio setups to identify any discrepancies. Unfortunately, without direct access to specific call log entries or error messages, providing a more targeted solution is challenging.