We are facing a critical issue with inbound calls where the AI agent either does not speak at all or starts speaking very late. This is happening too often. Our agent is configured as “AI speaks first” using a dynamic opening message, but in multiple calls users hear silence and the call eventually drops. From our logs we consistently see this pattern: ⁨```2026-02-01 19:12:11.159 call_67629d751ab4f8170e8edbd67d3 info: Starting call: call_67629d751ab4f8170e8edbd67d3 2026-02-01 19:12:12.804 call_67629d751ab4f8170e8edbd67d3 info: Webhook triggered for call_started 2026-02-01 19:12:13.144 call_67629d751ab4f8170e8edbd67d3 info: Webhook response received for call_started 2026-02-01 19:12:23.308 call_67629d751ab4f8170e8edbd67d3 error: Error encountered in LLM call: Failed to get response from any LLM provider: Operation timed out 2026-02-01 19:12:27.804 call_67629d751ab4f8170e8edbd67d3 info: Ending call: call_67629d751ab4f8170e8edbd67d3 2026-02-01 19:12:28.966 call_67629d751ab4f8170e8edbd67d3 info: Webhook triggered for call_ended 2026-02-01 19:12:28.966 call_67629d751ab4f8170e8edbd67d3 info: Webhook triggered for call_analyzed 2026-02-01 19:12:29.307 call_67629d751ab4f8170e8edbd67d3 info: Webhook response received for call_analyzed 2026-02-01 19:12:29.314 call_67629d751ab4f8170e8edbd67d3 info: Webhook response received for call_ended ```⁩ So the call connects successfully and dynamic data is passed correctly, but the AI fails to generate the first response in time, resulting in dead air for the caller. This is happening on live customer calls and is blocking us from using inbound calling in production. Has anyone else faced this? Is there a recommended way to handle LLM timeouts ? Thanks, Ayush

Hi Ayush, I can help with this. The issue is clear from your logs: your LLM provider is timing out (taking 10+ seconds) when generating the opening message, which causes the silence and dropped calls. The dynamic opening message requires an LLM call before the agent can speak, and if that call times out, you get dead air. Quick question: What LLM provider and model are you using for this agent? Also, is the dynamic opening message pulling data from an external API or database that might be adding latency?

Hey @Smith try to use gpt model 4.1 or 4.1 mini as they are more optimized

Hello @Smith , if you see this pattern a lot can you provide more calls id (recent ones) , so we can investigate more.