Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.

​ To learn more, see the sources I considered: [[1] Custom LLM Overview](https://docs.retellai.com/integrate-llm/overview) [[2] Discord message from @colinfortuner on 2024-04-16](https://discord.com/channels/1186424376300679209/1186462578310910112/1229936804761833552) [[3] New LLM Model Options, Modular Pricing, And More](https://www.retellai.com/changelog/new-llm-model-options-modular-pricing-and-more) ​

### Answer Based on the available sources, I found some relevant information about handling custom LLM interactions with Retell AI. When using custom LLM integration, the documentation mentions that your server needs to check for

interaction_type

in the JSON messages from Retell [(1) ]() . If the

interaction_type

is

response_required

, you need to send a response . Additionally, Retell sends messages with

interaction_type

as

update_only

when the AI is speaking, where you don't need to respond but can get the transcript from the JSON body . For handling user interruptions during response generation, a community discussion suggests that if you're in the middle of generating a response and the user interrupts, you can drop the current response being generated and generate a new response, though these implementations can be quite tricky and require careful thinking [(2) ]() . The documentation also mentions new ping pong events and reconnection mechanisms in the LLM websocket that track server roundtrip latency [(3) ]() . However, I don't have specific documentation about detecting whether a message is the final update for a turn or detailed configuration options to prevent multiple

response_required

messages in the same turn. For this specific issue, I recommend checking the [LLM WebSocket API documentation]() for detailed event handling specifications, or reaching out to Retell support for guidance on managing multiple response requests within a single turn.