Fallback on AI Failure

[lancerichards]

We were working with a customer on their AI Agent.
It was all working perfectly and then stopped working and just dumping the call.

Upon investigation, it seemed they had hit the context limit within gpt-realtime-1.5 and that was the root cause.
But it got me thinking, rather than dumping a call, it would be great to see a fallback method being able to be defined so that you're never dumping a call.

Has this been thought of or added to a roadmap?

Best
Lance

 

[nikolascx]

Hi Lance,

Yes, this is a valid point and it is already on our radar.

Just to clarify, the call being dropped is not really related to the context limit. In this case, it is more likely that OpenAI could not be reached or the realtime connection could not continue, so the AI Agent had no fallback path available.

We agree that the better behavior would be to fail safely instead of simply ending the call.

We do plan to add a fallback option for these cases, so that if the AI service cannot be reached or the AI call cannot continue, the call can be routed to a defined destination such as an extension, queue, or Digital Receptionist.

No ETA yet, but this is something we definitely want to improve. Great feedback!!