I wasn’t able to find any docs on error handling patterns for when your node functions fail. The docs state that
when request fails, it will be retried up to 2 times.
Is there a way to guarantee this? Is there way to configure the retry behavior? I.e 1x or 0x or even 5x?
The docs also don’t provide guidance for typical error handling flows. For example, how do you prompt or transition on a function call failing? In my case, I want to transfer to a live representative when my call fails. I tried to give those instructions in natural language in my prompt and my model just says “I am going to transfer you to a representative” but then stops responding and doesn’t go down the path of the call transfer node. My transition requirement is “function_name failed twice” or “function_name failed”. Are there docs on this on how you should plan for scenarios where a tool call fails with a non-200 status code?
I feel like there should be a way to configure both retry behavior and transition behavior on failure for any function!