A few quick questions on how the input/output processors, specifically the
SystemPromptScrubber
SystemPromptScrubber
: - Do input processors add latency to the time to first token streamed from the agent, or is it processed in parallel? It's honestly a non-starter for us if it blocks. - Do output processors run in parallel as the agent is streaming? - Is there any way to avoid using a model in an output processor and just rely on regex? - Is there a simple way to send a default response when we flag an output? We could catch the error and create a message ourselves but wondering if there is a built-in way to do this?