clean LLM output for conversational voice agent (agentics)
Hello! I'm using the Agent step to build a fully agentic conversational voice agent. I need to clean the original LLM output before it is reproduced by TTS. The problem is that the LLM output contains some formatting characters that are read aloud by the TTS. For example, if the LLM output is bla bla , the TTS reads it as 'asterisk asterisk bla bla asterisk asterisk'. I've added a function tool for cleaning the LLM output, but I'm unable to set the correct instructions to trigger it when needed. Any help? any working examples? Thanks!
