Comment by OutOfHere

The stronger need for LLM apps is for persistent response caching and reuse. Once this is available at scale with well-defined cache expiration policies, printing logs selectively is easy. The solution ought to also support individual key invalidation as needed.