It would be cool if there were some cache (invalidated by hand, potentially distributed across many users) so we could get consistent results while iterating on the later stages of the pipeline.
Do you mean you want responses cached to e.g. a file based on the inputs?
That’s a great idea. Store inputs/outputs in XDG_CACHE_DIR/runprompt.sqlite