I want to stipulate not sending previous user input to the model, and not sending the main text above the second floor except for the summary. What should I do if I can’t choose separately?
Setting the summary regex to depth 2 will prevent sending the main text above the second floor.
For blocking previous user input like this, perhaps you could consider tagging the user input content, hiding it with regex, and then setting the minimum depth (number of effective rounds). Maybe it could work? I haven’t tried it.
When AI responds, it acts as a message with a role, but it only takes effect for replies after activation, not before. Using regex from the start doesn’t affect it.