I am running a pretty significant KB based on documents. There is no way to get a decent checksum on them - so, when I do changes my only approach viable is deleting them all (ony be one, slow and really slow on the website) and then reinitializing them. This would allow a fast delete and reseed. Alternatively, please expose a decent (SHA256) checksum and management via API - then I could write a sync program.

@witty-football-93730 for visibility

Interressting Idea. It's in the talkings to expose the Documents API

It would be nice. I have to say - given that this is imho the biggest issue in the whole AI thing, I would also love a task or a setting hat can outsource the whole thing to a separate system. YOu try nice, but that is likethe most trivial approach to a KB that is possible. There are various additional steps one COULD do (that are not exactly trivial to do and run) to make the whole thing better for a lot of non-trivial problems that the chunking approach will neve rhandle.