It’s indeed a generative patch, without using ANY VCO at all.
The sounds are generated by 2 self-fed VCFs.
One feeds the high pass in and outputs the low pass.
The other one does the opposite.
They are “cross-faded”, therefore I called this patch “X”.
As usual, overlaying some time-delayed instances (i.e.: 30 secs or more per layer) enrichens the overall experience – try to match (or cleverly mismatch) the rhythms, if any.