It uses a custom algorithm and data format but uses a similar deduplication technique to restic. There is bit of info in the docs page about this [1].
Obviously this is needed to make block level deduplication possible, but it also means that for this "end to end encrypted" product someone can still tell whether or not you own a given file. This is probably fine for most people but worth calling out.
EDIT: I should make note of that on the docs page really. It currently gives the impression that I'm just straight up calculating the sha of the content.