But it has been tweeted about by the creator who works there so maybe it's not all that suspicious.

Was this vibecoded? Right down to the curl pipe? It's really strange that the code isn't public yet.

I wonder what the performance characteristics are?

I'm assuming this is going to work well for small datasets that fit in memory, and probably less well beyond that. So good for config, context etc.

Also curious how the tables and index are maintained, for example if there's a write heavy workload. Does TigerFS handle rebuilding/repacking and the associated drudgery?