What is the use case here? Are these operations so computationally intensive that normal servers/VMs can't handle it?
Turbopuffer is also a terrible name because it makes me think of someone performing oral sex or smoking weed. If that's what you're going for with the name then you have succeeded. It worked for Pk-Zip....
Typically, vector databases store the index in memory, requiring machines with a lot of RAM. There are ways to reduce the memory usage, e.g. with quantization, but more recently disk-first databases (LanceDB) and algorithms (DiskANN) are becoming more popular as they scale better and are cheaper.
This looks great but let's have a thread once the actual thing is there to be discussed. On HN there's no harm in waiting:
https://hn.algolia.com/?dateRange=all&page=0&prefix=false&so...
https://hn.algolia.com/?dateRange=all&page=0&prefix=true&sor...
This is in the interests of the project as well, btw, because if there's a big thread now, then a subsequent thread will probably get downweighted as a follow-up (https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que...) or buried as a dupe.