Hi,

Vec1 is at version 0.7 now:

https://sqlite.org/vec1

This is not a 1.0 release yet, but it's solid enough to stop calilng it "preview" or anything like that.

Latest version adds a "RabitQ" quantizer (vector compressor). This seems to work and has the advantage of not requiring training, but it doesn't really stack up to a proper trained quantizer like Product Quantization (PQ) or Optimized Product Quantization (OPQ) in terms of performance.

Vec1 should at this point be in-the-ballpark competitive with most ANN vector-search implementations for static datasets that can be trained on, or for data for which a representative sample is available to train on. It should also compare well to implementations of brute-force search.

Focus now shifts to adding a mode that does not require training at all - a mode that accepts slower writes in exchange for making them online and the index always available and searchable.

Any comments or questions are most welcome here!

Dan.