Also wondering about scalability. Saw your demo in haystack live which was awesome btw! The demo where you have 28 million records for book titles was pretty fast but a little bit slower (200ms Vs 20ms) than the 2-3 million records for recipes. For our use case we’re looking to move digitized textbook data, we have over 5000 textbooks. These translate to over 250 million records in our db (a record for each paragraph) about 150GB of data. Do you guys have any metrics on how performant typesense will be on such a large dataset ?

In my case, about 3M documents uses 1.5GB ram. Its performance is not problem, fast enough. I think the problem is your ram disk space. Even with simple calculation, 120GB RAM is needed.

Yes, the trade-off is speed vs memory. While we have certain items on our backlog to decrease memory consumption, ultimately you cannot search instantly on large number of records without adequate memory. One thing that can help is removing commonly occurring stop words from the dataset -- this should reduce memory consumption with negligible impact on search experience.