[mble_]

Author here. I deliberately didn't include one because I'm waiting for the dust to settle a bit.

There is a lot of activity in the space, from things like TurboPuffer (proprietary), Postgres extensions like VectorChord and pgvector, to established players like Elastic getting in on the action. Not to mention things like Pinecone.

[ako]

Fair point, but would be interesting to see what characteristics make for a best in class ai database, rag, vectors? Or do you foresee ai to replace databases? “ChatGPT, can you remember this data for me?”

[mble_]

Definitely doesn't replace databases, at least LLMs as they currently are. We're going to be stuck with relational algebra for a long time to come, but the interfaces to interact with store might change over time from SQL to more natural language.

> what characteristics make for a best in class ai database

As I said before, I think the space is moving too fast on what is "best in class" - see all the vector indexing work that has happened in the last ~6 months or so. Big, centralised vector stores for some applications will have very different profiles to vector stores on the edge/on device, for example.

As mentioned, I'm a big fan of boring technology, so I'm going to happily wait we have a boring "winner".