The semantic layer

So I know that semantic layers are, like, the feature of the moment right now, and they are really powerful ways to provide that governed data layer that everyone can be using as a single source of truth. That's really cool. The problem with them is that they drive up enormous cost. They are ultimately a SQL generating engine. Someone asks a query that SQL gets driven to the database, and that SQL is generally relatively poorly optimized, and it's just driving more compute into your data warehouse, which is quite expensive. And so when we looked at what we should do for count, when we built count metrics, we actually realized that that isn't the model you need. The model you need is to have your the ability at least to have, your semantic layer, running in memory in DuckDV, which is an incredibly fast in memory analytical engine so that you don't have to hit your data warehouse with all the extra cost to have blistering fast governed metrics. Now Count lets you do both. You can choose, but it's amazing how much money we're saving our customers when they are running their semantic layer in, Inductee b on our servers, and they are therefore querying their data warehouse about eighty percent less.