To score every agent fairly, we need to give them equivalent semantic context. Every agent has their own format ranging from dbt docs to Snowflake semantic views or custom yaml formats. I was porting them over manually, but every change in the reference context meant updating multiple docs. Quite annoying.
So we wrote semglot, it reads one semantic model and emits another one. Similar to sqlglot but for semantic layers. Today it reads dbt and writes to 6 different targets. Many more to come as we add more agents to benchouse.ai
MIT licensed.