I was going to say, while I have not been utilizing BAML heavily, I did some poking around on the model side with needle3 and it feels like Jev ≅ needle3 + BAML taking the perspective of a web app developer/ux, I want to "just sideload a small LLM helper" into the SPA. (i.e. in cyberchef I could say "write me a recipe for splitting these lines, base64 decode, then comma separate by whitespace" and the sidecar "just does it" without everyone needing to host their own inference backends)