boundary

howdy folks -- do y'all have any best practices on using BAML for codemode, RLMs, programmatic tool calling, or related concepts?

i know there's a turing-complete compiler rewrite in the works, but I'm trying to figure out whether it would fit into the above... 

...like, could we find ourselves asking LLMs to generate BAML code snippets and then interpreting the code?

generically yes, but is that a BAML specific feature?

why I ask -- i was curious whether BAML has any plans to have a more integrated experience for "llm generates code which calls tools in a sandbox"

currently, the experience is: ask llm to generate code, and then eval it in your own sandbox.

we are thinking about how to make eval quite good! specifically in the realm of type-safety

i.e. assert eval returns a specific type of value BEFORE you run it

I figured, especially after reading https://beps.boundaryml.com/beps/19

Thank you! Exited to see what’s coming 🙂