at Raindrop, we like to treat every prompt as a function: structured inputs, structured outputs every function used to prepare the input for the model lives in the same prompt file, and we have an extension for iterating on the prompt right inside our codebase.

@raindrop_ai have you checked out BAML? does that, but with a bit more tooling (cli testing, type-system based GEPA optimizer built in, type-safe streaming). And it ports right into typescript (and python, and go, etc). PS the git diff thing is sick. Gonna be adding that!

@raindrop_ai nice! overkill for what we need.

@vaibcode @raindrop_ai Can’t wait for the dspy crowd to find this tweet 😂😅

@vaibcode @raindrop_ai lmao

@raindrop_ai collab? :)

this is cool! totally agree on the prompts are functions piece in mirascope we take advantage of Python so automatically capture these functions so you don’t have to have it all in the same function. Simple decorator on an actual function definition takes care of everything else we’re close to having something similar for typescript

@raindrop_ai Yes, our abstraction handles this as well!

@raindrop_ai This is sick I bet you guys have super rich internal benchmarks

@raindrop_ai Raindrop looks great but there’s so many new agent ops platforms it’s hard to tell them apart on a feature by feature basis.

@raindrop_ai i think our next launch will tell the story for us