Mirascope

Building with Mirascope has been one of the most fun aspects of this project. Mirascope doesn't make bad assumptions about how you want to build agents. It's a framework that sits an abstraction layer between Langchain and APIs from the major AI labs. For our work, this is perfect. We want fine-grained control over our agents, so that we fully understand all the aspects of the AI systems we develop. If you're building agents and have not checked out Mirascope yet, I highly suggest you check out the docs: https://lnkd.in/gF9rTh6r

can you believe implementing an llm call with a web search tool is actually this easy?? docs: https://lnkd.in/gqvCttJS

we still think `FromCallArgs` is one of the coolest features of mirascope's structured outputs that puts it a cut above the rest docs: https://lnkd.in/gHAz5332

merry christmas! 🎄🎁 here's an ascii christmas tree from gpt-4o

since mirascope prompt templates are just python functions, you can use objects as inputs and template specific attributes we've made this work even for string templates docs: https://lnkd.in/gTTQ_HuB

mirascope makes it easy to log and debug sequences of llm calls through computed fields, which ensure the final output contains all metadata for the entire process docs: https://lnkd.in/ge-fK4ev

have you had a chance to try out the @OpenAIDevs's Realtime API? mirascope has the cleanest interface out there for Python developers docs: https://lnkd.in/gUM6FZwS

if you put a mirascope call inside of a class, you can use `self` to access attributes of the class this makes state management across calls simple docs: https://lnkd.in/gyvGT_9X

structured outputs don't get easier than with mirascope docs: https://lnkd.in/gSWadvqd

did you know you can use PIL images directly with mirascope? this makes using vision models super easy docs: https://lnkd.in/gvD95nMT