5 comments

[ 3.5 ms ] story [ 20.9 ms ] thread
Great concept! Have had to build some bare bones solutions to handle LLM routing but no reason I needed to roll my own. Planning to give this a go next time I run into that issue.

Are you thinking about integrating smarter cost monitoring and reporting into routing as well?

How does it determine task complexity? Or is this a parameter the user passes in?
Amount of emojis makes me uncomfortable
It's a bit annoying that every AI library implements it's own protocol handlers for every provider - OpenAI, Google, Anthropic.

I'm not talking about the routing part, thats cool, but about how you need to load the API keys in this library.

We need something like Language Server Protocol, where we have a single openai library, a single gemini library, and all libraries use that.