All it does is the work of quickly deciding which tool(s) might be relevant using embeddings instead of an LLM - but yeah should be easy enough to recreate!
2 Likes
Ah, yeah so Iām suggesting something slightly different, Iām suggesting: still allow the LLM to decide but pass it a smaller set to choose from to keep within a reasonable maximum.
Iām also not going to be branching to different code, the process remains the same, Im just filtering the function set. I donāt need to ārouteā.
So my proposal is not so much a ārouterā but a āsmart function filterā.
1 Like