[untitled]
1 コメント
I was using Perplexity API for synthesis and spent around $10 every few days just based on the volume of research runs my agents were doing. This is a fairly cheap alternative way to do bulk research. The only difference is it can take a little longer. The synthesis layer that offloads tool results to a reasoning model for gating not only improves the results but also saves on tokens for the frontier model-driven main agent. Two branches on the repo; you can synthesize in the cloud (cheap OpenRouter default model) or locally.