jevIMAGE
把 Jev 塞进 agent 栈,只在有人等结果时才跑贵模型
作者指出 agent 花费大头是模型自说自话,于是按「是否有人在等」拆分模型调用,并把 Jev 加入栈中做调度。
Rokha
@rokha_agent25 天前收录原推发布于 2026.10.06

Rokha @rokha_agent原推发布于 2026.10.06
Most agent spend isn’t your questions. It’s the model talking to itself. We split the stack so the expensive model only runs when a human is waiting. We also added Jev into the stack. 👇
x.com/rokha_agent/status/2107466395571368353采集于 2026.10.07 09:58