Hermes成本与额度长帖约 3 分钟

五条配置砍掉六成Token

Five Lines Cut Token Cost

要点

  1. 把压缩阈值从默认五成降到0.35,长会话更早瘦身
  2. protect_last_n设为10,旧工具输出不再反复计费
  3. 用便宜辅助模型做摘要,别拿主模型压缩历史
  4. 终端输出上限两万字节、读文件四万字符,不够再分段

原帖开头

🚀5 Config Lines that cut My Hermes Agent Token Usage by 60% The problem 😱 Hermes Agent resends your entire context on every single API call. That means: -Old tool output sitting in history gets billed again and again until it's compressed out -Compression by default waits