Cut Hermes token use by choosing the right cron mode

Hermes can use very different amounts of language-model capacity depending on how they are created. A plain job such as `hermes cron create "30m" "check X"` starts a full every time it runs, so every interval consumes tokens and may use more when Hermes calls tools. With `--script`, the script runs first, but its stdout is added to the agent prompt for interpretation, so an still happens.

With `--script --no-agent`, the script itself is the entire job, with no and no . Its stdout is sent unchanged to the chosen Discord, , or Slack destination. Empty output stays silent, while any output becomes a message.

Hermes calls this the classic pattern.

Key points

  • A plain starts a full on every scheduled run and consumes tokens each time.
  • Using `--script` alone still causes an because Hermes interprets the script output.
  • Using `--script --no-agent` runs only the script and avoids language-model usage.
  • Empty stdout sends no alert; non-empty stdout is delivered unchanged to the selected channel.
Read original