What should I do if the Hermes Agent token consumption is high?

The consumption of Hermes Agent tokens is suddenly high, so don't blame the model for being expensive. Recently, this problem has been very concentrated in the community, and the common triggers are: too many tool calls, too long contexts, hybrid memory pulling information, and no cheap model routing. When investigating, it is necessary to first locate which type is burning.

Fastest investigation sequence

1. Use /usage to see the approximate consumption trend first.
2. Execute /compress for long conversations first, don't pile up context indefinitely.
3. If you turn on Honcho hybrid memory, temporarily cut the local memory for comparison.
4. When there are many simple questions and answers, consider turning on smart_model_routing.
5. Fix the tool if it fails repeatedly, and don't let the agent keep trying again.

The cost issue should be taken apart: model price is part of it, and failed retries and long contexts are the most easily ignored invisible bills.

In a word: first look at /usage and context, then see Honcho and the tool to try again, and finally change the model.

Official open source address: https://github.com/NousResearch/hermes-agent; Official document entry: https://hermes-agent.nousresearch.com/.