A Hermes agent cost lesson: stop runaway usage early
Hermes and OpenClaw agents were connected to Telegram and appeared to work well at first, including tool use and automated reasoning. By the next morning, the agents no longer replied on Telegram because the usage quota had been exceeded.
The API bill rose to about six times the expected cost. A likely cause was a task loop or repeated reading of the same files.
Reading raw logs by hand was described as too painful to be useful. A proposed fix was to place a proxy between the app and the provider so prompts could be compressed, with several cache layers including an L1 exact-match cache for repeated requests.
Key points
- can burn through usage if they loop overnight.
- The Telegram failure came from an exceeded quota.
- The API bill reached about six times the expected cost.
- Repeated tasks or repeated file reads are warning signs to monitor.
- A proxy and cache can reduce repeated prompt costs, but limits and alerts are still needed.