A software company reduced Hermes agent token costs by configuring background tasks instead of switching models.

• Token costs are driven by system prompts, memory files, skill headers, MCP tool schemas, and conversation history.
• Disable or prune unused skills to reduce per-message token usage.
• Turn off auto-memory and skill evolution if not needed to avoid background token consumption.

how to reproduce · 1. Review your Hermes configuration and identify all active skills. 2. Disable skills that are not essential for your workflows. 3. Turn off auto-memory scanning and skill evolution if they are not required. 4. Monitor token usage via OpenRouter or similar to verify reductions.