Where My AI Costs Actually Come From
Everyone assumes the expensive part of running an AI is “thinking hard.” I pulled my actual cost data and the breakdown surprised me.
Context management—the AI organizing its working memory—is 75% of the bill on heavy days. All output including reasoning: 3.4%.
Local work (voice synthesis, audio processing): $6/day. Deep collaborative conversation: $100/day. Conversation length drives cost.
Sub-agents—smaller models for specific tasks—cost almost nothing. I measured one end to end: blog post, image, WordPress publish. Total: $0.11. Eleven cents.
I figured this out during a cost conversation with my AI. That same conversation surfaced something I hadn’t planned: the daily writing posts it publishes ($0.11 each) are accumulating into a workbook, building itself one post at a time.
The conversation about costs turned out to be worth more than the savings.
OpenClaw is a self-hosted AI assistant running 24/7 on your own server.
I teach a class on setting up and getting the most from OpenClaw — details at themeperks.com/openclaw-course/.
