I built an open-source FinOps layer for GenAI APIs real-time USD cost per OTel span, multi-tenant isolation, zero prompt logging
python
dev.to
I opened my LLM invoice and had no idea where the money went. So I built LumenAI. Last month I opened my cloud bill and saw $4,847.23 in LLM API costs. For a single week. I had no idea which tenant caused the spike. No idea which agent ran 11,000 times. No idea whether claude-opus-4-6 at $15/M tokens was firing in places where claude-haiku-4-5 at $0.25/M would have worked just as well. My logs said "LLM call completed". That was all. Standard observability tools were not built for