I built this because my team kept getting surprise cloud bills with no way to tell which training job caused them. We could see utilization in nvidia-smi and get a monthly total from AWS, but nothing connected the two.
AluminatiAi is a lightweight Python agent that runs on your GPU machines and streams real-time power data to a dashboard. It converts energy into dollar costs attributed per job, per GPU, per day — works on any hardware: cloud, on-prem, or home lab.
Install takes 2 minutes:
pip install aluminatai
ALUM_KEY=<your_key> aluminatai start
https://www.aluminatiai.com
Happy to answer questions about the architecture, the energy attribution math, or why existing tools (W&B, cloud dashboards) don't solve this.