Frugon

Frugon reads your model calls on your own machine and shows where the spend is actually going. The follow-up question it exists to answer is which of those calls a cheaper model could have handled just as well, so the bill comes down by substitution rather than by cutting usage. It works across different LLMs and keeps the interface plain. Developers, researchers, and companies watching an AI budget are the readers it assumes. Nothing is uploaded: it is free, open source, and runs locally.