I build YemHub, a set of free tools for working out what AI models actually cost to run. Most recent is an agent cost simulator that shows how token spend compounds across a multi-step agent run. https://yemhub.com/agent-sim/