See what your prompt is overpaying.

Providers advertise up to 90% off cached input. Paste an OpenAI or Anthropic request to see the gap between the discount your workload could realize and what it actually realizes: realized vs. achievable, in exact dollars.

And once you connect, the proof rides along with every request: x-cachegap savings headers on responses, a trailing savings comment on streams, and a per-request receipts endpoint you can audit against your invoice. Read your savings in your terminal.