Paying twice for the same answer
A documentation assistant answers the same handful of questions all day — how do I reset my password, what are your rate limits — and every single one is a fresh model call with fresh tokens billed and two seconds of latency.
The answers are identical. You are paying, repeatedly, to regenerate text you already have.
Place what should answer before the model is called at all.
Components — tap one, then tap a slot on the diagram
!Forty percent of the model bill is the same handful of questions, asked again and again.
Outside every boundary: Model (billed per token; FAILED: billed twice), User question, an empty slot for the before the model Connections: User question calls before the model (step 1) before the model calls Model — only on a miss (step 2)
Modelbilled per tokenbilled twice
User question