API proxy that optimizes prompts before they hit the LLM — 48% avg token savings with 3-layer injection defense. Change one line of code.