Caliber lives inside Claude, ChatGPT, Gemini, and Perplexity. It scores your prompt 0 to 100 as you type, tells you what's holding it back, and rewrites it in one click. Then it's the one tool that measures whether that rewrite actually helped, and publishes the number.
Caliber attaches to the composer you already use. No new app, no copy-paste. Score, optimize, and send, right where you type.
A gauge scores your prompt 0 to 100 as you type and flags what's missing, right beside the composer.
The score breaks into what changes an answer: specificity, context, constraints, output format.
Pick a mode: add a role, go detailed, or keep it tight. Accept drops the rewrite into the box.
A blind benchmark: the same tasks, original prompt versus optimized, judged by a stronger model in both orders to cancel position bias. This is the real pilot result, including that most comparisons were inconclusive. No other prompt tool shows you this.
Scoring runs in your browser. Your conversation memory lives in the page and is erased when the tab reloads.
You control your data. Usage stats are anonymous and off-switchable; sharing prompt text is opt-in only.
Never sold, never used to train other models. No exceptions.
Caliber is in a small private beta. Leave your email and we'll send you access as we open more spots.