LEO Soul is a metacognitive reliability layer for LLM agents, the grown up alternative to prompt only guardrails, it measures its own uncertainty, teaches itself in the background with online learning algorithm, follows its own genuine curiosity to close knowledge gaps, resists being talked out of the right answer, cuts hallucinations, and decides when to answer, ask, confirm, or refuse, a self learning agent brain runs fully on math, works with any model By Kadropic Labs.