Metacognitive Uncertainly Estimation and Hallucination Control for AI Agents.
Estimate confidence levels for candidate responses and actions.
Reduce overconfident responses through uncertainty-aware gating.
Includes ECE, Brier Score, entropy metrics, and Bayesian updates.
Adjust behavior based on domain-specific risk profiles.
Determine whether to answer, retrieve evidence, or request review.
Runs without LLM APIs, vector databases, or ML frameworks.