teolex2020/AuraSDK

Implements a five-layer cognitive engine (Belief→Concept→Causal→Policy) that derives structured abstractions and policy hints directly from stored records without requiring an LLM, enabling bounded self-adaptation and explainability. Features salience weighting, contradiction governance, experience capture with anti-hallucination guards, and ChaCha20 encryption—all operating offline with ~270× faster recall than vector-based alternatives. Exposes advisory surfaces for operator correction, plasticity risk scoring, and provenance chains, integrating as a local sidecar runtime for any frozen model via Python SDK.