Abstract
This publication discloses five novel architectural mechanisms for persistent AI memory systems: (1) a versioned fact chain architecture for exact-fact stores that preserves full temporal history with provenance timestamps, enabling temporal queries unavailable in flat vector stores; (2) a source-trust provenance field that weights facts by declaration origin and prevents model-inferred data from incorrectly overriding user-stated facts; (3) an asymmetric confidence gate for conflict resolution that uses source trust to apply differential correction thresholds; (4) a hardware-tier gradient search escalation algorithm that expands retrieval scope from fastest-tier (VRAM) to slowest-tier (remote VPS) when data is not found at the current tier; (5) a per-person-pair dual emotional dataset with Quality-of-Life Index and Relationship Index gating that forecasts relationship consequences before executing interpersonal actions. All five claims were demonstrated in a production LOCI deployment on 2026-07-08.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Sanyal, Arko and Roza, "LOCI Persistent AI Memory Engine: Versioned Exact-Fact Chains, Source-Trust Provenance Weighting, Hardware-Tier Gradient Search, and Interpersonal Quality-of-Life Gating", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/10843