Universal Memory Protocol: A Probabilistic Framework for Cross-Platform AI Memory and Identity Resolution

Read full text Summarise with VIRAT
Authors: Devansh Verma
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

The rapid proliferation of specialized Artificial Intelligence (AI) agents has exposed a critical flaw in human-computer interaction: contextual amnesia. Current AI systems operate as isolated Markov decision processes, lacking access to a user's historical state across different platforms, devices, and sessions. Furthermore, even within continuous sessions, standard models suffer from severe context decay. This paper introduces the Universal Memory Protocol (UMP), a highly performant, universal memory infrastructure layer that utilizes a novel probabilistic identity resolution engine. Rather than relying on deterministic, walled-garden authentication schemas, UMP links disparate user interactions through information entropy, Gaussian-boosted semantic similarity, and a cascading 3-Layer context retrieval model driven by client-agnostic hardware anchoring. Through rigorous empirical benchmarking, UMP completely eliminates in-chat context decay (boosting retention by 41%), successfully ports high-fidelity memory across completely disjoint applications, and achieves an F1-score of 0.82 for probabilistic identity linking with zero false positives. Operating at sub-4ms infrastructure latencies for approximately $3.00 USD per million requests, our results indicate that deterministic accounts are no longer a prerequisite for persistent, personalized AI consciousness.

Fields of Study & Subjects

Speech and dialogue systems User Authentication and Security Systems AI in Service Interactions

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
30 March 2026
Language
EN
Document type
Preprint