Memorivex AI: An Edge Native AI Architecture for Local Memory Deduplication and Zero Knowledge Cryptography

Current conversational artificial intelligence systems rely on cloud-hosted vector databases to manage long term user context. This dependency introduces structural flaws, primarily compromised user privacy, continuous API latency, and "context overlap" a condition where conflicting facts cause the model to output hallucinated data. This paper details the engineering architecture of Memorivex, a proprietary mobile framework developed on React Native. Memorivex functions as a decentralized cognitive extension, eliminating cloud database dependency by running a Zero Shot Deduplication Pipeline directly against a local SQLite instance. The system injects historical memory nodes into the language model's pre inference prompt, forcing the model to identify logical contradictions. The application features an LLM agnostic routing system and an asynchronous Human in the Loop (HitL) execution block to govern database mutations securely. This paper documents the framework's complete technical implementation, including a Zero Knowledge credential vault utilizing PBKDF2 block ciphers, threat model analysis, computational complexity metrics, and an evaluation of system limitations.