Abstract

We disclose LOCI ∞, a universal memory middleware system for cross-platform AI memory synchronization. LOCI extracts entity knowledge from conversations on any LLM platform, stores it in a user-owned typed dual-stream profile (informational stream: heat-decaying associative entities; quantifiable stream: exact typed facts with zero-hallucination retrieval), and injects relevant context back into any LLM platform at any point during or between sessions. The system implements hardware-aware storage tiering mapped to entity semantic salience scores (VRAM/RAM/NVMe), graph-neighbor heat propagation via spreading activation, generation-time confidence-gated correction, keystroke-level predictive context prewarm, and a privacy-locked population memory layer using federated learning with differential privacy. A local API proxy (AXON) intercepts LLM API calls, injects LOCI context using a natural-language identity anchoring technique, captures session data for training, and generates synthetic decoy traffic to obscure injection events. The combination of cross-platform middleware, dual-typed streams, confidence-gated supervision, and hardware-tier mapping constitutes a novel system not anticipated by prior art including Mem0, Zep, Letta/MemGPT, LangChain memory, and OpenMemory MCP. This disclosure is published to establish prior art and is dedicated to the public domain to prevent third-party patenting.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.

Download

Share

COinS