Arcstone Continuity Core: An Air-Gapped, Multimodal Local RAG Reference Architecture for Zero-Egress Autonomous Workstations

Le résumé fourni par la source

Arcstone Computational Spine (ACS) — Academic Whitepaper Suite (Paper 2) Modern cloud-based Retrieval-Augmented Generation (RAG) paradigms introduce severe data leakage hazards, vendor lock-in, and unpredictable latency bounds for privacy-critical enterprise workflows. This paper presents Arcstone Continuity Core, a hardware-accelerated, fully air-gapped local AI substrate. The architecture unifies local vector embeddings (nomic-embed-text), deterministic local language generation (llama3.1), and direct multimodal clipboard ingestion via vision models (llava) over an isolated loopback interface. We detail the operational mechanics of real-time image payload serialization via Base64 memory buffers, local state recovery invariants, and a modular application interface (Streamlit). The platform achieves absolute data egress elimination (Cₒₚₛ = 0) while preserving real-time user-driven context ingestion on commodity hardware. Linked to Primary Master Anchor (10.5281/zenodo.22665852) and the arcstone-continuity-core open-source repository. Core Mathematical & System Invariants:• Zero-Egress Bound: Data_Egress_Sensitive = 0 (Loopback Only: http://localhost:11434 / http://localhost:8501)• Fixed Precision Control: FIXED_PRECISION_DIGITS = 8• Payload Serialization: Payload_base64 = Base64Encode(Bytes_PNG)• Cₒₚₛ = 0 (Zero Operational Drag)• Data_Egress_Sensitive = 0 (Zero Sensitive Egress)

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.