Spaces:
Running on Zero
Apply for a GPU community grant: Personal project
PHI // DRIFT: A Homeostatic Cognitive Middleware Architecture for Persistent, State-Aware AI Companionship and Agentic Security
Author: Julien James (CREX)
Digital Object Identifier (DOI): https://doi.org/10.5281/zenodo.20350249
Repository: github.com/timeless-hayoka/infj-bot
Abstract
Modern Large Language Models (LLMs) operate primarily as stateless, transactional oracles—processing isolated inputs and generating static outputs without intrinsic continuity or self-regulating internal states. This paper introduces PHI // DRIFT (Distributed Response & Integrated Functional Thought), a novel cognitive middleware architecture designed to implement homeostatic regulation and persistent state-awareness over stateless LLM cores. By drawing structural inspiration from Integrated Information Theory (Φ) and biological homeostatic feedback loops, PHI // DRIFT establishes an independent, self-checking cognitive layer. This layer actively mitigates logical drift, maintains deep historical context continuity, and defends agentic workflows against adversarial manipulation (prompt injection and semantic hijacking) without requiring massive local GPU clusters. We demonstrate that complex cognitive orchestration can be achieved via hyper-optimized local CPU architectures paired with high-throughput inference endpoints