Sovereign: A Unified GPU Inference Substrate (Fractal Memory, Manifold Routing)

Sovereign is a unified GPU inference architecture that fundamentally reimagines AI memory and processing by treating intelligence as a geometric substrate rather than a statistical model, replacing traditional components like KV-caches and fixed context windows with persistent, fractal memory systems. The project, developed solo over 3.5 years, addresses core limitations of modern AI—including statelessness, context window constraints, and massive computational overhead—through a complete ecosystem of mathematical frameworks and GPU-native implementations that enable continuous identity and cross-session memory without requiring model retraining or hyperscale infrastructure.