Declarative AI Architecture - Knowledge Artifacts as System Logic for Generative Systems

Generative AI systems have achieved remarkable progress in language understanding and reasoning, yet their architecture remains fundamentally model-centric. System behavior is typically determined by a combination of model parameters, prompts, retrieved documents and application-level logic, resulting in fragmented and implicit representations of knowledge and control. This paper introduces Declarative AI Architecture, a new architectural paradigm in which system behavior is defined through structured knowledge artifacts rather than embedded in model parameters or procedural code. In this approach, generative models function as semantic reasoning engines that interpret explicit knowledge structures representing operational system logic. Unlike retrieval-augmented generation, where external documents serve as contextual information, knowledge artifacts in Declarative AI Architecture act as explicit carriers of system logic. They define rules, constraints and domain structures that govern how generative reasoning is applied within the system. The architecture establishes a clear separation between generative reasoning and knowledge definition, enabling knowledge to be represented as a modular, inspectable and governable component of the system. This shift allows generative systems to evolve through modifications of structured knowledge artifacts rather than through repeated model training. The paper outlines the architectural principles, system structure and implications of this approach, including its impact on system design, knowledge governance and organizational adaptability. Declarative AI Architecture thus represents a transition from model-centric to knowledge-centric AI systems, providing a foundation for controllable, transparent and reconfigurable generative AI.