GitPetri : A State-Driven Runtime Kernel for Knowledge-Intensive Systems. Context Is Everything

Abstract Large Language Models have fundamentally changed software engineering and knowledge work. Current research largely concentrates on improving model architectures, reasoning capabilities and agent orchestration. This paper explores a different hypothesis. We propose that the primary state of an intelligent system is neither the language model nor its prompt, but its context. From this perspective, a Large Language Model is analogous to a computational engine. It provides general reasoning capability, while its observable behavior is determined by the active context presented to it. This paper introduces GitPetri, a runtime architecture that externalizes context, versions it using Git, activates it through Petri-Net state transitions and projects it into interchangeable Large Language Models. GitPetri is proposed as the runtime kernel of the broader pVoid project.