This paper introduces a modular cognitive architecture for distributed collective intelligence via peer-to-peer coordination, federated memory, and semantic knowledge integration. It operationalizes a unified model of distributed intelligence by integrating semantic negotiation, context-sensitive reasoning, and modular memory federation across decentralized agents. Virtual Cognitive Agents (VCAs) utilize dynamic memory graphs and semantic reasoning capabilities to interact through a shared semantic layer supporting faceted classification, contextual tagging, and inter-agent negotiation. The architecture addresses epistemic fragmentation of existing AI systems by preserving traceability, adapting across conceptual domains, and sustaining semantic coherence at scale. It emphasizes explainability, ethical modularity, and interoperability through privacy-aware, metadata-rich protocols. By supporting participatory sensemaking and distributed reasoning, the architecture enables agents—human and artificial—to co-create, refine, and align knowledge across domains. By modeling general-purpose cognition as a modular and distributed process, the architecture facilitates the co-emergence of shared knowledge within multi-agent epistemic environments—advancing general-purpose cognitive infrastructures and contributing to the evolution of Artificial General Intelligence (AGI).