UNaIVERSE: A Peer-To-Peer Network For Human-AI Agents

We present UNaIVERSE (https://unaiverse.io/), a decentralized platform designed to implement a peer-to-peer model of the Web, based on human-AI agent communities. We detail the platform’s architecture, design principles, communication protocols, and demonstrate its capabilities through multiple use cases that showcase distributed problem-solving and human-AI interaction. UNaIVERSE is aimed at addressing emerging limitations of the current Web, social networks, and Artificial Intelligence (AI) infrastructures. While these technologies have expanded global communication and knowledge exchange, they also raise concerns regarding privacy, concentration of control, and escalating energy demands. The coexistence of humans and AI further remains uncertain and largely unregulated, exposing risks that extend beyond technical boundaries. UNaIVERSE is designed around a shift from a Web of “document-like” resources offered via a client-server model to a network of agents (human and artificial) participating in a peer-to-peer network. Agents run on different types of devices, including energy-efficient edge devices, thereby avoiding dynamics that depend on centralized intermediaries and consequently improving privacy preservation. The platform introduces “Worlds”, autonomous communities of agents organized around shared goals, topics, or intents. Within these Worlds, agents learn, reason, and plan both individually and collectively through a peer-to-peer communication protocol, while preserving the integrative information-sharing properties of contemporary social platforms. We discuss the relationship to related work on agent-oriented intelligent-systems approaches, and illustrate through multiple use cases how the platform enables new forms of distributed problem solving and human-AI interaction, offering a new perspective for research activities.