Agent-OSI: A Layered Protocol Stack Toward a Decentralized Internet of Agents

TL;DR

This paper introduces Agent-OSI, a six-layer decentralized protocol stack for interoperable, trustable, and pay-per-use agent networking on the Internet, leveraging blockchain for settlement and provenance.

Contribution

It proposes a comprehensive layered architecture for decentralized agent communication, integrating blockchain-based settlement and semantic interoperability without new network protocols.

Findings

Reduces on-chain session costs by ~51% compared to Web3 baseline.

Keeps negotiation and delivery off-chain to improve efficiency.

Blockchain confirmation latency is not the main bottleneck for generative workloads.

Abstract

Large Language Models (LLMs) are accelerating the shift from an Internet of information to an Internet of Agents (IoA), where autonomous entities discover services, negotiate, execute tasks, and exchange value. Yet today's agents are still confined to platform silos and proprietary interfaces, lacking a common stack for interoperability, trust, and pay-per-use settlement. This article proposes \textit{Agent-OSI}, a six-layer reference stack for decentralized agent networking built on top of the existing Internet. Agent-OSI combines secure connectivity and A2A messaging, decentralized identity and authorization, settlement and metering, verifiable execution and provenance, and semantic interoperability for orchestration. In particular, we treat HTTP 402 (Payment Required) as an application-level payment challenge (analogous to HTTP 401 for authentication) that triggers escrow-based settlement and verifiable receipts (instantiated via a blockchain escrow in our prototype), rather than introducing a new network-layer protocol. We implement a prototype and evaluate cost and latency. Results show that keeping negotiation and delivery off-chain while preserving verifiable settlement reduces on-chain session costs by approximately 51\% compared with a standard Web3 baseline in our prototype setting, and that blockchain confirmation latency is often not the dominant factor for generative workloads.