Agent Coupling Protocol
The trust and verification layer — identity, attestation, and coordination primitives that allow autonomous agents to verify each other and coordinate without human intermediaries.
Agent Coupling Protocol: A Trust and Verification Layer for Autonomous Agent Coordination
Foundation whitepaper for the ACP specification
We present the Agent Coupling Protocol (ACP), a decentralized trust and verification layer enabling autonomous AI agents to establish identity, attest capabilities, and coordinate without requiring human intermediaries. ACP defines a minimal cryptographic primitive set — Verified Agent Identifiers (VAIDs), attestation receipts, and capability manifests — that allows agent pairs to negotiate trust boundaries and execution contracts in constant time with respect to fleet size.
Formal Model of Agent Coupling: Verified Coordination Under Byzantine Conditions
Mathematical foundations and security proofs
This paper presents the formal model underpinning the Agent Coupling Protocol. We develop a process calculus for agent interaction — the Agent Interaction Calculus (AIC) — and prove that ACP achieves verified coordination under Byzantine conditions with O(log n) attestation overhead. We derive the necessary and sufficient conditions for coordination liveness and prove termination bounds for nested delegation chains.
ACP Reference Implementation: Design Notes and Benchmarks
Engineering notes on the open-source ACP reference library
We describe the design of the open-source reference implementation of the Agent Coupling Protocol. The implementation provides a minimal, auditable codebase for teams integrating ACP into agent orchestration frameworks. We discuss key engineering decisions, performance characteristics under various agent fleet configurations, and report benchmark results from trials with LangChain, AutoGen, and CrewAI adapters.