agent-network-protocol/AgentNetworkProtocol

What it solves

Agent Network Protocol (ANP) addresses the lack of a standardized communication layer for large-scale AI agent networks. It prevents agents from being locked into isolated platforms by providing an open, secure, and machine-readable way for agents from different domains to authenticate, discover, and collaborate without needing to imitate human web browsing.

How it works

ANP is a three-layer architecture built on existing internet infrastructure (HTTP, DNS, TLS):

1. Identity and Encrypted Communication Layer: Uses W3C DID (specifically did:wba) and Web infrastructure to handle authentication and end-to-end encrypted messaging.
2. Application Protocol Layer: Defines how agents describe their capabilities (Agent Description Protocol) and find each other (Agent Discovery Protocol).
3. Domain-Specific Protocols: Implements specialized functions like the AP2 agent payment protocol for agent-to-agent transactions.

It also includes a meta-protocol for semantic negotiation, allowing agents to agree on which interface or security profile to use before communicating.

Who it’s for

Developers building autonomous AI agents that need to interact across different platforms, as well as architects designing decentralized agentic ecosystems.

Highlights

• Standardized Identity: Implements did:wba for cross-platform agent authentication.
• Human-Readable Naming: Uses WNS handles (e.g., alice.example.com) to resolve to DIDs.
• Modular Messaging: A suite of nine interoperable profiles covering everything from basic JSON-RPC 2.0 binding to group messaging and E2EE.
• Agent-to-Agent Payments: Includes the AP2 protocol for secure mandates and receipts between agents.