This repository contains the draft of formal ontology for AI Agent interoperability, security, and communication, developed as part of the W3C Agent Semantic Communication Community Group work. It provides a machine-readable, logically consistent framework for defining agents, their capabilities, and the rules governing their interactions, built upon core W3C standards like Decentralized Identifiers (DIDs) and Verifiable Credentials (VCs).
This CG defines the semantic meaning of agent interactions, not the transport, execution, or reasoning mechanisms used to process them.
Our vision is to create a universal standard for safe and effective semantic communication among AI agents. This ontology serves as the foundational layer for a future where agents can securely delegate tasks, exchange value, and collaborate across different platforms and ecosystems.
Key Goals:
- Interoperability: Define a shared vocabulary that allows agents from different developers to understand each other.
- Protocol-Agnostic: Focus on the semantic payload of communication, ensuring the ontology can be used over any network protocol (HTTP, WebSockets, etc.) and is independent of specific agent implementations (e.g., LLMs).
- Security: Anchor digital trust in physical reality by enabling agent identities and actions to be bound to hardware roots of trust (e.g., TEEs like ARM TrustZone) through attestation, moving beyond purely digital security guarantees.
- Verifiability: Create a complete, auditable chain of responsibility from intent to execution, providing the foundation for automated compliance, online arbitration, and system self-regulation.
- Extensibility: Provide a modular core that can be extended to support diverse applications and domains.
The core Agent Semantic Communication Ontology, including all TTL, JSON-LD, and SHACL files, is now managed in a dedicated repository to better separate the formal standard from community governance and documentation.
➡️ Agent Ontology Core Repository
This repository contains all the normative artifacts of the standard.
This repository now focuses on the high-level specifications, use cases, and community governance for the standard. The formal ontology modules, which are developed in the core repository, are structured as follows:
graph TD
subgraph "Ontology Modules"
Core
Agent
Capability
Delegation
SecurityBinding["Security Binding"]
ExecutionContext["Execution Context"]
Intent
Ledger
Payment
Identity
AgentProfile["Agent Profile"]
end
Agent -- depends on --> Core
Capability -- depends on --> Core
Delegation -- depends on --> Core
Intent -- depends on --> Core
SecurityBinding -- depends on --> Agent
AgentProfile -- depends on --> Agent
ExecutionContext -- depends on --> Intent
To foster broader community engagement and provide a dedicated space for practical applications, all use cases and examples for the Agent Semantic Communication Ontology have been moved to a separate repository:
This dedicated repository allows for independent evolution of practical examples, making it easier for developers to contribute and explore real-world implementations of the ontology.
The behavior and design of this ontology are guided by several key specification documents.
- Minimal Threat Model (
specs/minimal-threat-model.md): This is a critical document that outlines the security threats the ontology is designed to mitigate. It is essential reading for understanding the "why" behind many of the security-related components. - Core Ontology Concepts (
specs/core-ontology.md): Describes the foundational classes and properties.
We welcome contributions from the community! Please read our CONTRIBUTING.md file to learn how you can get involved, whether it's by improving the ontology, adding examples, or enhancing the documentation.
This repository contains the Agent Interface Description Language (AgentIDL) and related tools for code generation and SDK development.