Skip to main content
Community Specification v0.2.0

The standard for trusted AI agents

The agent passport for identity, permissions, lifecycle, and compliance — in one auditable, machine-readable document.

Read the SpecExplore Governance
agent.adl.json
{
  "adl_spec": "0.1.0",
  "name": "Claims Processing Agent",
  "version": "2.1.0",
  "lifecycle": {
    "status": "active",
    "sunset_date": "2027-06-01"
  },
  "permissions": {
    "network": {
      "allowed_hosts": ["claims-api.internal"]
    },
    "data_access": {
      "allowed_sources": ["claims-db"],
      "restricted_fields": ["ssn", "dob"]
    }
  },
  "profiles": [
    "urn:adl:profile:governance:1.0",
    "urn:adl:profile:financial:1.0"
  ]
}

Trust, verify, deploy

Everything a CISO needs to say yes to AI agents

Governance & Compliance

Map agents to NIST 800-53, SOC 2, ISO 27001, and EU AI Act controls. Built-in compliance profiles for enterprise governance requirements.

Agent Identity

Cryptographic identity with W3C DIDs, attestation support, and deny-by-default permission boundaries for every agent.

Lifecycle Management

Track agents from draft to retirement with sunset dates, successor agents, and auditable status transitions.

Interoperable

Generate A2A Agent Cards, MCP configurations, and integrate with OpenAPI specifications. The definition layer for the agent ecosystem.

Extensible Profiles

Domain-specific profiles for governance, portfolio management, and more. Extend ADL without modifying the core spec.

Vendor-Neutral

Portable agent definitions work across providers and runtimes. Structured for submission to Linux Foundation, IETF, and ISO/IEC.

The governance layer above MCP and A2A

MCP describes how agents connect to tools. A2A describes how agents communicate. ADL describes who the agent is — identity, permissions, lifecycle, and compliance in one portable document.

  • Deny-by-default permissions with auditable boundaries
  • Compliance mappings for NIST 800-53, SOC 2, ISO 27001, EU AI Act
  • Lifecycle management with sunset dates and successors
  • Generates A2A Agent Cards and MCP configurations
Learn More

The Agent Bill of Materials

You wouldn't deploy software without an SBOM. ADL is the ABOM for AI agents — a complete, verifiable record of what an agent is, what it can do, and who authorized it.

ConcernA2A Agent CardsAgent SpecAGNTCYADL
Agent IdentityPartial—DIDs + VCsDIDs + attestation
Permissions ModelPartial—PartialDeny-by-default
Governance & Compliance——PartialNIST, SOC 2, EU AI Act
Lifecycle ManagementPartial—PartialStatus + sunset dates
Agent RelationshipsPartialFlowsPartialPortfolio profile

A2A Agent Cards (Google/LF) · Agent Spec (Oracle) · AGNTCY (Cisco/Outshift)

Built for Interoperability

ADL integrates with existing standards and protocols, enabling seamless agent deployment across diverse ecosystems.

A2A Protocol

Generate Agent Cards for agent-to-agent communication

JSON Schema

Full schema validation for all ADL documents

OpenAPI

Reference OpenAPI specs for tool definitions

MCP

Export Model Context Protocol configurations

See It In Action

One agent discovers another by reading its passport. No hardcoded integrations, no manual configuration — just a standard document that machines can read and trust.

1

Discover

An agent queries a domain and finds available agents — no prior knowledge needed.

2

Read the Passport

It fetches the ADL passport and learns the agent’s tools, permissions, and data sensitivity.

3

Evaluate Trust

Is the data public? Are tools read-only? Is auth required? The passport carries the answers.

4

Communicate

Trust established. The agents exchange tasks using capabilities discovered from the passport.

Try the Interactive Demo

Ready to trust your agents?

Define identity, permissions, and compliance for every agent in your organization. Read the specification or explore the governance profile.

Read the SpecificationView on GitHub