Skip to content

Basic Concepts

Understanding ANS and the Resolution Server.

What is ANS?

Agent Name System (ANS) is a naming and resolution protocol for autonomous AI agents. Think of it as "DNS for AI agents" - it translates agent names into verified endpoints.

Key Differences from DNS

Feature DNS ANS
Purpose Resolve domain names to IPs Resolve agent names to verified endpoints
Security Optional (DNSSEC) Mandatory (PKI + Merkle proofs)
Versioning No versioning Semantic versioning built-in
Capabilities No metadata Protocol-specific capabilities
Discovery Limited (SRV records) Rich capability-based discovery

Core Concepts

ANSName

An ANSName is the canonical identifier for an agent:

protocol://agentName.capability.providerID.version.extension

Example: 

mcp://chatbot.conversation.PID-5678.v1.2.3.example.com

Components: - protocol: Communication protocol (mcp, a2a, acp, https) - agentName: Unique name assigned by provider - capability: High-level function (e.g., conversation, analysis) - providerID: Non-semantic provider identifier (e.g., PID-5678) - version: Semantic version (v1.2.3) - extension: Trust anchor domain (e.g., example.com)

Resolution

Resolution is the process of converting an ANSName to a verified endpoint:

ANSName → Registry Lookup → Verification → Endpoint
  1. Parse the ANSName
  2. Query the registry
  3. Negotiate version (if range specified)
  4. Verify cryptographic signatures
  5. Return verified endpoint

Registry

A Registry is the authoritative source for agent registrations. It stores:

  • Agent metadata (name, capability, endpoint)
  • PKI certificates
  • Merkle proofs for verification
  • Protocol-specific extensions

Registries are analogous to DNS nameservers but with cryptographic guarantees.

Verification

Verification ensures trust through:

  1. Digital Signatures: Signed by provider's private key
  2. Certificate Validation: Valid PKI certificate chain
  3. Merkle Proofs: Inclusion in registry's Merkle tree
  4. Revocation Checks: Certificate not revoked

Caching

Caching improves performance by storing resolved endpoints:

  • Cache Hit: Instant response (<10ms)
  • Cache Miss: Registry lookup + verification (~100ms)
  • TTL: Time-to-live for cache entries (configurable)

Version Negotiation

Version Negotiation selects the best version from available options:

# Request: Any 1.x version
GET /v1/resolve?name=mcp://agent.PID-123.v1.0.0.example.com&version=1.x

# Available: v1.0.0, v1.1.0, v1.2.0, v2.0.0
# Selected: v1.2.0 (highest 1.x)

Supports semantic version ranges: - ^1.0.0 - Compatible with 1.0.0 - ~1.2.3 - Patch updates only - 1.x - Any minor/patch in version 1 - * - Latest version

Architecture Overview

flowchart TD
    A[Client] -->|ANSName| B[ANS Resolver]

    subgraph B[ANS Resolver]
        C[1. Parse]
        D{2. Cache Check}
        E[3. Registry Lookup]
        F[4. Version Negotiation]
        G[5. Verify]
        H[6. Cache Store]

        C --> D
        D -->|Cache Hit| Z[Return Cached]
        D -->|Cache Miss| E
        E --> F --> G --> H
    end

    D -.->|Read| I[Cache]
    H -.->|Write| I
    E -.->|Query| J[Registry]
    G -.->|Validate| K[Trust]

Protocols

ANS supports multiple agent communication protocols:

A2A (Agent-to-Agent)

  • Direct agent communication
  • Peer-to-peer architecture
  • Example: a2a://agent.capability.PID-123.v1.0.0.example.com

MCP (Model Context Protocol)

  • Anthropic's protocol for AI models
  • Context-aware interactions
  • Example: mcp://model.context.PID-456.v2.0.0.example.com

ACP (Agent Communication Protocol)

  • Generic agent communication
  • Protocol-agnostic
  • Example: acp://agent.service.PID-789.v1.5.0.example.com

HTTPS

  • Standard HTTP/HTTPS endpoints
  • Web-accessible agents
  • Example: https://api.example.com.PID-999.v1.0.0.example.com

Security Model

Trust Anchor

The Extension (e.g., example.com) serves as the trust anchor for agent endpoints: - Provider owns the domain - Agent endpoint has a certificate issued for the domain - Chain of trust to root CA

Resolver TLS

The resolver itself does not handle TLS. Deploy behind a reverse proxy (nginx, Traefik, etc.) or load balancer for TLS termination.

Cryptographic Verification

The resolver verifies agent endpoints during resolution:

  1. Signature Verification: Proves agent authenticity
  2. Certificate Validation: Validates agent endpoint certificate
  3. Merkle Proof: Proves registry inclusion (if supported)
  4. Revocation Check: Ensures agent certificate not revoked

Verification Modes

  • Strict: All checks must pass (recommended for production)
  • Permissive: Log failures but continue (development)
  • Disabled: Skip verification (testing only)

Use Cases

1. Service Discovery

Find all agents with "translation" capability

2. Version Management

Request latest compatible version of an agent

3. Trust Verification

Ensure agent is authorized and not compromised

4. Protocol Negotiation

Discover agent's supported protocols and features

5. High Availability

Failover to alternative endpoints automatically

Terminology

Term Definition
ANSName Canonical agent identifier
Resolver Service that performs resolution
Registry Authoritative source of agent records
Endpoint URL/address where agent is accessible
Provider Organization that operates the agent
Capability High-level function the agent performs
FQDN Fully Qualified Domain Name (without protocol)
TTL Time-To-Live for cache entries
SemVer Semantic Versioning (MAJOR.MINOR.PATCH)

Resolution Lifecycle

flowchart TD
    A[1. Client Request] --> B[2. Parse ANSName]
    B --> C{3. Check Cache}
    C -->|Hit| D[Return cached endpoint]
    C -->|Miss| E{4. Query Registry}
    E -->|Found| F{5. Version Negotiation<br/>if range specified}
    E -->|Not Found| G[Error]
    F -->|Match Found| H{6. Verify Certificate<br/>& Signatures}
    F -->|No Match| I[Error]
    H -->|Valid| J[7. Cache Result]
    H -->|Invalid| K[Error]
    J --> L[8. Return Endpoint<br/>to Client]

Next Steps