Imagine a future where autonomous AI agents don’t just assist us — they negotiate, orchestrate, and execute decisions across digital and physical realms in milliseconds. Now imagine trying to route, authenticate, and audit these trillions of agents using a system designed for 1980s-era websites. That’s the conundrum the creators of the NANDA index are confronting head-on.

The paper, Beyond DNS: Unlocking the Internet of AI Agents via the NANDA Index and Verified AgentFacts, presents a bold infrastructure vision that goes far beyond anything like DNS, HTTPS, or traditional service registries. Instead, it proposes a lean yet powerful framework for agent discovery, authentication, routing, and governance. The implications? A new kind of internet, tailored for machine-native, privacy-preserving, trust-aware autonomy.

From Domains to Agents: Why DNS Breaks at Scale

The current Domain Name System (DNS) works for static websites, but it breaks under the strain of:

  • Rapidly spawning agents (millions per second)
  • Ephemeral routing needs (blue-green deployments, geo-failover, DDoS protection)
  • Privacy constraints (zero-trust environments)
  • Rich credentialing (beyond “who owns the domain”)

To address this, NANDA introduces a three-tiered system:

Layer Role Analog Key Innovation
Lean Index Maps agent IDs to metadata links DNS Signed, 120-byte records with TTL and URLs
AgentFacts Stores signed capabilities & endpoints Service registry Verifiable JSON-LD claims with VCs
Dynamic Resolution Routes requests based on real-time logic CDN / Load balancer AdaptiveResolver URLs that can geo-balance or detect threats

AgentFacts: A Verified Resume for Every AI Agent

Think of an AgentFacts document as a cryptographically signed service manifest:

  • Endpoint lists (static, rotating, adaptive)
  • Capability declarations (skills, performance scores)
  • Authentication methods and telemetry config
  • Credentials from trusted issuers (e.g. audit trails, certifications)

Unlike Google’s A2A card, AgentFacts supports:

  • TTL-based endpoint agility (5min to 6h rotation)
  • Privacy-preserving access via PrivateFactsURL
  • Cryptographic guarantees with Verifiable Credentials v2
  • Modular hosting on IPFS, CDN, or enterprise-controlled domains

Quilt, Not Monopoly: Embracing Diverse Registries

NANDA is not a centralized gatekeeper. Its architecture supports a quilt model:

  • Some agents register natively on NANDA
  • Others are visible via enterprise stores (e.g., Salesforce, Google, Microsoft)
  • Web3 agents may use decentralized identifiers (DIDs)

This flexibility allows various governance models to coexist. For instance:

Agent Type Name Resolution Path Comment
Civil @agentx Native Public services, NGOs
Enterprise @company:shop Routed via store Controlled access with delegated facts
Web3 @DID:company:agent DID + NANDA Decentralized, auditable agent identity

Privacy and Governance: Not an Afterthought

The authors embed privacy preservation and trust governance directly into routing and metadata resolution:

  • Dual-path resolution (public & private facts URLs)
  • Auditable metadata using W3C Verifiable Credentials
  • Sub-second revocation via short-lived credentials
  • Split-horizon responses (what you get depends on who you are)

This is essential in regulated sectors like finance and healthcare, where discoverability must not imply surveillance.

Design Tradeoffs Made Explicit

Unlike typical infra proposals, this paper explicitly lists the pros, cons, and open issues of each architectural choice. For example:

  • ✅ Index write reduction (10,000× fewer than DNS)
  • ✅ Rich routing via AdaptiveResolvers
  • ⚠️ Added latency (30–60ms) from privacy mechanisms
  • ⚠️ Replay window risks in decentralized updates

This transparency makes it easier for implementers to mix-and-match layers, balancing agility, trust, and performance.

Why This Matters for Cognaptus Readers

The NANDA vision resonates strongly with Cognaptus’s core mission: building automation infrastructures that are modular, secure, and future-ready. Whether deploying autonomous customer agents, internal workflow bots, or decentralized knowledge workers, firms will soon face the same challenges this paper outlines:

  • How to verify the capabilities of agents you didn’t build
  • How to route trust-aware queries across federated teams
  • How to maintain auditability and revocation across jurisdictions

Adopting a NANDA-like model could help Cognaptus clients prepare for a world where human and agent workflows are tightly interwoven — yet demand fine-grained control over identity, permission, and delegation.

🧵 Final Thread

This is not just a naming system. It’s a new connective tissue for the AI-native internet. Whether or not NANDA itself becomes the standard, the ideas it articulates — minimalism, verifiability, privacy, and interoperability — will define the invisible architecture of the agentic future.

Cognaptus: Automate the Present, Incubate the Future.