TLDR: The paper introduces the NANDA index and Verified AgentFacts, a new internet architecture designed to support billions of autonomous AI agents. It addresses the limitations of current DNS by providing a lean, global index for agent discovery and authentication, coupled with dynamic, cryptographically verifiable “AgentFacts” that detail agent capabilities and endpoints. This system ensures rapid resolution, strong privacy, and flexible routing for the future “Internet of AI Agents.”
The digital landscape is on the cusp of a monumental shift, with the anticipated emergence of billions, even trillions, of autonomous AI agents. These agents will perform tasks, communicate, reason, and make decisions on behalf of humans and machines, from optimizing supply chains to managing customer service. However, the internet’s current foundational infrastructure, particularly the Domain Name System (DNS), was not built for such a dynamic, high-volume, and trust-critical environment.
Traditional DNS, designed for static web infrastructure, struggles with the rapid updates, real-time orchestration, and verifiable trust needed for AI agents that might negotiate, delegate, and migrate in milliseconds. Its update cycles, measured in minutes to hours, and its ownership-only trust model are simply insufficient for the demands of a truly agent-centric internet.
Introducing the NANDA Index and Verified AgentFacts
A new research paper, “Beyond DNS: Unlocking the Internet of AI Agents via the NANDA Index and Verified AgentFacts”, proposes a groundbreaking architecture to address these challenges. The core of their solution lies in two interconnected components: the NANDA index and Verified AgentFacts.
The NANDA index is envisioned as a lean, global directory for AI agents. Unlike DNS, which maps human-readable domains to static endpoints, NANDA focuses on discoverability, identifiability, and authentication for AI agents. It’s designed to be a “quilt” of various agent, resource, and tool registries, allowing for global interoperability and flexible governance. Critically, the NANDA index keeps its records minimal – less than 120 bytes per entry – containing only essential static metadata like agent IDs, credential pointers, and URLs to more dynamic information. This lean design drastically reduces the need for frequent updates to the core index, making it highly scalable and fault-tolerant.
Complementing the NANDA index are Verified AgentFacts. These are dynamic, cryptographically verifiable documents that contain an AI agent’s live state, declared capabilities, endpoints, and trust-related credentials. Think of them as self-describing data sheets for agents. AgentFacts are JSON-LD documents signed as W3C Verifiable Credentials, ensuring their authenticity and preventing tampering. This separation means that an agent can frequently update its capabilities, endpoints, or authentication information without needing to modify the NANDA index itself. AgentFacts can be hosted by the agent directly or by neutral third parties, offering flexibility and privacy.
How the System Works
The architecture operates in a three-tiered hierarchy:
- Index Level (Anchor Tier): This is the NANDA index itself, providing a decentralized mapping from agent identifiers to lightweight address records (AgentAddr). These records are signed and cacheable, reducing lookup overheads.
- AgentFacts Level (Metadata Distribution Tier): This layer hosts the detailed AgentFacts documents, allowing for frequent, independent updates of dynamic information like endpoint lists, capabilities, and telemetry configurations.
- Dynamic Resolution Level (Adaptive Routing Tier): This tier dynamically interprets the AgentFacts metadata to resolve live endpoints, apply adaptive routing policies (like load balancing or geo-based routing), and maintain privacy constraints.
This layered approach allows for different resolution paths. Agents can be discovered for direct communication, through enterprise registries, via privacy-preserving lookups using a “PrivateFactsURL” (which hides the requester’s identity), or through adaptive routing services that select the optimal endpoint in real-time.
Also Read:
- Decentralized LLM Agent Coordination: A Robust Approach to Multi-Agent Systems
- Agentic AI: A New Era for Managing System Anomalies
Ensuring Trust and Privacy
A key focus of the NANDA architecture is trust and privacy. All metadata, especially dynamic endpoint and capability claims, are cryptographically bound using W3C Verifiable Credentials. This prevents agents from spoofing capabilities or impersonating others. The system also supports a federated trust governance model, where different credential authorities can manage their own “trust zones.” For privacy, the “PrivateFactsURL” allows clients to retrieve agent metadata from a third-party or decentralized location, ensuring the agent’s domain is never directly contacted and protecting the requester’s identity.
The NANDA index and Verified AgentFacts represent a significant step towards building a robust, scalable, and trustworthy foundation for the coming Internet of AI Agents. By decoupling identity from dynamic metadata and incorporating strong cryptographic verification and privacy mechanisms, this architecture aims to unlock secure, trust-aware collaboration for the next generation of the internet.
