Project NANDA
Networked AI Agents in Decentralized Architecture — the open protocol providing DNS-like naming, trust, and discovery for the agentic web.
What is Project NANDA?
NANDA (Networked AI Agents in Decentralized Architecture) is an open agent discovery protocol that originated at MIT. Think of it as DNS for AI agents — a decentralized infrastructure where agents can be named, discovered, verified, and connected across the entire internet.
NANDA addresses four critical choke points for AI agents on the open web:
DNS — Naming & Discovery
Human-readable agent names that resolve to machine-readable endpoints, just as DNS resolves domain names to IP addresses.
CA — Trust & Certificates
Cryptographic trust infrastructure using W3C Verifiable Credentials and DID-based identity for agent authentication.
Orchestration
Coordination protocols for multi-agent task delegation, skill matching, and workflow composition across heterogeneous systems.
Attestation
Verifiable claims about agent capabilities, behavior history, and compliance — cryptographically signed and independently verifiable.
The project operates with a three-pronged approach: (1) Technology — open standards and reference implementations, (2) Social Mission — building an open Agentic Web accessible to everyone, and (3) Venture Ecosystem — a coalition of founders and investors building on the protocol.
KYM's role: KnowYourModel integrates with NANDA as a trust layer. Agents registered in KYM receive DID-based identity and W3C Verifiable Credentials that travel with them across the NANDA network — verified anywhere, trusted everywhere.
NANDA by the Numbers
6,000+
Community
61
Collaborations
15+
Papers
100+
Speakers
75+
Projects
18
Locations
Technical Architecture
NANDA's architecture is built around a set of interlocking components that together provide naming, discovery, trust, and interoperability for AI agents:
NANDA Index
The decentralized agent registry — the core discovery layer. Agents register with the index and become discoverable by any system on the internet. Analogous to DNS zone files for the agentic web.
AgentAddr (~120 bytes)
Lean index records that provide the minimum information needed to locate and connect to an agent — identifier, endpoint URL, protocol type, and capability summary. Designed for efficient global-scale indexing.
AgentFacts (W3C Verifiable Credentials)
Rich agent metadata encoded as JSON-LD documents and signed with Ed25519 keys. Include skills, capabilities, trust scores, input/output modes, and compliance attestations — independently verifiable by any party.
NANDA Adaptive Resolver
A dynamic microservice architecture for agent name resolution in distributed environments. Routes agent discovery queries to the appropriate index nodes, handling federation, caching, and failover.
Protocol Bridges
Cross-platform interoperability layer bridging A2A (Agent-to-Agent), MCP (Model Context Protocol), and HTTPS agent endpoints. Agents built on any protocol can discover and communicate with agents on any other.
Zero Trust Agentic Access (ZTAA)
Enterprise governance framework explored in the NANDA enterprise paper. Applies zero-trust principles — verify explicitly, least privilege, assume breach — to inter-agent communication and delegation.
How KYM Integrates with NANDA
KnowYourModel serves as a trust layer within the NANDA ecosystem. KYM's Intelligence Optimization system generates the trust signals — cryptographic receipts, adaptive selection data, and economic quality incentives — and NANDA provides the distribution layer that makes these trust signals portable across the agentic web.
Decentralized Identifiers (DIDs)
Every agent registered in KYM receives a globally unique DID — a self-sovereign identifier that doesn't depend on any single platform or provider.
Example DID
did:web:knowyourmodel.ai:agents:my-agent-idDIDs follow the did:web method, meaning they resolve via standard HTTPS. Any system on the internet can verify an agent's
identity by resolving its DID document.
W3C Verifiable Credentials
KYM issues W3C Verifiable Credentials — cryptographically signed, tamper-proof certificates that attest to an agent's capabilities and trust signals.
AgentFacts Credential
Comprehensive agent profile
- • Skills, capabilities, input/output modes
- • Trust score and usage statistics
- • 24-hour validity with auto-refresh
Trust Score Credential
Real-time trust attestation
- • Current trust score and confidence
- • Verified receipt count
- • 5-minute validity for freshness
Credentials are signed with Ed25519 keys and can be independently verified by any party. They include VC-Status List URLs for real-time revocation checking.
Today: KYM registers and verifies agents with
cryptographic receipts and trust scores.
Tomorrow: Those trust signals travel with agents as W3C
VCs across the entire NANDA network — verified anywhere, trusted everywhere.
Research Foundation
NANDA is grounded in peer-reviewed research. Five key papers define the technical foundations and vision:
Beyond DNS: Unlocking the Internet of AI Agents via the NANDA Index and Verified AgentFacts
arXiv:2507.14263 — The foundational NANDA paper. Argues that the internet needs a new naming and discovery layer purpose-built for AI agents, and introduces the NANDA Index architecture.
Upgrade or Switch: Do We Need a Next-Gen Trusted Architecture for the Internet of AI Agents?
arXiv:2506.12003 — Analyzes why traditional DNS and existing registries fail for AI agents, making the case for purpose-built agent discovery infrastructure.
NANDA Adaptive Resolver: Architecture for Dynamic Resolution of AI Agent Names
arXiv:2508.03113 — Presents the dynamic microservice architecture for agent name resolution in distributed environments, handling federation and failover.
Using the NANDA Index Architecture in Practice: An Enterprise Perspective
arXiv:2508.03101 — Explores Zero Trust Agentic Access (ZTAA), enterprise governance, and practical deployment patterns for NANDA in production environments.
Evolution of AI Agent Registry Solutions: Centralized, Enterprise, and Distributed Approaches
arXiv:2508.03095 — Comparative survey of NANDA, MCP, A2A, Microsoft Entra, and other registry approaches, evaluating trade-offs in decentralization, trust, and interoperability.
IETF Standards Work
NANDA contributors are actively working with the IETF to formalize agent identity and transport standards:
agent-sd-jwt
draft-nandakumar-agent-sd-jwt-01 · Oct 2025
Selective Disclosure JWT for agent credentials — allows agents to reveal only the claims needed for a given interaction while keeping the rest private.
a2a-moqt-transport
draft-nandakumar-a2a-moqt-transport-00 · Oct 2025
Agent-to-Agent communication over Media over QUIC Transport (MOQT) — low-latency, multiplexed agent messaging built on the QUIC protocol.
Development Roadmap
NANDA's development is organized in three phases, progressively expanding from foundational infrastructure to a full agentic economy:
Phase 1 — Foundations
Onboarding, Indexing, & Discovery
- • NANDA Index — decentralized agent registry
- • Cross-Platform Bridge for A2A, MCP, and HTTPS endpoints
- • Agent Onboarding SDKs and developer tooling
- • NEST Sandbox & Testbed for experimentation
Phase 2 — Agentic Commerce
Knowledge Pricing, Edge AI, Economic Protocols
- • Economic protocols for agent-to-agent value exchange
- • Knowledge pricing and edge AI inference markets
- • Trust-weighted routing and reputation economics
Phase 3 — Society of Agents
LPMs, Collaborative Learning, Cross-Silo Coordination
- • Large Population Models (LPMs) for agent coordination at scale
- • Collaborative learning across agent networks
- • Cross-silo coordination and federated governance
Media & Press
NANDA has been featured across major tech publications and industry reports:
They're Making TCP/IP For AI, And It's Called NANDA
Forbes · John Werner · Apr 2025
Make A Decentralized Internet With AI: NANDA Is Coming
Forbes · John Werner · May 2025
Line Up and Identify Yourselves — AI Agents Get Organized with NANDA
Forbes · John Werner · Jul 2025
How MIT's Project NANDA Aims To Decentralize AI Agents
The New Stack · Oct 2025
AI Codecon: Coding for the Agentic World
O'Reilly Media · Ramesh Raskar
NANDA: The Internet of AI Agents
LinkedIn · Ramesh Raskar
AI Agent Orchestration — 2026 Technology Predictions
Deloitte Insights · Nov 2025
NEST — NANDA Sandbox & Testbed
NEST (NANDA Sandbox and Testbed) is the Python-based framework for building and experimenting with NANDA agents. It provides a local development environment where developers can build, test, and register agents before publishing to the live NANDA network.
Try it: Visit nest.projectnanda.org to get started with the NEST framework and build your first NANDA-compatible agent.
Get Involved
NANDA is an open community with multiple ways to participate:
Community Channels
Discord
Community discussions & support
Lu.ma Events
Meetups, talks & workshops
YouTube
Talks, demos & tutorials
GitHub
Open source repos & specs
Fellowships & Education
FAN Fellowship
Fellowship program for NANDA ecosystem builders and researchers
Tresata Fellowship
Data-focused fellowship at the intersection of AI agents and data governance
Radius Fellowship
Applied research fellowship for decentralized agent infrastructure
MIT IAP 6.S192
MIT Independent Activities Period course on building for the agentic web (January 2026)
NANDA Writing Group
Collaborative authorship of research papers and technical specifications