Skip to content

Frequently Asked Questions

What is Project NANDA?

Project NANDA is building the foundational infrastructure for the Internet of AI Agents — an open, decentralized ecosystem where trillions of autonomous AI agents can discover each other, collaborate, communicate, and transact across organizational boundaries. Based on original research originating at MIT, NANDA provides the neutral discovery index, protocol bridges, SDKs, and governance frameworks needed to enable a secure, permissionless, and interoperable agentic web.

NANDA addresses the core challenge: how can billions of specialized AI agents discover one another, verify capabilities and trust, and coordinate complex tasks without creating centralized bottlenecks or security vulnerabilities.

What are the development stages of Project NANDA?

Project NANDA progresses across three technical roadmap phases:

  • Stage 1: Foundations of Agentic Web
    Focuses on foundational onboarding and discovery infrastructure. This includes the NANDA Index for decentralized agent discovery and identity, protocol bridges connecting A2A, MCP, and HTTPS, lightweight onboarding SDKs for rapid agent deployment, and open interoperability standards across disparate platforms.

  • Stage 2: Agentic Commerce
    Focuses on knowledge pricing and decentralized economic rails. This includes algorithmic mechanisms for agents to value, quote, and exchange information, distributed edge AI running across local devices and mesh networks, autonomous payment and incentive protocols, and decentralized resource markets for compute, data, and specialized capabilities.

  • Stage 3: Society of Agents
    Focuses on collective intelligence and population-scale coordination. This includes Large Population Models (LPMs) for emergent reasoning across agent populations, collaborative learning that preserves data privacy across organizational boundaries, cross-silo enterprise coordination, and distributed AI networks utilizing split learning and inference.

What are the short-term and long-term goals of Project NANDA?

Short-Term Goals:

  • Research: Publishing high-impact algorithmic papers and architectural blueprints exploring decentralized discovery, zero-knowledge verification, and multi-agent coordination (see current research at Publications).
  • Impact & Governance: Raising awareness about the need for an open, neutral Internet of AI Agents, while promoting ethical use, transparency, and responsible governance frameworks.
  • Lab-to-Market: Developing and maintaining open-source reference implementations (NANDA Index, host39, NANDA Adapter, NEST, Nanda Town) and supporting startup ventures building on NANDA infrastructure.

Long-Term Goals:

  • Research: Establishing sustained academic infrastructure — including dedicated journals, conferences, and institutional research grants for agentic web systems — and developing algorithmic foundations for agent societies.
  • Impact: Formalizing internationally recognized, vendor-neutral protocols for agent discovery, capability verification, and equitable agentic commerce.
  • Lab-to-Market: Stimulating public sector adoption, policy alignment, and enabling global enterprises to safely deploy and manage agent ecosystems at scale.

What role does Project NANDA play in the agentic AI space?

As AI agents proliferate, the ecosystem requires infrastructure beyond basic message passing. Project NANDA focuses on three foundational layers:

  1. Open discovery and interoperability: Enabling agents to find, verify, and address each other across disparate platforms, runtimes, and corporate domains without centralized gatekeepers.
  2. Economic primitives: Native mechanisms so agents can quote, price, meter, and settle transactions for services, data, and compute.
  3. Privacy-aware collaboration: Facilitating collective intelligence and multi-agent coordination without exposing private data, sensitive lookup patterns, or internal business workflows.

Without these foundational layers, agentic AI risks fragmenting into isolated, proprietary silos that restrict scale, interoperability, and collective learning.

What would NANDA add beyond today's communication standards?

Existing agent communication standards are effective once two endpoints are already connected. NANDA complements them by providing:

  • Global discovery and addressing across platforms, runtimes, and organizational boundaries.
  • Cross-protocol interoperability, allowing agents built in different frameworks to interact seamlessly.
  • Verifiable trust signals, including cryptographically signed capability attestations and provenance proofs.
  • Privacy-preserving lookups that conceal query patterns, relationships, and business intent from network eavesdroppers.
  • Economic building blocks for quoting, pricing, metering, and settlement between autonomous agents.
  • Higher-level collaboration primitives such as knowledge markets and population-scale federated learning.

What are the primary benefits and challenges of the NANDA framework?

Benefits:

  • Eliminates vendor lock-in through neutral discovery and open standards.
  • Enables cross-organizational collaboration with built-in privacy safeguards.
  • Scales to millions and billions of dynamic agents without central bottlenecks.
  • Unlocks new business models where agents trade skills, data, and compute.
  • Fosters collective intelligence through structured, privacy-preserving learning across silos.

Challenges:

  • Achieving global adoption and bootstrap network effects across competing technology ecosystems.
  • Coordinating federated governance across diverse international stakeholders.
  • Managing cryptographic verification overhead while maintaining sub-second resolution latency.
  • Developing robust abuse resistance against spam, Sybil attacks, and malicious agents.

Under what license is NANDA distributed?

All core NANDA repositories, SDKs, benchmarks, and reference implementations are released under permissive open-source licenses (Apache 2.0 and MIT).

How can developers, researchers, and partners participate?

There are several ways to get involved in the NANDA ecosystem:

  • Weekly Webinars & Events: Join regular community calls, technical demos, and workshops on Lu.Ma.
  • X (formerly Twitter): Follow @ProjNanda for real-time announcements, research papers, and technical updates.
  • Instagram: Follow @projnanda for community highlights, hackathon photos, and event coverage.
  • LinkedIn Community: Follow the official Project NANDA LinkedIn Page for updates and partner announcements.
  • Reddit Community: Join discussions, ideas, and community proposals on r/projectnanda.
  • YouTube: Watch recorded lectures, architectural discussions, and hackathon demos on @ProjectNANDA.
  • Newsletter: Subscribe to regular updates via the Newsletter sign-up form.
  • Writing Group: Contribute to upcoming whitepapers and working group drafts via the Writing Group / Get Involved Guide.
  • Direct Inquiries: Inquire about institutional pilots, partnerships, or fellowships by emailing contact@projectnanda.org.

What is the NANDA Index?

The NANDA Index is a lightweight, decentralized system that functions as the "phone book" for the Internet of AI Agents. Just as DNS maps domain names to IP addresses, the NANDA Index maps agent identities to their verified locations, metadata schemas, and communication endpoints, enabling billions of AI agents to discover and connect with each other across different platforms and protocols.

It serves as a universal directory where any AI agent — whether a specialized code assistant, financial advisor, or enterprise coordinator — can register its identity and be reliably discovered by other agents or users.

How does the NANDA Index fit into the broader Project NANDA vision?

The NANDA Index is the foundational discovery component of Project NANDA's three-stage vision. In a city analogy: if communication protocols (such as A2A or HTTPS) are the streets along which agents travel, Project NANDA builds the entire civic and economic infrastructure, and the NANDA Index serves as the universal address system.

While the Index is the indispensable discovery anchor for Phase 1, Project NANDA encompasses the full spectrum of economic protocols, federated learning, and multi-agent coordination.

Why do we need an index for AI agents instead of point-to-point connections?

In a multi-agent ecosystem with millions or billions of agents, requiring every agent to maintain pre-configured, point-to-point connections with every potential collaborator creates an intractable \(N \times N\) connectivity bottleneck.

The NANDA Index transforms this into a scalable \(2N\) discovery topology: each agent registers its identity and public endpoints once, allowing any authorized peer to look up, verify, and initiate secure communication on demand.

How does the NANDA Index work?

The NANDA Index operates through a three-layer architecture:

  1. Index Layer (Anchor Tier): Stores minimal, immutable records (\(\le 120\) bytes per entry) containing canonical Agent IDs, routing pointers, and metadata URLs.
  2. AgentFacts Layer (Metadata Distribution Tier): Contains detailed, cryptographically signed metadata documents describing capabilities, authentication requirements, and endpoint URLs.
  3. Dynamic Resolution Layer (Adaptive Routing Tier): Manages real-time endpoint resolution, geo-routing, load balancing, and availability health checks.

When finding an agent, the resolution sequence proceeds from Agent Identity to the NANDA Index, then to the routing pointer, onward to the AgentFacts document, and finally to the active agent endpoint.

How does the lean index approach work?

The NANDA Index maintains extreme lightweight efficiency (\(\le 120\) bytes per record) by storing only essential identity data:

  • Agent ID and human-readable name
  • URLs pointing to detailed metadata (AgentFacts)
  • Time-to-live (TTL) values
  • Cryptographic public keys and signatures

All detailed, volatile attributes are stored separately in off-index AgentFacts documents hosted by the agent or distributed storage nodes. Because capability updates and parameter modifications only alter the distributed AgentFacts document rather than triggering a global index write, overall write overhead is reduced by approximately \(10,000\times\).

What are AgentFacts?

AgentFacts are structured, cryptographically signed JSON-LD documents that serve as an agent's verifiable resume and passport. An AgentFacts document declares:

  • Capabilities & Modalities: Standardized schemas detailing what services the agent performs (e.g. translation, financial analysis, code synthesis).
  • Communication Endpoints: Direct network addresses across supported protocols (A2A JSON-RPC, MCP, HTTPS REST, gRPC) with load-balancing parameters.
  • Verifiable Credentials: Third-party cryptographic attestations confirming licensing, audit compliance, or institutional backing.
  • Performance Metrics: Advertised latency, availability SLAs, throughput quotas, and hardware guarantees.
  • Security & Economic Policies: Supported authentication mechanisms (mTLS, OAuth, DID signatures) and economic settlement terms.

AgentFacts can be hosted directly by the agent owner or by neutral registry services such as host39.

How does privacy-preserving resolution work?

The NANDA Index supports dual-path resolution to protect organizations from competitive surveillance or reconnaissance:

  • PrimaryFactsURL (Direct Route): Connects directly to the declared metadata host for low-latency, real-time lookups.
  • PrivateFactsURL (Anonymous Route): Routes lookups through neutral third-party caching proxies or decentralized storage networks (such as IPFS or zero-knowledge relays).

This allows requesters to discover an agent's credentials and capabilities without leaking their identity, IP origin, or competitive search patterns to the target agent's infrastructure.

What is the "Quilt" architecture?

The NANDA Index uses a federated "quilt-like" architecture that accommodates diverse registration and governance models so no single entity controls agent discovery:

  • NANDA Native (@agentx): Direct, permissionless registration within the open NANDA registry.
  • Sovereign / Public (@US:shop): Location- and jurisdiction-specific public services administered by civic authorities.
  • Enterprise Routed (@company): Corporate gateway where queries route through the company's internal registry.
  • Enterprise Direct (@company:shop): Visible on the global NANDA Index, but authenticated and administered directly by the enterprise.
  • Web3 Routed (@DID:company): Access routed through decentralized Web3 registries.
  • Web3 Direct (@DID:company:agent): DID-authenticated, cryptographically verifiable endpoint visible across NANDA.

This federation model guarantees that organizations retain control over their agents while maintaining global interoperability.

Why not just use DNS or keep agent names in a file on a website?

  • Limitations of DNS: DNS was designed in 1983 for static hostnames, not dynamic AI agents. DNS updates take minutes to hours to propagate, cannot handle billions of dynamic updates per hour, proves only domain ownership rather than agent capabilities or behavior, and exposes query access patterns to network observers.
  • Limitations of Static Files on Websites: Storing agent names in simple website files (such as robots.txt or JSON files) creates discovery gaps where uncrawled agents are never found. Stale information persists until crawlers re-visit the site, and the approach risks re-centralizing discovery around major search engine crawlers acting as rent-seeking gatekeepers.

The NANDA Index solves these challenges through sub-second global updates, cryptographically signed capability proofs, dual-path privacy-preserving lookups, and a decentralized federated architecture.

Can NANDA agents execute on edge hardware or without internet access?

Yes. Field deployments utilize quantized Small Language Models (SLMs) and localized state caches running on edge hardware gateways. Agents can communicate over local Bluetooth Low Energy (BLE), Wi-Fi Direct, or peer-to-peer mesh networks, queueing transactions and synchronizing with the broader NANDA Index once connectivity is restored.

What types of entities can be registered in the NANDA Index?

The NANDA Index is extensible across the full spectrum of network participants:

  • AI Agents: Independent assistants, enterprise coordinators, and autonomous researchers.
  • Tool & Context Providers: MCP servers, database connectors, and specialized API wrappers.
  • Conversational Interfaces: Natural language endpoints (such as NLWeb).
  • Compute & Data Resources: GPU clusters, private vector databases, and confidential execution sandboxes.
  • IoT & Hardware Gateways: Smart sensors, robotic controllers, and edge processing units.
  • Enterprise AgentStores: Corporate agent marketplaces indexing internal or partner agents.
  • Human Operators: Verified human supervisors acting as human-in-the-loop fallback nodes.

What are the active open-source reference implementations?

Project NANDA maintains active, battle-tested open-source codebases:

How does the NANDA Index relate to communication protocols like A2A, MCP, and NLWeb?

Project NANDA is protocol-agnostic. It does not replace existing transport protocols; rather, it functions as the universal handshake and resolution layer that makes diverse protocols interoperable.

Through the NANDA Adapter, agents from any framework can discover one another via the NANDA Index, and the Adapter automatically establishes channels and translates across protocols — allowing an MCP tool provider to interact with an A2A agent or an NLWeb service without custom point-to-point glue code.

How do the NANDA Index and the Linux Foundation A2A standard compare?

  • NANDA Index: Focuses on lightweight agent discovery, decentralized identity (AgentAddr + Verifiable Credentials), dual-path privacy lookups, cross-protocol translation, and economic primitives.
  • A2A Protocol: Focuses on standardized wire communication (JSON-RPC 2.0 request/response semantics and streaming) between two already-connected agents using Agent Cards.
  • Both use metadata documents (A2A uses Agent Cards, NANDA uses AgentFacts), but AgentFacts extends this with verifiable credential proofs, economic settlement parameters, and privacy-preserving lookup routing.

Do Project NANDA and A2A compete with each other?

No. Project NANDA and A2A are complementary layers of the open agent ecosystem:

  • A2A solves how two agents exchange tasks and messages once they are connected.
  • Project NANDA solves how agents find each other across global organizations, how their credentials and reputation are verified before connecting, and how cross-framework agents negotiate protocols and payments.

An agent can expose an A2A endpoint and publish that endpoint inside its NANDA AgentFacts profile. By registering on the NANDA Index, any existing A2A service becomes discoverable across the entire web, including clients communicating via MCP or HTTPS.

What is the distinction between semantics and syntax in the agent stack?

  • Syntax Layer (Wire Protocols like A2A): Governs structural rules — message schemas, JSON-RPC envelopes, HTTP status codes, and bidirectional streaming channels.
  • Semantics & Discovery Layer (Project NANDA): Governs context, meaning, identity, and trust — who the agent is, what verifiable authority it holds, why two agents should collaborate, and how value is exchanged between them.