By Ramesh Raskar and Pradyumna Chari (MIT Media Lab), Luca Muscariello and Vijoy Pandey (Cisco), Samuel Sharaf and Junjie Bu (Google), Karan Bharadwaj (Project NANDA) #
AI agents are increasingly being used by enterprises, small businesses, and individuals. Enterprises are deploying agents for business operations and customer services, while personal agents such as Muse from Meta, dots from OpenAI, Wajo.ai's Fo, and Hello Haven are beginning to help individuals manage tasks and communicate with services.
As agents become more common, they need ways to find and communicate with one another. Enterprises can publish AI Catalogs, configure DNS records, or operate gateways that expose their agents. Small businesses and individuals often lack the infrastructure or technical resources to do the same.
NandaIndex provides a shared resolution service, called the NandaIndex Switchboard, and an open protocol that organizations can also use independently.
The discovery path depends on the destination, regardless of which agent initiates the interaction. An agent contacting an enterprise can use the enterprise's existing discovery infrastructure, such as an AI Catalog, DNS, or a gateway. NandaIndex is not required when the enterprise already publishes a known discovery path. An agent contacting a small business or individual without such infrastructure can use the NandaIndex Switchboard to locate the relevant agent information.
For example, a personal agent contacting a bank can use the bank's published discovery information. When a bank agent needs to contact a customer's personal agent, it may use the NandaIndex Switchboard to locate the agent. Enterprises therefore have a reason to use the shared switchboard even when they maintain their own discovery infrastructure, because they also need to communicate with agents operated by small businesses and individuals. NandaIndex separates the shared resolution service from the open protocol used to locate agents.
The NandaIndex Switchboard is a shared service that maintains records that associate identifiers with pointers to agent information or discovery systems.
The NandaIndex protocol is open and can be implemented by anyone, including enterprises operating their own internal discovery infrastructure. The protocol describes a logical path through three levels.
- Index. The index maps an identifier to a record or pointer indicating where the relevant agent information can be found.
- Agent Facts Card. The card provides or references information about an agent, including its capabilities and available interfaces. Existing formats such as AI Catalog entries and A2A Agent Cards can be used.
- Endpoint. The endpoint identifies the service through which the agent can be contacted or the next discovery system to consult.
The protocol does not require every organization to operate separate services for all three levels. An index record may point directly to an existing catalog or discovery service.
The shared switchboard uses the protocol to resolve identifiers across organizations. Enterprises can use the same protocol internally without routing their requests through the shared switchboard.
Finding and contacting an agent involves several functions. NandaIndex focuses on resolution and works with existing systems that provide the others.
Identity establishes who a person, organization, or agent is. Identity may be based on domains, verified accounts, or other identity systems. NandaIndex uses identifiers but does not independently establish identity.
Resolution maps a known identifier to a pointer indicating where the relevant agent information can be found. The NandaIndex Switchboard provides resolution across participating systems.
Discovery provides information about available agents and their capabilities. Existing mechanisms such as AI Catalog, Agentic Resource Discovery (ARD), and A2A Agent Cards support different parts of discovery.
Search finds agents matching a particular need, such as a delivery service or booking assistant. Search may operate across discovery systems, but it is not a function of the NandaIndex Switchboard.
An application may first search for a suitable agent and obtain its identifier. It can then resolve the identifier, retrieve the relevant agent information, and contact the agent through its published endpoint. Authentication, access control, and consent remain separate from resolution.
A small business may want to make its agent available to customers without operating its own discovery infrastructure. For example, a neighborhood bakery might use an ordering agent hosted by an external provider, while having no AI Catalog or DNS records for agent discovery.
The bakery can register an identifier and a pointer with the NandaIndex Switchboard. Other agents can resolve the identifier, retrieve the relevant agent information, and contact the ordering agent. The bakery can also change its hosting provider while retaining its identifier by updating the corresponding records.
Individuals face a similar problem. A personal agent may be associated with an email account rather than a personally owned domain. Under the proposed registration model, a provider can publish the agent information under a domain it controls and associate the record with a verified user account.
The individual does not need to configure DNS or operate a discovery service. The provider handles publication, while the switchboard makes the agent information available for resolution. Registration still requires verification of the relevant account or domain.
NandaIndex uses existing standards and discovery systems. AI Catalog provides structured descriptions of agents and other AI resources. ARD supports resource discovery and referrals between discovery services. A2A Agent Cards and MCP descriptors provide information about how resources can be accessed.
Enterprises can continue using existing discovery mechanisms directly. The NandaIndex Switchboard can refer requesters to an AI Catalog, agent card, ARD directory, or another supported discovery system when the destination does not have a known discovery path.
The organization or provider responsible for an agent continues to maintain its information and access policies. Once resolution is complete, the requester can communicate with the destination through its supported protocols. The switchboard does not need to relay subsequent messages.
As more agents are deployed, communication between them will involve enterprises, small businesses, and individuals using different platforms and discovery systems.
NandaIndex provides a shared resolution service for agents whose operators cannot maintain their own discovery infrastructure. Its open protocol also allows organizations to implement the same approach within their existing systems.
Enterprises can continue using their own discovery infrastructure while using the shared switchboard to locate agents operated by small businesses and individuals. The common resolution approach allows independently operated systems to communicate without requiring every participant to use the same platform.
NandaIndex remains an evolving open architecture, with further work needed on identity verification, federation, and interoperability between independently operated services.
NandaIndex architecture and implementation: [https://nandaindex.org/](https://nandaindex.org/)
Technical paper: [https://nandaindex.org/paper.pdf](https://nandaindex.org/paper.pdf)
MIT Media Lab publication: https://www-prod.media.mit.edu/publications/a-global-switchboard-for-the-agentic-web-connecting-discovery-islands-across-enterprises-smbs-and-individuals/
Reference implementation: https://github.com/projnanda/nanda-index-v2 "Agents built by small businesses and individuals will soon outnumber enterprise agents, and most of them will never own a discovery endpoint. A combination of AI Catalog, ARD, and the thin resolution layer of NandaIndex will allow every agent to be found."
Sam Sharaf and Junjie Bu, Google
“The agentic web won’t run on one directory. It will be many independently operated registries and platforms, and agents have to be discoverable across all of them. AGNTCY’s Agent Directory and NandaIndex solve complementary halves of that problem, and doing it on open standards is how it scales.”
Vijoy Pandey, SVP/GM, Outshift by Cisco
“From net neutrality to expanding broadband access, I have long believed that the internet works best when it is open, accessible, and trusted. MIT helped build the foundations of the World Wide Web in the 1990s. The research at MIT Media Lab along with many of our member companies over the last few years are now creating a DNS-like system for agent discovery and verifiable identity that can help make the emerging internet of AI agents open, inclusive, and safe.”
Jessica Rosenworcel, Executive Director, MIT Media Lab
“Widespread adoption of AI agents – particularly as they become embedded in new devices and interfaces – will generate additional network traffic and create new connectivity demands. Mobile operators are working to address these challenges and will need interoperable, open solutions that can scale across networks. Just as importantly, trust and security must be built in from the outset, including reliable ways to verify an agent’s identity and permissions, safeguard personal data and protect consumers from fraud or misuse.”
Louis Powell, Director of AI Initiatives, GSMA
“The Neuro SAN framework at Cognizant enables networks of specialized AI agents to collaborate and solve complex enterprise problems. NandaIndex complements this vision by providing an open discovery and resolution layer, allowing agents built on different frameworks and platforms to find and collaborate with one another across organizational boundaries.”
Babak Hodjat, CTO, Cognizant
“At Nutanix, we see enterprise AI evolving toward distributed agents operating across private clouds, public clouds, and edge environments. NandaIndex offers an open approach to agent discovery that enables interoperability independent of the underlying infrastructure.”
Debo Dutta, Head of AI, Nutanix
“At Kyndryl, we have worked with MIT's NANDA ecosystem for 18 months. Our customers run mission-critical systems, so governance and trust come first. NandaIndex is an important step toward an open, interoperable, trusted agentic web, one that lets enterprises bring in public agents without exposing private agents or data.”Ilyas Iyoob, Global Head of Research, Kyndryl
“Merchants of every size should be able to participate in agentic commerce while keeping control of their data, customer relationships, and technology choices. Open, interoperable standards, and approaches like NandaIndex that make small-business agents easier to find across platforms, can lower the barrier for merchants to reach customers through any AI agent. It’s why any BigCommerce store with an active storefront can let AI agents search its catalog, build a cart, and hand off to the merchant’s own checkout through MCP.”
Anil Kamath, Board Member, BigCommerce
“At Hello Haven, we believe personal AI agents should operate in an open ecosystem, not in the walled gardens that characterized the early internet, such as AOL. We are committed to making all Hello Haven agents discoverable through NandaIndex by default, enabling them to connect and collaborate with other personal, business, and enterprise agents across the open agentic web.”
Frank Addante, Founder & Chief Human, Hello Haven
“AI agents need a verifiable identity to earn trust, and they need to be as easy to discover as a domain name, without their owners having to manage complex infrastructure. That's why Inkbox supports NandaIndex: anyone can register their agent and make it discoverable across the open agentic web.”
Ray Liao, Founder and CEO, Inkbox
“From launching the first MCP servers integrated with NandaIndex in April 2025 to supporting MIT courses and students working on the earliest versions of the NANDA platform, ME-US has been part of this journey from the beginning. We believe NandaIndex can play a foundational role in enabling an open agentic web where businesses, developers, and individuals can connect and collaborate across platforms.”
Zafer Sahinoglu, VP of Innovation, Mitsubishi Electric
“Nasiko has been a core contributor to NANDA, helping build the open infrastructure needed for AI agents to operate and collaborate across organizational boundaries. We believe the open agentic web must be supported by strong governance, secure agent runtimes, and verifiable identities, and NandaIndex is an important foundation for enabling agents to be discoverable, accountable, and trusted at scale.”
Rahul Todkar, CEO and co-founder, Nasiko
“At Tata Consultancy Services (TCS), we believe the future of enterprise AI depends on agents collaborating seamlessly across organizations, industries, and platforms. NandaIndex provides an open foundation for agent discovery and interoperability, enabling enterprises to participate in a connected and trusted agentic web.”
Harrick Vin, CTO, TCS
“At HCLTech, we see Agentic AI becoming integral to how enterprises operate, collaborate, and serve customers. As organizations deploy agents across different clouds, platforms, and business ecosystems, open discovery and interoperability will be essential. NandaIndex offers a promising approach to connecting these agents across organizational boundaries while preserving enterprise control, security, and governance.”
Vijay Guntur, CTO, HCLTech
“For the agentic economy to scale, AI agents need cross-platform interoperability. At Hexaware, our collaboration with the open-source Project NANDA reflects that commitment — and NandaIndex is a practical step toward making AI agents accessible, safe, and useful across industries, organizations, and communities.”
Erik Haug, Global Head of AI Partnerships, Hexaware Technologies
“By our research, 99 percent of the web isn't ready for AI agents, and most agents fail before they ever reach the business they're looking for. Discovery is where the agentic web breaks first. We need to reinvent it for agents, not retrofit search-era methods. NandaIndex takes a bold step in that direction, making businesses and individuals discoverable by any agent, without requiring their own infrastructure. It complements open standards like ARD and MCP Apps that we're helping build, and we're excited to work with the NANDA team to bring this vision to life.”
Liad Yosef, Co-founder & CTO, Ora
“Having contributed to the evolution of the internet from IP-based mobile networks to 5G and beyond, I see the emergence of AI agents as another important step in how networks connect people, businesses, and intelligent systems. At Ericsson, we have been working with MIT and NANDA to advance this vision, and NandaIndex provides a promising foundation for open agent discovery, interoperability, and trusted coordination across the emerging agentic web.”
Mallik Tatipamula, CTO, Ericsson Silicon Valley
“Finding an AI agent and trusting it are two different problems, and the agentic web needs both solved. WorkOS is building the identity and authorization layer for agents, while NandaIndex provides an open approach to discovery across platforms and organizations.”
Michael Grinich, Founder & CEO, WorkOS
“The internet succeeded because no single company had to own it for everyone to participate. We are working to bring that same openness to the Internet of AI Agents. NandaIndex provides a foundation for agents to find one another across platforms and organizations, so that individuals, startups, and enterprises can all participate in the agentic economy on equal terms.”
Sharat Chandran and Chris Pease, ProjectNanda
“Agentic commerce will transform how consumers and businesses discover products, negotiate services, and transact through AI agents. For this new economy to scale, buyers’ and sellers’ agents must be able to find and engage with one another across platforms. NandaIndex provides an important building block for enabling open, interoperable agent-to-agent commerce.”
Sandhya Arun, Chief Technology Officer, Wipro Limited
“We're moving toward a world with billions of autonomous AI agents, each needing its own identity, infrastructure, and place in the digital world. Maritime makes it possible to run these agents at scale, while NandaIndex makes them discoverable and reachable. Together, we're building the foundation for an open agentic web.”
Maria Gorskikh, Founder & CEO, Maritime (YC F26)
“We are entering an era in which AI is becoming a new form of productive labor, fundamentally transforming how companies, institutions, and entire economies operate. At Wand AI, we are defining and building this future through autonomous AI workers, a unified operating system for human and artificial intelligence, and the infrastructure required to integrate AI labor into national economies. As billions of AI agents emerge across organizations and countries, open discovery and interoperability become essential. Initiatives such as NandaIndex represent an important step toward connecting these independently governed AI workforces and enabling a truly global agentic economy.”
Rotem Alaluf, Founder & CEO, Wand AI