Why do we need a global switchboard to locate agents on the agentic web?
Vision
The agentic web needs a common, reliable, and interoperable way to locate an AI agent given its identifier. Agent discovery is the automated process by which an external system resolves a known agent identifier to the information needed to communicate with that agent, including its network endpoint, supported protocols, and security requirements. Discovery should also enable verification of the authenticity and integrity of this information. Discovery is not semantic search. Search asks which agent can perform a task; discovery asks how to reach a particular agent.
1. Why do we need to enable interoperability without requiring uniformity for an open agentic web?
Enterprises, small businesses, and personal agents are increasingly operating across different providers, but no single registry can be expected to contain them all. A global switchboard can connect these systems without requiring everyone to use the same infrastructure. Without common resolution, each discovery system would need separate integrations or prior knowledge of how to reach agents in other systems.
The ecosystem is already fragmenting: AI Catalogs, DNS-based discovery, platform registries, and enterprise gateways provide different ways to publish and locate agents. Each addresses different requirements, and the agentic web should preserve this diversity.
DNS provides a useful analogy. A domain name can remain stable even when the underlying service moves to a different hosting provider or its IP address changes. DNS resolves the domain name to records that help clients locate the service. However, unlike DNS, the global agent switchboard does not define or administer a single global namespace, assign identifiers, or guarantee identifier uniqueness across namespaces. Instead, it provides a common resolution mechanism across existing, independently managed identification and discovery systems.
The idea of a global switchboard and federated resolution is inspired by research on NandaIndex. NandaIndex is an open resolution architecture designed to serve as a global switchboard. Given a resource identifier, it directs the requester to the appropriate discovery system or published agent information. There is no change in the existing agent discovery among enterprise to enterprise systems. Existing systems can retain control over their resources, and direct discovery continues to work wherever it is available without the global switchboard in the critical path.
2. Why should every agent have the opportunity to be discoverable?
The internet allows individuals, small businesses, and large enterprises to be discovered and reached without depending on a single platform or provider. The agentic web should offer the same openness. As personal and small-business agents become more common, everyone should have the opportunity to make their agents discoverable, regardless of where they are hosted or which technology provider they use. Discoverability should not be limited to agents operating within large platforms or closed ecosystems.
Consider a small bakery that wants to use an AI agent to manage customer orders without needing technical expertise or its own hosting infrastructure. Its ordering agent might be associated with orders@moonbakery.com. A global switchboard can resolve this identifier to discover information describing the agent, its supported communication protocols, and where it can currently be reached.
Personal agents should have the same freedom. For example, a personal agent associated with ramesh@hotmail.com should be reachable by businesses or other authorized agents, because its owner does not control a hotmail.com domain. In both examples, the switchboard enables discoverability without requiring individuals or small businesses to manage dedicated discovery infrastructure. They can change providers or hosting services while retaining stable identifiers, allowing others to find their agents without being tied to a particular platform.
3. Why do we need agent discoverability to work without DNS?
DNS-based discovery works well for organizations that control their own domains. For example, a large airline such as the fictional BitAirlines could publish its agent information at https://bitairlines.com/.well-known/agents.json. However, many agents do not have their own DNS-based discovery paths. Personal agents associated with identifiers such as ramesh@hotmail.com may run on OpenClaw or platforms such as Meta Muse. Small businesses may rely on hosted agent services, while other agents may use identifiers managed by independent registries. Requiring every agent to have a DNS discovery anchor would limit discoverability and create barriers to participation, particularly for individuals and small businesses.
NandaIndex provides a complementary approach through a global switchboard that resolves registered agent identifiers to discovery information, without requiring agent owners to manage their own domains or discovery infrastructure. The goal is not to replace DNS, but to make agents discoverable across both DNS-based and non-DNS-based systems.
4. Why should agents remain discoverable over long time horizons?
Unlike a website visit or a single API request, interactions between AI agents may involve multiple turns and continue over days, weeks, or months. A supplier agent may coordinate recurring deliveries with a retailer, a customer agent may return to modify an order, or an airline may need to contact a passenger's personal agent about a flight change.
These multi-turn interactions cannot rely on permanent connections or previously stored endpoints. Agents may change providers, move to new servers, or become temporarily unavailable. Even when earlier sessions have ended, agents should be able to find one another again.
Stable identifiers and a common resolution mechanism enable long-term discoverability. A global switchboard helps agents locate each other's current discovery information, even as their underlying infrastructure changes. It does not maintain conversations or preserve their state, but enables agents to reestablish contact over time.
5. Why does open discoverability not compromise trust and autonomy?
Making an agent discoverable does not mean making it publicly accessible. Enterprises may allow verified suppliers to contact procurement agents while keeping internal financial agents private. Personal agents may similarly restrict interactions to authorized services.
A global switchboard does not require centralized trust or governance. NandaIndex resolves known agent identifiers without providing an openly searchable directory or unrestricted enumeration, reducing the risk of scraping and unwanted targeting.
Discoverability, identity verification, and authorization remain separate functions. Existing security systems and access controls determine who can communicate with an agent and what they can do. Agent owners retain control over which agents are discoverable and which interactions are permitted.
6. Why should the agentic web support different discovery systems?
The agentic web will extend beyond personal and enterprise agents to connected devices, robots, autonomous vehicles, telecommunications networks, and space systems. Each environment has different requirements for discovery, connectivity, security, and scale.
Connected devices may use local registries, robotics platforms may maintain specialized directories, and satellite systems may require discovery mechanisms that work with intermittent connectivity. No single discovery system is likely to meet all these requirements. Requiring everyone to adopt one system would limit interoperability and make it harder to accommodate existing and emerging technologies.
Conclusion
The agentic web will bring together independently operated agents, platforms, and discovery systems. A global switchboard provides a common resolution mechanism across these systems without requiring a single registry, protocol, or provider. Each system retains its own protocols, security policies, and governance, enabling interoperability without requiring uniformity. The goal is an open agentic web where individuals, small businesses, and enterprises can make their agents discoverable while retaining control over how they operate and who can interact with them.
MIT - Ramesh Raskar, Pradyumna Chari, Abhishek Singh, Ayush Chopra, Dario Fenoglio, Vedh Krishnan, Maria Gorskikh
Independent - Vint Cerf, Paul Mockapetris
Columbia University - Karthik Sreedhar
Cisco - Luca Muscariello, Vijoy Pandey, Sichao Wang
Google - Samuel Sharaf, Junjie Bu
Nasiko - Karan Bharadwaj, Ankit Kumar Nath, Chaman Singhal, Gaurav Tiwari, Rahul Todkar
Project NANDA - Umamaheswar Edara, Chris Pease, Sharath Chandra
Akamai - John Zinky, Hema Seshadri
UnifyDynamics - Mahesh Lambe
Tata Consultancy Services (TCS) - Rekha Singhal, Harrick Vin
Dell - Robert Lincourt
Synergetics AI - Raghu Bala
Flower AI - Dimitris Stripelis
HCLTech - Bhuwan B, Sumit Kumar, Vijay Guntur
PayPal - Arjun Chittoor
GSMA - Louis Powell
Motorola Solutions - Mahesh Saptharishi
TELUS - Amit Nag
Cognizant - Babak Hodjat
Nutanix - Debo Dutta
Kyndryl - Ilyas Iyoob
BigCommerce - Anil Kamath
Hello Haven - Frank Addante
Inkbox - Ray Liao
AG2 - Qingyun Wu
Mitsubishi Electric - Zafer Sahinoglu
Hexaware - Erik Haug
Ora - Liad Yosef
Ericsson - Mallik Tatipamula
WorkOS - Michael Grinich
Wipro - Sandhya Arun
Wand AI - Rotem Alaluf
Suki - Punit Soni
TinyFish - Sudheesh Nair
AGI House - Rocky Yu
L&T Technology Services - Piyush Jain
Nebius - Narek Tatevosyan