Agent discovery is diverging into multiple useful approaches — a natural consequence of innovation across different communities, use cases, and deployment environments. AI Catalog is a useful starting point for public enterprise agent and AI-resource discovery. It provides a typed, nestable, machine-readable container for heterogeneous AI artifacts: A2A Agent Cards, MCP server descriptors, nested catalogs, tools, skills, datasets, gateways, and other AI resources.

However, AI Catalog needs to be strengthened by allowing other discovery mechanisms to connect through it. These include DNS-AID, ANS, .well-known, A2A cards, MCP, gateways, and platform registries. In addition, telecom directories, EdgeAI registries, IoT systems, and country- or sector-specific discovery systems will each continue to solve part of the problem. These non-compatible or emerging approaches are valuable — the product of real innovation — but they can also create discovery islands when left unconnected.

A NandaIndex Federated Resolution Architecture is proposed as a global switchboard that uses the AI Catalog format itself as the common index-record format. NandaIndex strengthens AI Catalog by making it the preferred framework through which other discovery systems can be bridged. The switchboard proposed by NANDA and AGNTCY, built jointly by Outshift (Cisco) and MIT Media Lab, provides the starting point for this architecture. The architecture considers improvements to AI Catalog and extends the switchboard model to cover enterprises, SMBs, and individuals.

The focus is on three practical needs:

• Enterprise heterogeneity, where some companies use AI Catalog and others use DNS-AID, gateways, .well-known, or legacy registries.
• SMBs such as MoonBakery, where website maybe listed on Wix, Squarespace or Shopify, agents may be hosted on AWS while agent cards are hosted by a third party such as host39.org.
• Individuals such as john@hotmail.com, who do not own a domain name, where a personal agent may be hosted on AWS, Azure, GCP, or another provider, while the agent card lives elsewhere.

NandaIndex presents an architecture proposal and initial design. It identifies the missing bootstrap layer between identity and discovery, defines the role of NandaIndex within that layer, and outlines the identity, trust, and federation mechanisms that must be formalized.