Software agents are increasingly able to discover, compose, and invoke one another’s capabilities across organizational boundaries, but the mechanisms used to discover them are diverging, a natural consequence of innovation across different communities, use cases, and deployment environments. AI Catalog is a strong starting point for public enterprise discovery. It provides a typed, nestable, machine-readable container for heterogeneous AI resources: A2A Agent Cards, MCP server descriptors, nested catalogs, tools, skills, gateways, and other AI resources.

But AI Catalog alone does not connect the wider ecosystem. Enterprises also rely on DNS-based service discovery, gateways, and platform registries, while telecom directories, EdgeAI and IoT systems, and sovereign or sector-specific registries continue to emerge. These approaches are valuable, the product of real innovation, and each is authoritative within its own domain. Left unconnected, however, they form discovery islands.

NandaIndex is a Federated Resolution Architecture that acts as a global switchboard. Given a stable resource identifier, it determines which discovery mechanism a requester should use, then defers to that mechanism. It reuses the AI Catalog format as its record type, so it introduces no new schema, and connected systems do not need to adopt AI Catalog internally. NandaIndex complements discovery protocols such as ARD and federated infrastructure such as the AGNTCY Agent Directory. It builds on the switchboard proposed by NANDA and AGNTCY, developed jointly by Outshift (Cisco), MIT Media Lab and Nasiko, and extends that model with improvements to AI Catalog.

Simple cases stay simple. An enterprise that publishes an AI Catalog at a well-known endpoint and owns its domain resolves directly, and NandaIndex stays out of the critical path. NandaIndex is a distributed index of pointers, not a centralized registry, and adds value where direct resolution is unavailable:

• Enterprise heterogeneity, where companies use DNS-based service discovery, gateways, or legacy registries instead of a published AI Catalog.
• SMBs such as Moon Bakery, whose website may be on Wix, Squarespace, or Shopify, whose agent runs on AWS, and whose agent card is hosted by a third party such as host39.org.
• Individuals such as john@hotmail.com, who do not own a domain, with a personal agent on AWS, Azure, GCP, or another provider and an agent card hosted elsewhere, discoverable when their account identity is verifiably bound to the resolution record.

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