TLDR: The AGNTCY Agent Directory Service (ADS) is a distributed directory designed for discovering AI agent capabilities, metadata, and provenance across diverse Multi-Agent Systems. It uses content-addressed storage, hierarchical taxonomies, and cryptographic signing (Sigstore) to enable efficient, verifiable, and multi-dimensional discovery. Built on the Open Agentic Schema Framework (OASF) and leveraging OCI/ORAS infrastructure, ADS decouples capability indexing from content location through a two-level mapping over a Kademlia-based Distributed Hash Table (DHT), supporting federated operation and extensible agent modalities.
In the rapidly evolving world of Artificial Intelligence, Multi-Agent Systems (MAS) are becoming increasingly common. These systems rely on various AI agents, each with specialized capabilities like natural language processing, computer vision, or reasoning. However, as these ecosystems grow, finding the right agent for a specific task becomes a significant challenge. Imagine trying to find an agent that can perform ‘regulated clinical summarization with streaming JSON output’ or a ‘reasoning agent with signed provenance and acceptable latency profiles.’ This is precisely the problem the AGNTCY Agent Directory Service (ADS) aims to solve.
What is the AGNTCY Agent Directory Service (ADS)?
The ADS is a groundbreaking distributed directory designed to help discover AI agent capabilities, their associated metadata, and their origin (provenance). Developed by Luca Muscariello, Vijoy Pandey, and Ramiz Polic from Cisco Systems, ADS acts as a central hub for agents to register their skills and for users to find them efficiently and securely. It’s built to handle the complexity of diverse agent types, from large language model (LLM) prompt agents to more traditional AI components.
How Does ADS Work?
At its core, ADS uses a clever two-level mapping system. First, it maps an agent’s capabilities (like ‘summarization’ or ‘image classification’) to a unique Content Identifier (CID). This CID is essentially a cryptographic fingerprint of the agent’s definition, ensuring its integrity and immutability. Second, it maps these CIDs to their actual storage locations, which are typically in existing, robust OCI (Open Container Initiative) registries. This separation means that even if an agent’s physical location changes, its capability index remains stable.
The system leverages several proven technologies:
- Open Agentic Schema Framework (OASF): This provides a structured and extensible way to describe agent capabilities, operational traits, and performance signals. It allows for future evolution without breaking existing records.
- Content-Addressed Storage: Every agent definition is stored based on its content’s cryptographic hash. This makes records immutable and verifiable; any tampering would change the hash, making it immediately detectable.
- Kademlia-based Distributed Hash Table (DHT): This decentralized network helps efficiently route queries and locate agent records and their storage peers.
- OCI / ORAS Infrastructure: ADS reuses mature Open Container Initiative (OCI) registry standards and ORAS tooling for distributing agent artifacts. This means it benefits from existing caching, access control, and security scanning ecosystems.
- Sigstore for Provenance: For enhanced security, ADS integrates Sigstore, a system that provides cryptographic signing and transparency logs. This allows users to verify the origin and integrity of an agent, ensuring it hasn’t been tampered with in the supply chain.
Key Benefits and Features
The ADS offers several significant advantages:
- Capability-Centric Discovery: Users can search for agents based on specific skills, domains (e.g., ‘healthcare’, ‘finance’), and features (e.g., ‘streaming-output’, ‘low-latency’). This multi-dimensional querying allows for precise and efficient agent discovery.
- Verifiable Integrity and Provenance: Thanks to content addressing and Sigstore, all agent records are immutable and their origin can be cryptographically verified, building trust in the agent ecosystem.
- Federated Operation: Different organizations can operate their own autonomous registries while still participating in a shared discovery network. This promotes decentralization and avoids a single point of control.
- Interoperability: By reusing existing OCI standards, ADS ensures compatibility with a wide range of tools and infrastructure, avoiding the need for bespoke solutions.
- Scalability and Performance: The two-level mapping, combined with hierarchical taxonomies and flexible replication strategies, ensures that discovery performance scales efficiently even with a large number of agents.
Also Read:
- Agentic AI: Reshaping the Future of Computing Environments
- AI-Powered Analytics for High-Performance Computing Operations: Introducing the EPIC Platform
Beyond Discovery
ADS is designed to complement other emerging initiatives in the AI agent space. While projects like NANDA focus on human-readable naming and resolution, and protocols like MCP and A2A address runtime communication between agents, ADS specifically targets the challenge of capability-centric discovery. It provides the foundational layer for finding the right agent, which can then be integrated into broader agentic systems.
The future of ADS includes further integration with naming systems, advanced ranking and evaluation mechanisms, privacy and confidentiality extensions, and continued performance optimizations. This service is poised to become a crucial component in enabling the next generation of interconnected and intelligent multi-agent systems. To learn more, you can read the full research paper here.
