Terminology for the Discovery of Agents With Names (DAWN)

Introduction Distributed systems increasingly rely on the dynamic composition of services, agents, and workloads that may not have pre-configured connectivity relationships. In an AI-based system, an AI agent may need to find another agent with specific capabilities. In these cases an AI agent needs knowledge of remote agents before interaction can proceed: what they are, what they offer, and whether they can be trusted. Such knowledge could be obtained through static configuration, but this approach is impractical at scale and across organisational boundaries. Automated discovery mechanisms are therefore needed. Today, where automated discovery of AI agents exists, it is typically handled through proprietary directories or platform-specific mechanisms. These approaches do not scale across organisational boundaries and create fragmented ecosystems where entities cannot find entities managed by other organisations. An interoperable mechanism for the discovery of AI systems and resources (e.g., AI agents) is needed that builds on existing protocols and tools, benefits from an established trust model, supports proven delegation and federation architectures, and allows organisations to independently publish discovery information. This document defines common terminology for use in documents that discuss Discovery of Agents With Names (DAWN).

Discovery Scope [Editor's Note: This section probably belongs in an Architecture document and not here.] It will be helpful to understand the broader discovery ecosystem. Information must be registered to the discovery system, may be managed and updated, can then be discovered by other AI agents. But the discoverable information must be balanced as a compromise between too much detail which will overload the discovery process or be duplicative and thus cause maintenance issues, and too little information which would make discovery suboptimal. Additional detailed information may be exchanged following initial discovery, and this part of the process may also include negotiation of capabilities. In the end, this means that the "discovery ecosystem" can be broken down into at least the following functions:

Registration and management of an entity to be discovered.
Search based on a formalised encoding of search criteria.
Discovery lookup based on a minimum viable dataset (protocol, identifying properties) with return of a set of minimum discoverable information (including reachability).
Reading of additional information.
Capability exchange and negotiation.

From an implementation perspective, some of these functional components might be included within the same module, and it is possible that some elements might be achieved using the same protocols. However, for the sake of this document and the DAWN work, only the discovery lookup is in scope.

Terminology The terms presented in this section are in alphabetic order for ease of reference. For those wishing to read this document to gain an understanding of the DAWN scenery, it is recommended to read the terms in the order presented in .

Key DAWN Terms in a Readable Order

Agent:: A software entity that acts autonomously or semi-autonomously on behalf of a user, organisation, or system. An agent may initiate interactions with other entities, make decisions, and perform tasks.
AI Agents:: A specific class of agent that employ artificial intelligence techniques.
AI Resource:: An identifiable entity or service that provides AI-related functionality, enables access to such functionality, or facilitates the discovery of other AI resources. Examples include AI agents, AI models, and AI inference services. An AI resource does not necessarily perform AI processing itself and may represent an entire AI system or an individually accessible component of one.
AI system:: A system that uses AI to process inputs and generate outputs such as predictions, content generation, recommendations, or decisions, to achieve objectives. An AI system may comprise multiple interacting components such as AI agents or AI resources.
Attributes:: The properties, features, capabilities, skills, etc., that an entity possesses or may have access to, such as skill type, communication language, capacity, task description, contact information, ID, etc. The attributes of an entity form a full detailed set of information about the entity not all of which is necessarily available to be discovered or communicated to other entities. (See also, properties and capability.)
Capability:: A description of the functions, services, or actions that an entity can perform. Capabilities may be described using structured schemas such as capability cards.
Capability Card:: A structured, machine-readable description of an entity's capabilities and interface. Variants include agent cards, task cards, resource cards, tool cards, and skill cards depending on the type of entity. A capability card may also include human-readable information as well as authentication/security certification data.
Capability Exposure:: The processes by which entities expose their capabilities. Such exposure may be part of the registration or management of discovery information prior to the discovery processes, or achieved through an interaction with an entity. (See also, Capability Exchange.)
Capability Exchange:: The processes by which entities exchange details between each other of what they can do, dynamic status information, and which particular features or functions they wish to engage.
: Capability exposure, exchange, and negotiation are out of scope for DAWN, but will form an essential part of the entity ecosystem enabling the selection and operation of entities.
Discoverable Object:: An information object that is discoverable and includes information that defines what an entity is, what attributes it possesses, how to reach the associated entity, etc. May be represented as a capability card.
Discoverable Object Validation:: The process that verifies a discoverable object, ensures its compliance to referenced standards, and makes it available to be discovered.
Discovered Entity:: An entity whose properties are returned as the result of a discovery process. A discovered entity may be a specific instance or a member of a class of entities that can perform a desired function. In the context of DAWN, a discovered entity is an AI system or AI resource (e.g., an AI agent).
Discovering Entity:: An entity (or its operator) that initiates the discovery process in order to find other entities to interact with.
Discovery:: The process by which an entity or its operator locates other entities that are capable of performing a desired function or providing a desired service, and obtains sufficient information to initiate interaction. In the context of DAWN, the discovery process is about how a client can discover public properties of AI agents or AI resources to facilitate subsequent interactions such as selection, invocation, or communication. Specifically, this not include open-ended searches, semantic matching ranking, or selection of an AI resource based on capabilities or user intent.
Discovery Information:: The information returned by a discovery mechanism that allows the discovering entity to decide whether later interaction is possible and desirable. This should be considered in terms of the Minimum Discoverable Information (q.v.) of which Discovery Information is a subset, and compared with the broader set of information that forms part of the Capability Exchange (q.v.).
Discovery Mechanism:: A protocol, system, or method used to perform discovery. Examples include Domain Name System (DNS) based service discovery, directory services, and distributed registries.
Discovery Scope:: The explicit domain over which discovery is performed. Discovery scope may be specified in one or more dimensions, including but not limited to administrative identifiers (e.g., DNS domain names, AS numbers), trust domains, topological or distance metrics, geographic or jurisdictional boundaries, and temporal constraints. Discovery scope bounds the search space and supports scalability, relevance, and policy enforcement.
Entity:: A system component that communicates with other entities in a peer-to-peer or client-server relationship. In the context of DAWN, entities include, but are not limited to, AI agents and AI resources. However, see the definition of "Discovered Entity."
Function:: The processing that an entity offers. Examples include task execution, data transformation, inference, routing, steering, storage, and orchestration.
Minimum Discoverable Information (MDI):: The minimum amount of information an entity needs to provide to become discoverable.
Minimum Viable Dataset (MVD):: The data that must be provided to the discovery mechanism to enable it to return indications of Discovered Entities.
Named Entity:: An entity that is identified by a stable name within a naming system. The naming system may be hierarchical (e.g., DNS) or flat.
Properties:: The discoverable characteristics of an entity. Properties include, but are not limited to, communication protocols, capability cards, location, trust indicators, and operational status. (Compare with attributes.) In DAWN, the properties are those needed to select a communication means, i.e., a protocol and achieve communication.
Registrar:: An entity or system responsible for accepting and maintaining records about entities that wish to be discoverable. Acts as a go-between for registrants and registries through the process of registration. Note that not all registrations require the use of a registrar, and that the registrar in a DAWN system is not necessarily the DNS registrar .
Registration:: The steps by which agents can register their existence via a registrar. This should include attestation and other security mechanisms.
: Registration is out of scope for DAWN, but the information that can be discovered and the trust with which that information is treated are key to any complete system.
Selection:: The mechanisms and policies by which an entity determines which discovered entities it will interact with.
: Selection is out of scope for DAWN, but depends on information obtained through discovery.
Trust Indicator:: Information associated with an entity that allows a discovering party to assess the trustworthiness or provenance of the entity and its advertised properties. Examples include digital signatures, certificates, and attestations.

IANA Considerations This document does not make any requests of IANA.

Security Considerations This document only defines a set of terms. It does not introduce any issues that require security consideration. However, this document does define Trust Indicator, attestation-dependent Registration, and scope-based policy enforcement, and so it should be clear that discovery has significant security implications. It is anticipated that these will be addressed in the DAWN solution/architecture documents.

Privacy Considerations This document only defines a set of terms. It does not introduce any issues that require privacy consideration. However, this document does use the concepts of names and identities, locations, reachability, operational status, capabilities, and trust indicators, so it should be clear that discovery has significant privacy considerations. It is anticipated that these will be addressed in the DAWN solution/architecture documents.

Operational Considerations This document only defines a set of terms. It does not introduce any issues that require operational consideration.

Acknowledgements The authors wish to acknowledge the contributions of participants in the DAWN discussions that shaped this document. Jim Mozley, Med Boucadair, Chenguang Du, Daniel King, Benoit Claise, Kaveh Ranjbar, Muhammad Awais Jadoon, and Aditya Dogra provided useful reviews of this document.