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. For example, an AI agent may need to find another agent with specific capabilities, a workload orchestrator may need to locate compute resources in a particular jurisdiction, or a service consumer may need to discover providers that support a required protocol or a data schema version. Further use cases and scenarios may be considered, but it is out of scope to enumerate them. In each case, an entity needs knowledge of remote entities 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 discovery mechanism 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 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 entities. But the discoverable information must be balanced as a compromise between too much detail which will overload the discovery process, be duplicative and thus cause maintenance issues, and too little information which would make discovery suboptimal. Additional detailed information may be exchanged between entities 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 data set (protocol, identifying properties) with return of a minimum set of 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 are a specific class of agent that employ artificial intelligence techniques.
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.
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.) 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, tools, skills, tasks, compute workloads, software services, task owners, network functions, and application endpoints. However, see the definition of "Discovered Entity."
Function:: The functional processing capability 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. Think of it as common header of a data structure.
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.)
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.
Task:: A legacy term kept for continuity with earlier work. A task may be considered as software service.
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.
Workload:: A unit of compute or processing that is deployed and executed within a hosting environment. Workloads may be transient or long-lived and may move between hosting environments.

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 and privacy 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.

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, and Kaveh Ranjbar provided useful reviews of this document.