Glossary of organisational AI

A

AI agent: An AI system that pursues a goal by deciding on steps and using tools, such as search, email or a database, rather than only replying with text.
AI gateway: A single controlled point through which AI applications reach business systems and data.
Approval: A decision by an authorised person that an action prepared by an AI or another person may go ahead.
Artifact: A piece of work produced by a person or an AI, such as a report, draft, analysis or article, held as a governed record with a status, versions and an owner.
Audit trail: A reliable, chronological record of what happened: who acted, through which agent, on what resource, with what action, under which rule and with what result.

Channel: A combination of an audience, a projection, artifacts, events and Skills through which an organisation exposes information and capabilities.
Connector: The component that tells Kroy how to communicate with another system — its authentication, resources, actions, events and limits.
Context engineering: The practice of deciding what information, tools, instructions and history an AI receives for a task, and in what form.

Delegation: The explicit authority a person gives an AI agent, or another principal, to act on their behalf, limited in scope, actions and time.

Entity: A structured record of something an organisation works with, such as a client, project, matter, service, contact or decision, with defined fields and relationships.
Event: A record that something meaningful has changed in the organisation, which authorised people, systems or agents can be told about.

Guest: A person or organisation outside your own, such as a client, auditor, adviser, contractor or supplier, given controlled access to selected information and capabilities.

Human in the loop: An arrangement in which a person reviews or decides at a defined point in an AI-driven process, rather than the AI acting entirely on its own.

Identity: The verified answer to "who is this?" for a person, an AI agent, a service or an organisation making a request.

MCP (Model Context Protocol): An open standard for connecting AI applications to external tools and data through MCP servers, which advertise tools, resources and prompts an AI can use.

Policy: A rule that decides whether a request is allowed, based on who is asking, which agent, what resource, what action and in what circumstances.
Principal: Any party that can be identified and granted permissions: a person, an AI agent, a service, a guest or an organisation.
Projection: The portion of an entity’s information that a particular audience is permitted to receive.
Provenance: The record of where a piece of information came from, who or what recorded it, when, and on what basis.

Relationship: A defined link between two entities or organisations, such as a contact belonging to a client, a task belonging to a project, or one firm acting for another.

Skill: A packaged business capability that bundles the context, permissions and connector actions needed to do one job safely.
State: What is true now about an organisation’s work — its decisions, tasks, artifacts and relationships — held independently of any single AI tool.
Subscription: A standing request by an authorised person, system or agent to be notified when a particular kind of event occurs.