Glossary

Terms

Foundations

Data

Values recorded without interpretation applied to them. On its own it answers no question. See DIKW.

Information

Data with context or structure attached, enough to answer a question but not yet enough to act on. See DIKW.

Knowledge

Information that is current, scoped, attributable and usable for a decision or an action. See DIKW.

Wisdom

Judgement about when to apply knowledge, and when to abstain from applying it. See DIKW.

DIKW

The ladder from data to information to knowledge to wisdom, used here as a reminder that each step adds a property the step below does not have. See DIKW.

Knowledge layer

The part of an agentic system that states what is true, what is current, what the system is allowed to use, and what to do when it does not know. See Why Knowledge.

Grounding

Tying what an agent says or does back to knowledge it can point at, so the result is auditable rather than merely fluent. See Standards.

Currentness

Whether a statement is still the operative one — a property an agent checks rather than assumes. See Mementos.

Deprecation

Retiring knowledge that is no longer true so it leaves the usable picture instead of lingering in it. See Adoption.

Representation

Knowledge bundle

A directory of concept files carrying typed relations between them and a source for each claim, checked by a conformance linter. See Knowledge bundles.

Knowledge format

The serialisation a knowledge layer reads and writes — a graph syntax, a JSON-LD document, or a bundle of files. See Knowledge formats.

Taxonomy

A hierarchy of categories used to classify content, usually without the formal constraints an ontology carries. See Taxonomies.

Ontology

A formal description of the classes, properties and relations in a domain, written so that the structure is machine-checkable. See Ontologies.

Knowledge graph

Entities and typed relations stored as a graph, so a question can be answered by traversing relations rather than by scanning documents. See Knowledge graphs.

Triple

A single statement in subject–predicate–object form: the smallest unit of a graph. See RDF.

Concept scheme

A set of concepts with labels and relations between them — the unit of vocabulary a taxonomy or thesaurus is published as. See SKOS.

Lifecycle and trust

Provenance

The record of where a statement came from and what produced it. See PROV-O.

Memento

A resource that encapsulates a frozen prior state of another resource, retrievable by datetime. See Mementos.

Entity resolution

Deciding when two records refer to the same thing, so that facts about it are not split across duplicates. See Knowledge graphs.

Retrieval

Selecting, out of everything the system holds, the knowledge an agent is given for a task. See Token savings.

Agent systems

Agent

A model that takes actions in a loop against tools and an environment rather than answering once. See Agents.

Agentic AI

The domain in which actions, tool use and multi-step execution are the product, rather than a single response. See Agentic AI.

A2A

Agent-to-agent communication: how one agent negotiates, shares and verifies knowledge with another. See A2A.

MCP

The Model Context Protocol, an open standard that connects assistants to tools and data through hosts, clients and servers. See MCP.

Semantic communication

Exchanging meaning rather than bytes — sending the assertion and its qualifications instead of a document to be reinterpreted. See Semantic communication.

Delivery and cost

Context window

The amount of text a model can hold at once, which sets a hard ceiling on how much knowledge can arrive with a request. See Context engineering.

Context engineering

Deciding what knowledge enters the window, in what shape, and what is left out. See Context engineering.

Token savings

The reduction in tokens moved and re-read when the right knowledge is sent in the right shape, instead of a larger volume of the wrong kind. See Token savings.