Glossary

This glossary defines terms used repeatedly in the D.Hub interface and user documentation. Interface labels are used as the headwords, with an abbreviation included only when it helps distinguish a concept.

Core assets and data

Term	Description
Alias	A display name set separately from the system identifier. When an alias exists, it is shown first in trees, lists, and search results.
Collection	A work scope that groups assets used together, such as datasets, code, pipelines, knowledge, dashboards, and agents.
Collection item	A dataset, code asset, pipeline, knowledge, dashboard, or agent contained in a collection or folder.
Data Connection	Connection settings used to access an external database, file store, or API. Also called a connector in technical contexts.
Vector DB	A database that stores text or image embedding vectors for similarity search. Under Data Connections, D.Hub registers connection information for an external vector database such as Qdrant.
Secret	A separately stored authentication value that must not be exposed again, such as a password, API key, or access key. Data Connections and tools use references to stored secrets.
Batch pipeline	A pipeline run directly by a user or on a saved schedule. Used for file loading, recurring aggregation, and bulk transformation.
Event pipeline	A pipeline run when a registered event arrives. Used for real-time ingestion and follow-up event processing.
Pipeline node	A step that performs one operation in a pipeline, such as data input, transformation, filtering, or output. Connect nodes to build the data flow.
Data version	A snapshot created when dataset data is committed. Preview a sample from a version or download it as CSV or Parquet.
Metadata version	A version that records changes to a dataset's alias, type, description, tags, options, and schema. Restore metadata restores only this information and does not roll back table data.
Version History	A tab that shows change history for a dataset, code asset, pipeline, or knowledge. Available actions, such as preview, restore, and download, depend on the asset type.
Manifest	A JSON definition file that transfers asset metadata and configuration when importing a collection scenario or shared resource.
Dashboard	An application that displays dataset or query results in widgets such as charts, tables, and maps.
Date Range Selector (DRS)	A control that sets the dashboard query period. The selected range applies to a widget query unless the widget has its own time range.
Report	An application that generates and manages document-style results from data. The organization-wide model used to generate reports is selected under Assistant Models.

Ontology

Term	Description
Entity type	Defines the structure of a business object such as a customer, product, or sensor. Configure properties, an identifier key, a display column, and a data source. In a graph, instances of the type appear as nodes.
Relationship type	Defines the connection structure and direction between two entity types. In a graph, relationship instances of this type appear as lines between two nodes.
Entity and relationship instance	An actual data row loaded according to an entity- or relationship-type schema. The type defines the structure; an instance contains values in that structure.
Identifier key	A column that uniquely distinguishes an entity or relationship instance. It is required to use an entity or relationship as pipeline input or output.
Display column	The column whose value represents an entity or relationship instance on screen. Only one can be set per schema.

Agents and Semantic Layer

Term	Description
D.Hub Assistant	The built-in D.Hub assistant used to ask questions and get help with product features. It uses the organization's active assistant model and operates independently from user-created agents.
Agent	An asset that combines an LLM, tools, actors, prompts, and semantic resources to process requests. Create it in AI Agent or Workflow mode. The mode cannot be changed after creation.
Tool	A capability invoked by an agent during a task. Configure it to run Python code, another agent, or a pipeline.
Actor	A Python function-like execution unit invoked by an agent. The actor's confirmation policy determines whether human approval is required before execution.
Prompt	An asset that stores system instructions and variables for reuse across multiple agents.
Semantic Layer	An area for managing semantic tables, value dictionaries, and query templates used to translate natural-language questions into data queries.
Semantic table	A semantic resource that defines the meaning and relationships of tables and columns used in natural-language queries. Semantic tables registered on an agent define its data-query scope.
Value dictionary	A dictionary that maps multiple expressions for the same value to one normalized value, for example mapping two alternate city names to the same value.
Query template	A resource that stores a recurring question pattern, slots, target semantic table, and query. Use it to reuse a validated query for similar questions.
Human approval (HITL)	A pattern that displays an approval request before execution when an actor's confirmation policy is set to Confirm. The user selects Approve or Reject. The user-input node in a workflow agent is a separate feature for receiving a value.

Knowledge and search

Term	Description
Knowledge Management	An area that collects documents for search and as evidence in AI Chat. After parsing, chunking, and indexing documents, manage them on the Documents, Chat, Search, Settings, and Sharing and Permissions tabs.
Chunk	A piece of text created by dividing an added document into searchable units. Choose a chunking strategy and size for the document format and purpose.
Embedding	A numeric vector representation used to compare the meaning of text. Select an embedding model when enabling semantic search.
Keyword search	A search method that finds document chunks containing entered words or exact terms.
Semantic search	A search method that finds document chunks with meaning similar to the query.
Hybrid search	A search method that combines keyword and semantic search results. Select it based on the on-screen method and results, rather than an internal search-algorithm label.
RAG	A method that retrieves documents related to a question and supplies them as evidence for an LLM answer. In Knowledge Chat, review the cited source together with the answer.
LLM	Abbreviation for Large Language Model, a language model that generates answers or code from input text. D.Hub distinguishes LLM Models and Assistant Models by where the model is used.
LLM Model	A model resource selected individually by an agent, workflow agent, or Knowledge AI Chat. Register it for a collection or for organization-wide use.
Assistant Model	An organization-wide model shared by D.Hub Assistant, AI code generation and editing, and report generation. An administrator selects one active model.
Model Serving	A feature that deploys an ML model for training or inference so it can serve requests. It is a different screen from LLM Models, which registers conversational models.

Accounts and access control

Term	Description
User type	The Administrator, Manager, or User classification that determines which administrative features an account can access. It differs from a sharing role, which controls access to individual assets.
Sharing role	The Owner, Editor, or Viewer role that determines what a user or group can do in a collection or asset. A collection role is inherited by child assets.
Service account	An automation account that does not use interactive sign-in. Issue an access token for CI, pipelines, or external integrations to call the D.Hub API.
OIDC client	Registration information used when an external system or SSO application obtains a D.Hub token. Used for delegated user sign-in and server-to-server calls.
Marking	An access-restriction label applied to a sensitive resource. Only users and groups with an access grant for the marking can open the resource.

Next steps

• Core concepts — Understand how D.Hub components relate to each other.
• Quick start — Upload sample data and review the result in a dashboard.
• API overview — Review the REST API reference.