Learn
Metadata glossary: terms for getting data ready for AI agents
Short answer
Thirty-four terms that come up when you prepare enterprise data for AI agents, each defined in one to three sentences. Where a term comes from a specific vendor's product, we say so and link the source.
A
- Active metadata
- Metadata that is used by other systems, not only read by people: for example a label that tells an agent an asset is certified, or a trigger that alerts an owner when a schema changes.
- Agent context
- The information an agent receives with a task: which tables exist, what the columns mean, which metrics are certified and what the requesting user may see.
- AI context layer
- A layer between your data stack and AI agents that assembles lineage, definitions, trust signals and access rules into context an agent can request at query time. Vendors also call it a context platform or knowledge layer.
Lesson: Catalog, semantic layer or AI context layer: which does what
B
- Business glossary
- A governed list of business terms and their agreed meanings, linked to the data assets that implement them.
C
- Certification
- A mark that an asset has been reviewed and approved for use, often applied by a data owner. Some platforms apply certification automatically from rules.
- Column-level lineage
- Lineage traced to individual columns, so you can see which source columns feed a field in a dashboard or a metric an agent uses.
- Context drift
- When the context an agent relies on no longer matches reality, for example after a model is renamed or a metric definition changes upstream. DataHub describes detecting drift and routing it to experts.
Source: DataHub Cloud context platform
- Context graph
- A graph of assets, their relationships, definitions, owners and rules, built so that a slice of it can be served to an agent. Euno describes computing its graph ahead of time from APIs and query logs.
D
- Data catalog
- An inventory of data assets with descriptions, owners and search, built mainly for people to find and understand data.
Lesson: Catalog, semantic layer or AI context layer: which does what
- Data contract
- An agreement between data producers and consumers on schema, quality and freshness, often checked automatically.
- Data lineage
- A record of where data comes from and how it moves and changes between systems, from source tables through transformations to reports.
- Data product
- A dataset packaged for reuse with an owner, documentation, quality signals and access rules. Collibra describes certified data products as the unit agents should use.
Source: Collibra homepage
- dbt
- A widely used transformation tool in which teams define models and tests in SQL. Its project files are a common source of lineage and definitions.
- Deployed semantic object
- AtScale's pricing unit: the semantic objects you deploy, instead of seats or queries.
Source: AtScale pricing
I
- Impact analysis
- Using lineage to see what breaks downstream before you change a table, column or model. Euno lists impact analysis with CI integration.
Source: Euno pricing
K
- Knowledge graph
- A graph of entities and relationships with explicit meaning, often used to unify data from many systems. Stardog and ServiceNow Data Catalog both describe knowledge graphs.
Source: Stardog homepage
- Knowledge layer
- A newer label for a layer that gives agents business context; vendors use it alongside context layer and semantic layer.
L
- LookML
- Looker's modeling language for dimensions, measures and joins. It holds business logic that agents may need to see.
M
- MCP (Model Context Protocol)
- An open protocol that lets AI applications call tools and data sources through a standard interface.
- MCP server
- A service that exposes tools or data to AI agents over MCP. Several platforms on this site ship one, including Euno, Atlan, Collibra, DataHub, Cube and AtScale.
- Metadata
- Data about data: names, types, owners, descriptions, lineage, usage and quality.
- Metric view
- A governed metric definition stored in the data platform. Databricks describes metric views as part of Unity Catalog business semantics.
Source: Databricks: Unity Catalog business semantics (read as a search summary)
O
- Ontology
- A formal model of the concepts in a domain and how they relate. Alation's Ontologies use OWL and RDF with SHACL constraints.
Source: Alation AIOS: Ontologies
P
- Persona-scoped context
- Context filtered to the role, task and access of the person or agent asking, so an agent sees only what it needs and may use. Euno lists personas as a context layer.
Source: Euno: Context for AI Agents
- PII
- Personally identifiable information. Tracking which assets depend on PII lets you mark them as not ready for agents.
Q
- Query log
- The record of queries run against a warehouse. Context platforms mine it for lineage, usage and join patterns.
R
- RBAC
- Role-based access control: permissions granted by role rather than by individual.
- Row- and column-level security
- Rules that hide specific rows or columns from users who should not see them. AtScale applies them to agent queries.
S
- Semantic layer
- A layer that defines metrics, dimensions and business logic once so BI tools and agents compute them the same way.
Lesson: Catalog, semantic layer or AI context layer: which does what
- Semantic model
- The set of definitions inside a semantic layer: measures, dimensions, joins and their names.
- Semantic view
- A semantic definition stored inside the warehouse. Jedify names Snowflake Semantic Views among the Cortex AI features it integrates with.
- SOC 2 Type II
- An audit report on a service provider's security controls over a period of time, commonly requested by enterprise buyers.
T
- Table-level lineage
- Lineage traced between tables only, without showing which columns feed which. Faster to build, less useful for agents that need field meaning.
- Text-to-SQL
- Turning a natural-language question into a SQL query. Without context about definitions and trust, the query can run correctly and still answer the wrong question.