Horizon Context¶

Horizon Context is the governed context layer built into Horizon Catalog. It collects metadata from across your data estate, enriches it with business meaning, and activates it so that people, BI tools, and AI agents all work from the same trusted definitions.

Context doesn’t live in one place. It’s spread across your BI tools, your query history, your pipeline definitions, and the knowledge your teams have built up over years. Horizon Context connects to those sources and keeps the results in one governed system. When an AI agent queries data, Snowflake enforces the role’s access control, masking, and row-access policies at the query engine. Access to metadata is managed separately from access to the underlying data.

Why use Horizon Context?¶

A table name and a column list tell you what a dataset is called, not what it means or whether you should trust it. That gap is what makes data hard to find, reports inconsistent, and AI answers unreliable.

Horizon Context addresses that gap in three ways:

One catalog for the whole estate. Snowflake objects, external databases, BI dashboards, and pipeline models are described in one place, so you don’t have to know which system a dataset lives in before you can find it.
One set of definitions. Business metrics, dimensions, and relationships are defined once in semantic views and reused by every consumer, instead of being redefined in each dashboard and each prompt.
One governance model for data queries. Snowflake applies the same role-based access control, masking, and row-access policies to queries from people, BI tools, and AI agents. Metadata visibility can require separate grants.

Layers of context¶

Horizon Context gathers four kinds of context. Each layer answers a different question about a data asset, and an agent or analyst usually needs all four to pick the right dataset and interpret it correctly:

Layer	Question it answers	Examples
Structural	What exists, and how is it connected?	Tables, columns, dashboards, joins, lineage
Operational	What’s happening to it?	Query history, freshness, data quality metrics
Semantic	What does it mean?	Business definitions, metrics, relationships in semantic views
Behavioral	How is it used, and by whom?	Popularity, query patterns, downstream consumers

How Horizon Context works¶

Horizon Context works in three stages: it collects raw context, enriches it with business meaning, and activates it for the tools and agents that consume your data.

Collect¶

Snowflake continuously gathers metadata from objects in your account. To cover the rest of your estate, Horizon Context adds two more sources:

Metadata connectors: Depending on the connector and its configuration, out-of-the-box connectors collect schemas, query logs, dashboard definitions, and popularity data from external databases, BI tools, and data pipeline systems, including PostgreSQL, SQL Server, Tableau, Power BI, and dbt. Metadata connectors are in private preview. Contact your account team for availability and setup details.
External lineage: Any OpenLineage producer, such as Apache Airflow, can post lineage events to a Snowflake REST endpoint. Snowflake folds those events into the same lineage graph it builds from its own query history.

Enrich¶

Collected metadata becomes useful only after it carries business meaning. Horizon Context enriches it with the following:

Semantic views: Governed, business-aligned definitions of your metrics, dimensions, and relationships. You can build them in Semantic Studio, generate them with Autopilot, or ingest them from Tableau and Power BI files.
Auto-generated descriptions: Snowflake Cortex writes table and column documentation from metadata and sample data, so undocumented objects don’t stay that way.
Object tagging: Tags record project, sensitivity, environment, and certification status. Snowflake-provided tags give you a consistent vocabulary across every account. For more information, see Snowflake-provided tags.
End-to-end lineage: Snowflake stitches lineage mined from Snowflake query history, external query logs, BI systems, and OpenLineage feeds into a single column-level graph.

Activate¶

Enriched context is only valuable when something consumes it. The same context layer serves the following:

People browsing the catalog in Snowsight or searching with Universal Search.
Cortex Agents and Cortex Analyst, which answer natural language questions against semantic views.
Snowflake CoCo, which uses catalog context when it writes SQL, traces lineage, or reviews governance posture.
BI tools and applications that read semantic definitions instead of reimplementing them.

Example use cases¶

1. See metadata and cross-platform lineage for data outside Snowflake¶

A retail analytics team runs reporting on Snowflake, but raw orders land in PostgreSQL first, dbt builds the transformation layer, and executives read the results in Tableau. Nobody can answer a simple question, such as which dashboard a given PostgreSQL column ends up in, without opening three tools.

With metadata connectors configured for PostgreSQL, dbt, and Tableau, Horizon Context collects the schemas, query logs, model definitions, and dashboard definitions from each system and indexes them alongside Snowflake objects. Lineage from all of those sources is stitched into one column-level graph, so the team can start at a PostgreSQL column and follow it through dbt models and Snowflake tables to the Tableau dashboards that display it.

Metadata connectors are in private preview. Contact your account team to find out which connectors are available to you. If your pipelines emit OpenLineage events, you can send them to Snowflake today. For more information, see External lineage.

2. Certify trusted assets so that search and AI point to them¶

A finance organization has seven tables with revenue in the name. Three are sandbox copies, two are staging, and only one is the table that the close process uses. New analysts pick the wrong one, and so do AI agents.

A data steward applies the SNOWFLAKE.TAGS.CERTIFICATION_STATUS tag to mark the authoritative table 'CERTIFIED' and the sandbox copies 'DEPRECATED'. Because the tag’s allowed values are fixed and can’t be modified, every account interprets them the same way, and governance and AI workflows can act on them without custom mapping.

Certification then pays off in two places:

Search and discovery. Universal Search searches object and column tags along with names and comments, so searches can match certification status as part of an asset’s metadata.
AI understanding. Agents that retrieve context from the catalog get an explicit, machine-readable trust signal that separates the production table from the sandbox copies, instead of guessing from the table name.

Pair certification with auto-generated descriptions and a semantic view over the certified table. The description explains what the asset holds, the semantic view defines how to aggregate it, and the certification tag says it’s the one to use. For more information about the tag and its allowed values, see Snowflake-provided tags.

3. Run cross-platform impact analysis with Snowflake CoCo¶

A data engineer needs to drop a column from a staging table. The Snowflake dependencies are easy enough to find, but the risk is in what sits beyond Snowflake: a dbt model that selects the column and an executive dashboard built on the result.

Because Horizon Context already stitches lineage from Snowflake, external databases, BI tools, and OpenLineage feeds into one graph, the engineer doesn’t have to assemble that picture by hand. They can ask Snowflake CoCo a question in natural language, such as “What will break if I change raw_db.sales.orders?” CoCo reads the same lineage that the catalog exposes and reports the downstream objects, ranked by risk, along with how often each is queried and how many users it affects.

The same lineage supports the reverse question. When a dashboard shows a number that looks wrong, CoCo can trace the value upstream through the transformation layers and flag recent schema or data changes along the path.

For the lineage skill and its example prompts, see Lineage. For the full set of skills available to CoCo, see CoCo CLI bundled skills.

Getting started¶

Review what the catalog already knows about your Snowflake objects. Browse them in Snowsight and check the lineage graph for a table your team depends on.
Fill in the gaps outside Snowflake. Configure external lineage for pipelines that emit OpenLineage events, and ask your account team about metadata connectors for your external databases and BI tools.
Enrich what you collected. Generate descriptions for undocumented objects, and apply Snowflake-provided tags to record project, sensitivity, and certification status.
Define your business logic once. Create semantic views for the metrics your organization reports on, starting with the ones that are defined inconsistently today.
Put the context to work. Add semantic views as tools on an agent, and use Snowflake CoCo for lineage, impact analysis, and governance tasks.