Snowflake Data Catalog: What Horizon Covers and What It Misses

Yes, Snowflake has a data catalog. It is called Horizon Catalog, and it covers discovery, tagging, classification, access policies, and column-level lineage without buying anything extra. Column-level lineage requires Enterprise Edition or higher, retains one year of history, and captures nothing before November 2024. The boundary is Snowflake itself: the moment a dashboard, an ingestion tool, or a second warehouse enters the picture, Horizon stops seeing the path.

This question comes up in almost every Snowflake catalog evaluation, usually phrased as whether the team needs to buy anything at all. The honest answer for a lot of single-platform teams is no. Below is what Horizon actually includes, checked against Snowflake's own documentation, where it stops, and how to tell which side of that line you are on.

What is Snowflake Horizon Catalog?

Horizon Catalog is Snowflake's governance and discovery layer, described in its documentation as the agentic catalog for all your data, whether that data sits inside or outside Snowflake. The useful way to think about it is as an umbrella brand over capabilities that are each documented separately, rather than a product you install or switch on. Under that umbrella you get object tagging, sensitive data classification, masking and row access policies, role-based access control, the Trust Center, data metric functions for quality, semantic views, and lineage.

That framing is not a criticism. It means there is no deployment project, no agent to run, and no separate contract. If you already use Snowflake, most of Horizon is configuration rather than procurement. It also means the marketing name and the feature set do not map one to one, so when a vendor comparison says a team already has Horizon, the more precise question is which Horizon features they have actually enabled and which edition they are on.

Does Snowflake have column-level lineage?

It does. Column-level lineage went generally available in the 9.3 release in February 2025 after a preview that started in late 2024. In Snowsight you can trace the relationship between columns in a source object and columns in a target object, walking the graph in either direction.

Three constraints decide whether that is enough for you:

Process lineage covering stored procedures and tasks reached general availability in October 2025, which closed a real gap: work orchestrated inside Snowflake now shows up rather than appearing as an unexplained jump in the graph.

Is Horizon Catalog free?

Horizon itself is not a billable line item. Search Snowflake's service consumption table and you will not find an entry for Horizon or for data lineage, because lineage, tagging, masking, and access history are edition features rather than metered services. What you pay for is Enterprise Edition, which you were likely paying for anyway.

Some components under the umbrella do meter serverless compute. Data quality monitoring runs at a two times compute multiplier, the Trust Center at one times, and sensitive data classification at zero point nine. Those are consumption charges that scale with how many objects you monitor and how often, so a broad quality program has a cost shape closer to a warehouse bill than a license. We worked through the specifics of that in Snowflake data quality checks, including why you are billed for a scheduled data metric function but not for creating one.

Snowflake Horizon, Open Catalog, and Polaris are not the same thing

These three names get used interchangeably in blog posts and they refer to genuinely different products. Getting them mixed up will send an evaluation in the wrong direction, so here is the clean version.

The practical takeaway: if someone recommends Open Catalog as the way to catalog your Iceberg tables, that advice is now out of date. New Snowflake customers cannot create an Open Catalog account, and Horizon is where Iceberg cataloging lands.

Where Snowflake's native catalog stops

Every warehouse-native catalog shares one limitation, and it is structural rather than a missing feature. Horizon knows what happens inside Snowflake. It does not know what happens on either side of it.

Consider a path that is completely ordinary. Fivetran lands a Postgres table into a raw schema. dbt builds three models on top of it. A Looker explore reads the last one, and an executive dashboard reads the explore. Horizon sees the middle of that path clearly and can trace columns through the dbt models. It cannot tell you that the Postgres source column was renamed at the origin, and it cannot tell you that the dashboard the CFO opens every Monday depends on the column you are about to drop. Those are exactly the two ends where trust questions get asked.

The second gap is multi-platform reality. Plenty of organizations run Snowflake alongside BigQuery or Databricks, usually because of an acquisition or a team that chose differently. A native catalog gives you one accurate map per platform and no way to join them, which means the person answering a question about data provenance still ends up assembling the story by hand. This is a general pattern with tools scoped to one system, and it is the same reason teams eventually reach for something that searches across every place work lives rather than one silo, whether that is a warehouse or the sprawl of documents and apps the rest of the company runs on.

The third is retention. One year of lineage is enough for operational work and short of what a regulated audit sometimes asks for.

When do you need a third-party data catalog for Snowflake?

A short test that usually settles it. If all four of these are true, use Horizon and buy nothing:

• Snowflake is your only warehouse, and it is where analytics actually happens.
• You are on Enterprise Edition or higher, so lineage is available to you.
• Nobody needs lineage that reaches into the BI layer to answer impact questions.
• One year of history covers your audit and investigation needs.

If any one of those is false, the gap is real and worth pricing. The most common failing condition by a wide margin is the third. Teams discover it the first time somebody drops a column that looked unused in Snowflake and breaks a dashboard nobody had mapped, because the dependency lived in Looker or Power BI where Horizon does not look.

Worth knowing before you shortlist: Snowflake announced its acquisition of Select Star in November 2025 and is folding it into Horizon, so the native option is getting broader. If you are evaluating Select Star today, that direction is either reassuring or a lock-in question depending on whether you plan to stay single-platform. The full landscape, including which vendors publish real prices and which open-source projects are still maintained, is in our data catalog tools comparison.

How Datatrail fits alongside Horizon

We are not a replacement for Horizon and it would be dishonest to pitch it that way. Horizon does access policies, classification, and masking, and those belong in the platform. What we add is the part that stops at the Snowflake boundary.

Datatrail connects with a read-only role, parses query history alongside your dbt manifest, and builds a column-level lineage graph that continues past the warehouse into the dbt project and the BI layer, then keeps a catalog derived from that graph rather than curated by hand. The payoff is the question Horizon cannot answer: before you drop or retype a column, which models, exposures, and dashboards break. If you are multi-warehouse, the same graph spans them, so provenance is one map instead of three. The Snowflake lineage page covers exactly how the connection and parsing work.

For a single-platform Snowflake team on Enterprise Edition whose analysts live in Snowsight, Horizon is genuinely sufficient and we would tell you so. The moment the path crosses a boundary, it is not, and that is the whole decision.