Snowflake Horizon Catalog

Overview

Horizon Catalog is governance built into Snowflake rather than a product you buy for your wider estate, and that scope boundary is the first thing to establish when comparing it with an independent platform.

It is anchored to a Snowflake account. It reaches beyond Snowflake only through Snowflake-mediated paths: Iceberg tables, the Iceberg REST catalog API, external tables, catalog-linked databases and external lineage.

There is no Horizon licence and no separate purchase. Instead, access to each capability is gated by your Snowflake edition, then billed as consumption when a feature runs compute.

That edition gate is the real commercial story and it is stricter than summaries suggest. Masking, row access policies, classification, data quality monitoring and lineage all need Enterprise edition or higher.

Only basic tag creation works on Standard, so anyone told that Snowflake includes governance should check which edition they are actually on before planning around it.

What it does better than any independent platform is enforce. Masking policies and row filters are evaluated by the engine itself at query time, with nothing pushed to a third party and nothing merely documented.

Openness is genuine rather than rhetorical, built on Apache Polaris with Iceberg REST and scan APIs so Spark, Flink and Trino can read the same governed tables bidirectionally.

Discovery and sharing are unified through an internal marketplace, where teams publish organisational listings and consumers request access with an approval workflow that reached general availability in 2025.

Lineage is column-level and includes external lineage, tracking how data flows in from outside tools, which is used for root-cause analysis in quality monitoring.

The 2026 direction is AI: semantic views and a semantic studio, guardrails detecting and redacting personal data, agent identity, AI security posture management, and support for an open semantic interchange standard.

Key features and capabilities

The same headings are used for every governance & stewardship entry, so two tools can be read side by side.

What it catalogues

• Catalogues Snowflake tables, dynamic and external tables, Iceberg tables, views, semantic views and ML objects
• Metadata is largely intrinsic, since objects are catalogued because they exist in the account
• Catalog-linked databases and the Iceberg REST API extend reach to external catalogues and lakes
• Auto-generated table and column descriptions, plus object tagging for consistent metadata
• BI and application coverage is the thinnest here; there is no published scanner list for Tableau or SAP

Ownership and stewardship

• An internal marketplace shares data products within the organisation as organisational listings
• Ownership runs through Snowflake's own role-based access control rather than a separate steward model
• Only roles holding ownership or modify privileges on a listing can approve access requests
• Semantic views carry business definitions so people and AI agents share the same meaning
• Data domains as a first-class object are not published, unlike its independent competitors

Glossary, policy and classification

• Object tagging is the classification backbone, with published limits of 50 tags on an object
• Sensitive data classification auto-tags columns, with an AI mode that adds token charges on top of compute
• Policies are genuinely enforced by the engine at query time, through masking and row access policies
• Tag-based masking policies apply protection by classification rather than column by column
• A business glossary product and sensitivity labels in the Microsoft sense are not published; tags fill that role

Lineage and impact

• End-to-end column-level lineage, tracing upstream and downstream column relationships
• Coverage spans tables, dynamic and external tables, Iceberg tables, views, stages and ML objects
• External lineage tracks flows from outside tools and databases into Snowflake objects
• Lineage drives root-cause analysis in data quality monitoring and traces AI answers back to source
• Impact analysis is delivered through the lineage graph rather than as a separate named feature

Workflow and collaboration

• A request and approval workflow for marketplace listings, generally available since May 2025
• Consumers click request access and give a reason; the decision is emailed to them
• Providers can manage requests in Snowflake or delegate to an external address or URL
• Automation through scheduled data metric functions and scheduled security scanners
• General workflow design, issue management and a data helpdesk are not published

Integrations

• The Iceberg REST catalog API with published support for Spark, Flink and Trino, and a scan API
• Catalog-linked databases and external tables reach external catalogues, lakes and relational databases
• Support for an open semantic interchange standard so definitions reach external agents and BI tools
• Interoperability runs both ways, with Collibra enforcing natively here and Purview scanning it
• A dedicated management REST API is not published; governance is configured in SQL and the interface

Where it runs and what it costs

• Software as a service only, on AWS across 24 regions, Azure across 18 and Google Cloud across 8
• Government deployments on AWS GovCloud supporting high-level federal authorisation, and Azure Government
• A fully isolated private environment is available as a separate edition
• Cross-region replication with failover, the latter needing Enterprise edition or higher
• Billed in credits for compute plus storage per month after compression

Pricing

Usage-basedIncluded with Snowflake; Enterprise edition gates most features

There is no Horizon licence, but nor is it published as free. Almost everything that makes it a governance product needs Enterprise edition or higher: masking, row access policies, classification, quality monitoring, lineage and tag propagation. Only basic tagging works on Standard, and enhanced protection for health data needs Business Critical. On top, quality monitoring and classification consume serverless credits when scheduled, with the AI classification mode adding token charges. No credit or edition price appears on the pages read.

Vendor pricing page →

Demos and videos

• Horizon demo for data governors and stewards Product demo · YouTube
• AI data governance and interoperability with Horizon Catalog Webinar or talk · YouTube

About Snowflake

Horizon Catalog is a feature set of the Snowflake platform rather than a separate company or product, from Snowflake Inc., founded in 2012 by Benoît Dageville and Thierry Cruanes and publicly traded as SNOW. It reports over 10,000 staff across more than 40 offices, 12,062 customers and 828 of them spending more than a million dollars a year. It remains independent, with no ownership change found for 2025 or 2026. The product was called Snowflake Horizon at its 2024 launch and is now consistently Horizon Catalog, repositioned in June 2026 around semantic context, agent identity and Iceberg interoperability.

Founded 2012 · snowflake.com