Snowflake Horizon Context Defined
Snowflake Horizon Context is the governed semantic and context layer Snowflake announced at Summit 2026 as the foundation for its AI products. Snowflake describes it as "a new capability within Horizon Catalog that offers a connected, governed semantic foundation with active context for AI and BI."
Translated into plain terms, Horizon Context is where Snowflake stores governed business meaning (what "revenue" means, how "active customer" is calculated, which dimensions matter for which report) and exposes that meaning to every consumer of Snowflake data, including CoCo, CoWork, and any application that queries Snowflake directly.
What it is: Snowflake's governed semantic and context layer, built on top of Horizon Catalog. It collects metadata, enriches it with business meaning, and activates that meaning for AI and BI consumers.
What it's not: A BI tool. Horizon Context defines and governs meaning, but it doesn't render that meaning in the tools the business uses every day.
How Horizon Context Fits With Horizon Catalog and Semantic Views
Snowflake's architecture stacks like this:
- 1
Horizon Catalog holds the metadata: tables, columns, lineage, ownership, tags.
- 2
Horizon Context turns that metadata into governed business meaning, with metric definitions, dimensions, relationships, and the rules that govern who can see what.
- 3
Snowflake Semantic Views are the native construct inside Horizon Context where the metrics, dimensions, and relationships are defined and stored.
- 4
CoCo uses Horizon Context to answer questions.
- 5
CoWork uses Horizon Context to take action.
When a user asks CoCo about gross margin, CoCo pulls the definition from Horizon Context, queries the underlying tables through Snowflake Semantic Views, and returns an answer grounded in the data team's governed logic. When CoWork executes a multi-step task involving the same metric, it works from the same definition rather than reinventing one along the way.
What Horizon Context Solves
Horizon Context closes the meaning gap inside Snowflake, which is a real and meaningful step forward.
Single source of truth for governed definitions. Metric definitions live in one place, versioned, audited, and tied to the data they describe.
Governance applied at the meaning layer. Row-level masking, role-based access, and column-level controls follow the definition, not the query path.
Native consumption for Snowflake's AI surfaces. CoCo and CoWork inherit governed definitions automatically, with no separate integration work.
Lineage and traceability. Every Horizon Context definition can be traced back to the underlying tables and columns, which gives audit and compliance teams the evidence they need.
Where Horizon Context Stops
Horizon Context governs meaning inside Snowflake. Most enterprises don't run the business inside Snowflake. They run it in Power BI, Excel, Tableau, Looker, ThoughtSpot, and a growing set of external AI agents (Claude, ChatGPT, Cursor, custom agents). Horizon Context's definitions don't natively reach any of those tools.
"Snowflake Semantic Views centralize governed metric definitions inside Snowflake, but Microsoft tools like Power BI and Excel do not natively consume those definitions."
The result is that the metric drift Horizon Context was built to eliminate reappears one layer out, between Snowflake's AI surface and the BI tools the rest of the business actually runs on. CoCo and Power BI can disagree on the same question, and the reporting cycle ends up spent reconciling them.
How to Extend Horizon Context to Every Tool the Business Uses
A universal semantic layer like AtScale for Snowflake reads Horizon Context's governed definitions and serves them to Power BI, Excel, Tableau, Looker, Google Sheets, and external AI agents through standard interfaces. The architecture splits cleanly:
Horizon Context is the system of record for governed definitions inside Snowflake.
AtScale is the system of consumption that extends those definitions across every BI tool and AI agent the enterprise uses, with Snowflake RBAC passthrough, adaptive aggregate caching, and no data movement.
Together they form what Snowflake calls a composite context layer: governed meaning that travels with the definition rather than being a feature of any single platform.
What the Anthropic Benchmark Says About the Value of a Governed Context Layer
The accuracy case for governed context stopped being theoretical when Anthropic published its own numbers. In How Anthropic's AI Accuracy Went from 21% to 95%, AtScale CTO Dave Mariani walks through two benchmarks worth working through.
Claude given raw access to SQL files, dashboards, transforms, and notebooks answered correctly 21% of the time. After a semantic layer went in and Claude was told to consult it first, accuracy moved to ~95%, with some domains near 99%.
AtScale's production benchmark cut compute by up to 21,000× and lifted accuracy from ~70% to 100% on five common questions — with ~$9M projected annual savings on that question set.
Horizon Context delivers a meaningful piece of that benefit inside the Snowflake perimeter, and a universal semantic layer carries the same accuracy and cost profile out to every tool and agent that asks the same question elsewhere.
Frequently Asked Questions
What's the difference between Horizon Catalog and Horizon Context?
Horizon Catalog holds the metadata about Snowflake objects (tables, columns, lineage, ownership). Horizon Context sits on top of Horizon Catalog and turns that metadata into governed business meaning, with metric definitions, dimensions, and relationships that AI and BI consumers can use directly.
Are Snowflake Semantic Views the same as Horizon Context?
Snowflake Semantic Views are the native construct inside Horizon Context where governed metrics, dimensions, and relationships are defined. Horizon Context is the broader semantic and context layer that includes Semantic Views along with governance, lineage, and the active context that CoCo and CoWork consume.
Does Horizon Context replace the need for a separate semantic layer?
Inside Snowflake, Horizon Context is the semantic layer. Outside Snowflake, in Power BI, Excel, Tableau, and external AI agents, a universal semantic layer like AtScale is what extends those same governed definitions to the tools the business actually runs on.
Can Power BI read Horizon Context directly?
Power BI doesn't natively read Snowflake Semantic Views or Horizon Context definitions. A universal semantic layer is the bridge that exposes governed Snowflake definitions to Power BI through XMLA, MDX, and DAX.
Is Horizon Context part of the Open Semantic Interchange standard?
Snowflake leads the Open Semantic Interchange initiative, with 54 participating organizations including AtScale, dbt Labs, and ThoughtSpot. The standard is intended to make semantic definitions portable across vendors, so a definition governed in Horizon Context can be consumed by tools that speak the standard.
Key Takeaways
Snowflake Horizon Context is the governed semantic and context layer built on top of Horizon Catalog, where Snowflake stores metric definitions, dimensions, relationships, and the governance that goes with them.
Horizon Context is what makes CoCo and CoWork trustworthy inside Snowflake, with one definition feeding both products and audit-grade lineage on every answer.
The gap that remains is outside Snowflake, in Power BI, Excel, Tableau, and external AI agents, and a universal semantic layer like AtScale is what extends Horizon Context to those tools without data movement or rebuilt logic.