Snowflake Semantic Views: A Complete 2026 Enterprise Guide

Q: How do I create a Snowflake semantic view?

Two methods: SQL DDL using CREATE SEMANTIC VIEW with five blocks (TABLES, RELATIONSHIPS, FACTS, DIMENSIONS, METRICS), or the Snowsight UI wizard for guided creation. Production best practice is to start with the 20% of data that answers 80% of questions, validate with subject matter experts, then scale. There is no automatic sync between existing table comments and semantic view definitions.

Q: How do Snowflake semantic views work with Cortex Analyst?

Cortex Analyst reads the semantic view's metadata including descriptions, synonyms, predefined join paths, and verified query examples, and uses it to generate SQL against physical tables. Snowflake reports approximately 85-90% accuracy on natural language queries against a well-defined semantic view. Cortex Analyst reads the view's metadata to inform SQL generation; it does not query the semantic view directly.

Emily Winks

Data Governance Expert

Updated:06/03/2026

Published:06/03/2026

16 min read

Watch Context Layer Live Get the Context Layer Ebook

Key takeaways

Snowflake semantic views are schema-level objects encoding TABLES, RELATIONSHIPS, FACTS, DIMENSIONS, and METRICS.
Standard SQL querying reached GA on March 2, 2026, removing the last major adoption barrier.
Semantic views are warehouse-bounded: they see only Snowflake data, not the 15-30 other systems in your stack.
Atlan's Enterprise Data Graph ingests semantic view definitions and extends them to 100+ cross-system sources.

What are Snowflake semantic views?

Snowflake semantic views are schema-level database objects that store business metrics, dimensions, and entity relationships directly in the warehouse. They power Cortex Analyst, CoWork, CoCo, and Cortex Sense by giving AI agents a governed vocabulary for business queries. Standard SQL querying reached GA on March 2, 2026. But semantic views are warehouse-bounded: they see only Snowflake data, not the 15-30 other systems in an enterprise stack.

The five DDL building blocks:

TABLES map logical entity names to physical tables and declare primary keys.
RELATIONSHIPS define explicit join paths that eliminate guesswork for AI query generation.
FACTS capture row-level quantitative attributes before aggregation.
DIMENSIONS provide categorical slicing context: product category, region, order date.
METRICS are business KPIs derived by aggregating facts, such as total revenue.

Is your data estate AI-agent ready?

Assess Your Readiness

Snowflake semantic views are schema-level objects that store business metrics, dimensions, and relationships directly in the warehouse. Standard SQL querying reached GA on March 2, 2026, and 83% of enterprise AI pilots never reach production (Atlan research, 2026). Semantic views are warehouse-bounded: Snowflake data only, not the 15-30 other systems in your stack. The context layer closes that gap.

Attribute	Detail
What It Is	Schema-level Snowflake database object storing metrics, dimensions, and entity relationships
Core DDL GA	Mid-2025
Standard SQL Querying GA	March 2, 2026
Five DDL Blocks	TABLES, RELATIONSHIPS, FACTS, DIMENSIONS, METRICS
Works With	Cortex Analyst (~85-90% NL accuracy), CoWork, CoCo, Cortex Sense, any SQL client
Key Limitation	Warehouse-bounded: no cross-system context, no cross-account replication, 1 MB YAML cap
Context Layer Relationship	Atlan’s Enterprise Data Graph ingests semantic view definitions and extends them across 100+ systems

Get the blueprint for implementing context graphs across your enterprise.

Get the Stack Guide

How Snowflake semantic views work

Snowflake semantic views are schema-level database objects that store business concepts directly inside Snowflake: how tables relate, which columns are metrics, which are dimensions, and what the business calls them. They act as a translation layer between physical table storage and the business language data teams and AI agents use.

AI agents querying raw SQL tables hallucinate business logic. “Total sales” means different things to different teams: retail only versus retail plus wholesale minus returns. Semantic views encode the authoritative definition once, inside the database, so every agent and analyst queries from the same governed source. As of Summit 2026, Snowflake CoWork, Snowflake CoCo, Cortex Analyst, and Snowflake Cortex Sense all depend on them.

The five DDL blocks

TABLES declare business entities (customers, orders, products) mapped to physical tables with primary key declarations. RELATIONSHIPS define explicit join paths (for example, orders(customer_id) REFERENCES customers), telling Cortex Analyst how to join tables rather than inferring it. FACTS capture row-level quantitative attributes at the most granular level before aggregation: individual sale amounts, session durations. DIMENSIONS provide slicing context: product category, customer region, order date. METRICS are business KPIs derived by aggregating facts, such as total revenue or average order value. Derived metrics reached GA in September 2025.

Summit 2026 context

Semantic View Autopilot reached GA on February 3, 2026, reducing semantic model creation from days to minutes. At Summit on June 2, 2026, Snowflake announced Advanced Semantics (level-of-detail calculations, composable definitions) and Semantic Studio (AI-assisted IDE with CoCo integration), both entering private preview. Snowflake Horizon Context is the overarching platform context layer; semantic views are its governed foundation. ThoughtSpot announced native Semantic Views integration (“Spotter Semantics”) on June 3, 2026 (ThoughtSpot, 2026).

As Christian Kleinerman, EVP of Product at Snowflake, stated at the Autopilot launch (Snowflake, February 2026): “AI is quickly becoming part of the operating fabric of the enterprise, not a side project.”

Snowflake builds the data plane. Atlan is the context plane. Semantic views are where those two planes connect.

How do Snowflake semantic views work?

You create a semantic view using a CREATE SEMANTIC VIEW DDL statement with five blocks, or via the Snowsight UI wizard. Since March 2, 2026, you can query it with either the SEMANTIC_VIEW() clause or standard SQL with AGG() functions, with no special connector required. Cortex Analyst reads the view’s metadata (descriptions, synonyms, predefined join paths, verified examples) to achieve approximately 85-90% accuracy on natural language queries (Snowflake, 2026).

Creating and querying a semantic view

CREATE SEMANTIC VIEW ecommerce_analysis
  TABLES (
    orders PRIMARY KEY (order_id),
    customers PRIMARY KEY (customer_id),
    products PRIMARY KEY (product_id)
  )
  RELATIONSHIPS (
    orders (customer_id) REFERENCES customers,
    orders (product_id) REFERENCES products
  )
  FACTS (
    orders COLUMNS (sale_amount, quantity_sold)
  )
  DIMENSIONS (
    customers COLUMNS (region, customer_segment),
    products COLUMNS (category, brand),
    orders COLUMNS (order_date)
  )
  METRICS (
    total_revenue AS SUM(orders.sale_amount),
    avg_order_value AS AVG(orders.sale_amount),
    total_orders AS COUNT(orders.order_id)
  );

Since March 2026, you can query the same view using either the semantic clause or standard SQL:

-- Semantic clause
SELECT * FROM SEMANTIC_VIEW(ecommerce_analysis
  METRICS total_revenue, avg_order_value
  DIMENSIONS customers.region, products.category
);

-- Standard SQL (GA March 2, 2026)
SELECT products.category, SUM(orders.sale_amount) AS total_revenue
FROM ecommerce_analysis
GROUP BY products.category;

One constraint to note: you cannot mix FACTS and METRICS in the same query clause.

How Cortex Analyst and CoWork use semantic views

Cortex Analyst reads the view’s metadata to generate SQL against physical tables. Descriptions, synonyms, and predefined join paths reduce AI agent hallucination substantially. CoWork uses semantic views as the governed foundation for conversational analysis; CoCo uses them to understand schema meaning, not just structure; Cortex Sense enriches agents at runtime using semantic view definitions, but only within Snowflake’s warehouse boundary.

Production note from phData (2026): start with the 20% of data that answers 80% of questions. Validate with subject matter experts before deployment. There is no automatic sync between existing table comments and semantic view definitions.

For how the context layer for Snowflake Cortex extends Cortex Analyst accuracy, the key is what becomes available once agents cross the warehouse boundary.

Snowflake semantic views vs the dbt semantic layer: what’s different?

The core split is architecture: Snowflake semantic views are a native database object, with zero additional cost, queryable by any SQL client, but Snowflake-only. The dbt semantic layer is a middleware proxy: multi-warehouse, accessed via MetricFlow API, approximately $100 per developer per month via dbt Cloud. According to Paradime’s 2026 technical comparison: “The semantic layer vs context layer landscape has shifted from a single-vendor story to a genuine architectural choice.”

Dimension	dbt Semantic Layer	Snowflake Semantic Views
Architecture	Middleware proxy via dbt Cloud	Native database object, no external infra
Warehouse scope	Multi-warehouse (Snowflake, BigQuery, Databricks, Redshift, Postgres, Trino)	Snowflake-only
Cost	~$100/developer/month (dbt Cloud required)	Zero additional cost (included with Snowflake)
Query access	Via MetricFlow API (JDBC/GraphQL)	Any SQL client, no connectors needed
GA status (SQL)	October 2023	March 2, 2026
Metric types	5 native types incl. conversion funnel, cumulative	Simple, derived, semi-additive (preview Mar 2026)
Version control	Native (YAML in Git)	Via dbt_semantic_view package or manual YAML
AI/NL querying	Limited native NL support	Cortex Analyst integration (~85-90% accuracy)
Cross-account	Yes	No: cannot replicate across accounts
Auto-time grains	Yes	Manual dimension definitions required

For Snowflake-only shops, semantic views win on simplicity and cost. Multi-cloud shops need dbt or a warehouse-agnostic layer. The dbt_semantic_view package (Snowflake Labs, Apache 2.0, October 2025) bridges both: define metrics in dbt YAML, publish to native Snowflake semantic objects. It is best-of-both for shops running both tools.

OSI (Open Semantic Interchange, launched September 2025 with 54 participating vendors) is Snowflake’s acknowledgment that semantic context needs to travel across tools. Atlan is a founding partner. As Prukalpa Sankar, Co-founder of Atlan, stated at the OSI launch (Atlan, September 23, 2025): “By joining OSI, we’re helping ensure that context travels across every tool, dashboard, and AI agent.”

For a broader view, best semantic layer tools covers where semantic views fit in the enterprise tooling landscape. The ontology layer above semantic views is where enterprise-wide entity resolution lives.

Snowflake builds the data plane; the semantic layer is the vocabulary that makes it legible. The enterprise context layer carries that vocabulary across every other system in your stack.

What Snowflake semantic views don’t cover

Semantic views are warehouse-bounded by design: they see only Snowflake data, not dbt models, Looker metrics, Salesforce objects, or Jira data. The YAML spec is capped at 1 MB with a recommended ceiling of 50-100 columns for Snowflake Cortex Analyst. Business Glossary is on Snowflake’s H2 2026 roadmap. Cross-system entity resolution and certified definitions traveling across 100+ systems require the enterprise context layer.

Technical caps to plan around

YAML specification: 1 MB size limit, 32K token constraint after removing sample values
Recommended maximum: 50-100 columns total across all tables
Cannot mix FACTS and METRICS in the same query clause
No native row-level filter declarations in DDL
No native cumulative or conversion metric types
No automatic time grains: manual dimension definitions required
Cannot be replicated across Snowflake accounts via standard replication

The cross-system gap

Enterprises do not run on Snowflake alone. According to Atlan research (2026), organizations operate across 15-30 tools in their data stack. The authoritative definition of “customer” lives in at least four different places (Snowflake, Salesforce, dbt models, a Confluence wiki), and they drift apart over time. Semantic views solve the problem inside the warehouse. Cross-system column-level lineage requires a dedicated unified context layer. See Snowflake data lineage best practices for what that full picture requires.

Snowflake builds the data plane; semantic views are the vocabulary on that plane. The enterprise context layer extends that vocabulary beyond the warehouse boundary.

How the enterprise context layer extends Snowflake semantic views

Atlan’s enterprise data graph ingests Snowflake semantic view definitions and connects them to 100+ cross-system sources with column-level lineage. Context agents auto-generate descriptions for the assets feeding those views: 690K+ generated, 87% rated on par with human writing across 50+ enterprise customers (Atlan, 2026). Context Engineering Studio validates semantic definitions before they reach agents. Context Lakehouse serves certified context to Snowflake agents via MCP server for Snowflake, including definitions that originated outside Snowflake. Snowflake = data plane. Atlan = context plane.

The four products that extend semantic views

Atlan Product	How it extends Snowflake semantic views
Enterprise Data Graph	Pulls semantic view definitions into a unified graph alongside 100+ other sources (dbt, Looker, Databricks, Tableau, Salesforce) with column-level lineage. 100+ connectors vs Horizon Context’s 5 in private preview Wave 1.
Context Agents	Auto-generates descriptions, metrics, and ontology for the assets feeding semantic views. 690K+ descriptions generated, 87% rated on par with human writing, 50+ enterprise customers (Atlan, 2026).
Context Engineering Studio	Bootstraps, tests, and versions the context semantic views depend on. CI-integrated eval suites validate definitions before they reach agents. One insurance customer compressed a 1-year context build to 1 month.
Context Lakehouse	Iceberg-native context store serving certified semantic definitions to Snowflake Cortex agents via MCP, including definitions that originated outside Snowflake.

Atlan AI Labs measured a 5x accuracy improvement in agents grounded in an enterprise context layer versus agents relying on semantic view definitions alone (Atlan AI Labs, 2026). For medium-complexity queries, that is a 2.15x improvement across 522 enterprise queries, with 38% of needed SQL accuracy gains coming from cross-system context that semantic views cannot carry.

Workday deployed Atlan MCP to expose shared business language to AI agents, solving the definitional ambiguity blocking a revenue analysis agent. As Jeff Miller at BlackRock Aladdin noted (Snowflake Horizon Context blog, 2026): “As AI becomes increasingly embedded across our enterprise, it’s essential that applications, analytics and agents operate from the same trusted understanding of the business.”

The context layer does not replace Snowflake semantic views. It completes them. Learn how the enterprise context layer and the semantic layer differ, or start with context layer implementation. The Snowflake-specific integration path is at context layer for Snowflake.

Snowflake builds the data plane. Atlan is the context plane: the governed infrastructure that delivers enterprise knowledge to every model, every agent, and every team from a single source of truth.

See how teams are building production-ready AI analysts with Atlan's Context Studio.

Save your Spot

Real stories from real customers: semantic layers in production

Workday: Shared business language at enterprise scale

"We're excited to build the future of AI governance with Atlan. All of the work that we did to get to a shared language at Workday can be leveraged by AI via Atlan's MCP server...as part of Atlan's AI Labs, we're co-building the semantic layer that AI needs with new constructs, like context products."

Joe DosSantos, VP of Enterprise Data & Analytics

Workday

See how Workday leverages context as a foundation for AI

Watch Now

DigiKey: Context as a company-wide operating system

"Atlan is much more than a catalog of catalogs. It's more of a context operating system...Atlan enabled us to easily activate metadata for everything from discovery in the marketplace to AI governance to data quality to an MCP server delivering context to AI models."

Sridher Arumugham, Chief Data & Analytics Officer

DigiKey

See how DigiKey activated context across their full data estate

Watch Now

Semantic views are infrastructure: what your team builds on top matters

Snowflake made an unambiguous architectural bet at Summit 2026: semantic views are the load-bearing primitive under every AI capability they ship. Teams that have not invested in a semantic foundation at the warehouse layer are now behind every Snowflake-native AI feature release. Context engineering for AI agents is the methodology that carries those investments into production.

But warehouse-level semantic consistency is necessary, not sufficient. According to Atlan research (2026), 83% of enterprise AI pilots never reach production. The 17% that do make it are the ones who solved the cross-system context problem, not just the within-warehouse problem. Semantic views get you to consistent definitions inside Snowflake. The context layer for Snowflake gets you to consistent definitions across every system your agents need to reason about.

The path from “semantic views defined” to “production-ready context layer for AI agents” runs through what is context engineering: the methodology that treats enterprise knowledge as infrastructure, version-controlled and validated just like code.

Snowflake builds the data plane. Atlan is the context plane. Semantic views are the connection point between them.

Book a demo

FAQs about Snowflake semantic views

1. What are Snowflake semantic views?

Snowflake semantic views are schema-level database objects that store business metrics, dimensions, and entity relationships directly in the database: a translation layer between raw table storage and the business language data teams and AI agents use. Core DDL reached GA mid-2025. Standard SQL querying reached GA on March 2, 2026. They are the foundational primitive under Cortex Analyst, CoWork, CoCo, and Cortex Sense as of Summit 2026.

2. How do I create a Snowflake semantic view?

Two methods: (1) SQL DDL using CREATE SEMANTIC VIEW with five blocks (TABLES, RELATIONSHIPS, FACTS, DIMENSIONS, METRICS); (2) the Snowsight UI wizard for guided creation. Production best practice (phData, 2026): start with the 20% of data that answers 80% of questions, validate with subject matter experts, then scale. There is no automatic sync between existing table comments and semantic view definitions.

3. How do Snowflake semantic views work with Cortex Analyst?

Cortex Analyst reads the semantic view’s metadata (descriptions, synonyms, predefined join paths, verified query examples) and uses it to generate SQL against physical tables. Snowflake reports approximately 85-90% accuracy on natural language queries against a well-defined semantic view (Snowflake, 2026). Cortex Analyst reads the view’s metadata to inform SQL generation; it does not query the semantic view directly.

4. What is the difference between the dbt semantic layer and Snowflake semantic views?

Architecture is the core difference. Snowflake semantic views are a native database object: zero additional cost, any SQL client, Snowflake-only. The dbt semantic layer is a multi-warehouse middleware proxy, approximately $100 per developer per month via dbt Cloud, accessed via MetricFlow. The dbt_semantic_view package (Apache 2.0, October 2025) enables a hybrid approach: define in dbt, publish to native Snowflake semantic objects.

5. What are the limitations of Snowflake semantic views?

Semantic views are warehouse-bounded: Snowflake data only, no cross-system context. Key constraints: 1 MB YAML spec limit, 50-100 column ceiling for Cortex Analyst, no cross-account replication, no native conversion or cumulative metric types, no automatic time grains, no Business Glossary (H2 2026 roadmap). The enterprise context layer closes the cross-system gap by extending definitions across 100+ systems.

6. What is Semantic View Autopilot?

Semantic View Autopilot reached GA on February 3, 2026. It analyzes existing Snowflake tables and auto-generates semantic view DDL, mapping entities, inferring relationships, and suggesting metrics. It reduces semantic model creation from days to minutes. Early adopters include eSentire, HiBob, Simon AI, and VTS (Snowflake, February 2026). Advanced Semantics and Semantic Studio entered private preview at Summit 2026 as the next evolution.

Sources

Snowflake documentation. “Semantic Views Overview.” docs.snowflake.com/en/user-guide/views-semantic/overview. Accessed June 2026.
Snowflake. “Snowflake Launches Semantic View Autopilot, Making Enterprise AI Analytics Accessible.” Press release, February 3, 2026.
Snowflake. “Horizon Context: Building Governed Context for the AI Enterprise.” snowflake.com/en/blog/horizon-context-governed-context/. June 2026.
Paradime. “dbt Semantic Layer vs Snowflake Semantic Views: A Complete Technical Comparison.” paradime.io/blog/dbt-semantic-layer-vs-snowflake-semantic-views-a-complete-technical-comparison. 2026.
phData. “Snowflake Semantic Views in Production: Real-World Insights and Best Practices.” 2026.
Colrows. “Why Snowflake and Databricks Can’t Be Your Enterprise Semantic Layer.” January 2026.
InfoWorld. “Snowflake’s Horizon Context aims to give AI agents a common understanding of the business.” June 2026.
ThoughtSpot. “ThoughtSpot Expands Governed Enterprise AI with Snowflake Cortex AI and Semantic Views.” June 3, 2026.
AtScale. “AtScale Announces Snowflake Semantic Views XMLA Endpoint.” June 2, 2026.
Atlan AI Labs. “Context Layer Accuracy Benchmarks: 522 Enterprise Queries.” Internal research, 2026.
Atlan. “83% of Enterprise AI Pilots Never Reach Production.” Atlan Research, 2026.
Atlan. “OSI: Open Semantic Interchange Launch Partners.” September 23, 2025.
Constellation Research. Miller, Mike. “Snowflake Summit 2026: Context, custom model training, Iceberg V3.” June 2026.

Share this article

Atlan is the next-generation platform for data and AI governance. It is a control plane that stitches together a business's disparate data infrastructure, cataloging and enriching data with business context and security.

Book a Demo Context Studio Live