What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
To give AI a reliable business view of Snowflake data, define its meaning and relationships explicitly in a Semantic View. Map business entities to tables and columns, specify how they join, and encode dimensions and metric formulas. A Cortex Agent can then use that governed context to generate SQL for structured questions. Snowflake does not offer a separate first-party product literally called an ontology; Semantic Views provide the business-meaning layer this pattern needs.
What “ontology” means in Snowflake
An ontology here is a model of business concepts and how they relate—not a feature that makes an AI independently understand an organization. A Semantic View is a schema-level Snowflake object that maps physical data to logical business concepts an AI application can use. It can define logical tables, relationships, dimensions, facts, metrics, descriptions, and synonyms. Snowflake’s Semantic View overview describes this structure and its use in analytics.
As an Amazon Associate I earn from qualifying purchases.
For example, a physical column named cust_nm may be exposed as “Customer name,” described in business language, and given synonyms such as “customer” or “account name.” A defined relationship tells SQL generation how a customer connects to orders. A metric can encode the organization’s chosen calculation for “Net Revenue” rather than leaving the model to infer a formula from column names.
The pieces of the model
- Logical tables: business entities such as customers, orders, or suppliers, mapped to underlying Snowflake tables or expressions.
- Relationships: declared connections between entities, including the keys needed to join them.
- Dimensions: attributes used to filter or group results, such as market segment, region, or order date.
- Facts: row-level measures, such as an order amount.
- Metrics: defined aggregations or calculations built from facts, such as order count, total order value, or average order value.
What Semantic Views can—and cannot—do for AI
Semantic Views give an AI application explicit names, descriptions, join paths, metric definitions, and examples to use when turning a question into analytics SQL. Snowflake describes the approach as combining LLM reasoning with rule-based definitions. That can reduce ambiguity—for instance, by giving “revenue” one governed formula—but it does not establish a particular accuracy rate for your workload or guarantee that every generated query is correct. Snowflake’s overview and Cortex Analyst documentation describe the relevant mechanisms.
#1 Best Overall
This pattern is best suited to structured questions answerable with SQL, such as “What is the average order value by market segment?” It is not a substitute for evaluation, human review, or a system designed to retrieve unstructured documents. The standalone Cortex Analyst documentation also describes limits around broad business-insight prompts and using a prior query’s result as a value in a later question. Because Snowflake’s product direction now points new integrations toward Cortex Agents, check the current Cortex Agents documentation before applying those Analyst-specific limits to an Agent workflow.
Plan the business model before writing SQL
Start with questions the AI should answer, not with every table in the warehouse. Snowflake recommends beginning with a simple star schema where appropriate. A small, well-defined model is easier to validate than a broad semantic layer whose metrics, joins, and business terminology are unclear. Snowflake’s Semantic View best-practice index covers modeling guidance.
- Choose representative questions. Write down the actual questions users will ask, such as “What is the average order value by market segment?” or “What is the month-over-month revenue growth for 2021 in Asia?”
- Identify entities and grain. Decide what each logical table represents and what one row means. List the entities needed for those questions, such as customers and orders.
- Map connections. Record the keys and relationships between entities. Check that the intended join path produces the right result without unintended duplication.
- Define dimensions and facts. Identify the fields users should group or filter by and the row-level measures from which metrics are calculated.
- Set metric rules. Write each KPI’s formula and aggregation explicitly. Clarify edge cases—such as which amounts count toward net revenue—according to your organization’s definitions.
- Use business language. Add useful descriptions and synonyms that match how analysts and users phrase questions, rather than relying on cryptic physical names.
Create and validate a Semantic View
Snowflake supports authoring through Semantic Studio, SQL, and guided Snowsight workflows; YAML can also be used as an authoring format. These are ways to create and manage the semantic layer, not separate products with equivalent strategic status. For new implementations, Snowflake recommends native Semantic Views. Its overview and YAML specification explain the options.
Free tools Windows power users keep installed
One-click scans. No signup required.
- Build the object from your model. Define logical tables, their columns and business descriptions, relationships, dimensions, facts, and metric formulas. Add synonyms for important terms people use interchangeably.
- Query it directly. Test the Semantic View against known data and expected answers. Confirm the joins, filters, and aggregations behave as intended. Snowflake’s Cortex Analyst getting-started guide advises fixing a failing view before using it with an agent.
- Connect it to an agent. Add the Semantic View as a Cortex Analyst tool in a Cortex Agent workflow, then test representative natural-language questions and inspect both generated SQL and results.
- Improve from evidence. Add verified question-and-SQL pairs, refine descriptions and custom instructions, and evaluate known cases as the model changes. Treat the semantic definition as maintained data infrastructure, with an owner and a deployment process.
Choose native Semantic Views or keep a legacy YAML model
Snowflake’s native Semantic Views are its recommended path for new work. Stage-based semantic-model YAML remains a compatibility option for existing workflows, so a migration is not automatically required. The relevant differences are about how the model is managed and used:
| Approach | Management and access | Querying and integration | When it fits |
|---|---|---|---|
| Semantic View | Schema-level object integrated with Snowflake privileges, sharing, and catalog features. | Can be queried directly; Snowflake positions it as the preferred approach for new implementations. | New models or teams that want native schema-object management. |
| Stage-based semantic YAML | Model is maintained as YAML on a stage rather than as a native schema-level Semantic View. | Remains supported for backward compatibility with existing workflows; native-object capabilities such as schema-level management do not apply in the same way. | Existing clients or workflows that depend on the legacy format. |
See Snowflake’s Semantic View overview and YAML specification for implementation details.
Use Cortex Agents for the current application direction
In its August 28, 2026 release note, Snowflake recommended transitioning from standalone Cortex Analyst to Cortex Agents. This is product-direction guidance, not a statement that Analyst has been removed: Snowflake said existing applications continue to work and the Cortex Analyst REST API remains available. Semantic Views and verified queries carry over unchanged.
| Option | Role in an application | Relevant capabilities and migration context |
|---|---|---|
| Standalone Cortex Analyst | Invokes natural-language analytics over structured data. | The REST API remains available, according to Snowflake’s August 28, 2026 release note. |
| Cortex Agents | Provides the current workflow for combining the semantic layer with tools and orchestration. | Snowflake’s release note describes retrieval over unstructured data with Cortex Search, tool calling, conversational threads, and multistep orchestration. Semantic Views and verified queries carry over from Analyst. |
Use an Agent when the application needs those additional capabilities; a structured-data question still relies on the Semantic View as its governed foundation. For current setup details, refer to Snowflake’s Cortex Agents documentation.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Build governance and evaluation into the workflow
Semantic Views are schema objects that use Snowflake’s privilege system. Cortex Analyst documentation says generated and executed SQL adheres to established role-based access controls, but the actual outcome depends on your account’s roles, grants, and policies. Validate the access model with the same identities and permissions the application will use; do not assume that semantic metadata itself grants data access. See the Semantic View overview and Cortex Analyst guide.
Quick Recap
- Test known questions against expected SQL and results, including joins and aggregation edge cases.
- Review generated SQL before relying on answers in consequential reporting.
- Maintain verified query examples for recurring or business-critical questions.
- Assign ownership for metric definitions, access reviews, and changes to underlying tables.
- Re-run evaluations when the semantic model, source data, or application behavior changes.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
