At a glance
- Create a semantic view with CREATE SEMANTIC VIEW, with Semantic View Autopilot in Snowsight, or from a YAML specification.
- Cortex Analyst and Cortex Agents read semantic views to write SQL, and BI users get the same metrics and dimensions.
- A semantic view covers only the subjects someone modelled, so agents also need the rest of the warehouse mapped, and a role and a log of their own.
What is a Snowflake semantic view?
Snowflake’s documentation describes a semantic view as a schema-level object that stores business meaning alongside physical data. It holds logical tables for business entities such as customers and orders, the relationships between them, facts (row-level numbers such as a sales amount), dimensions to group and filter by, and metrics such as total revenue or margin. Because it is a Snowflake object, Snowflake roles govern it.
How do you create a Snowflake semantic view?
You can write a CREATE SEMANTIC VIEW statement that names the logical tables, relationships, facts, dimensions and metrics. Semantic View Autopilot in Snowsight drafts a view for you and can import a YAML specification, a Tableau file or a Power BI file, and you can also author a view in Semantic Studio in Workspaces. Whichever route you take, start with one subject, such as orders, define the joins on shared keys, add three or four metrics people already argue about, then test them against the report finance trusts.
How does a semantic view work at query time?
Cortex Agents and Cortex Analyst read the definition and generate SQL against the physical tables, so a question about revenue uses the metric you defined. BI users can query the view with a SELECT and get the same metrics and dimensions. The definition is only as current as the last person who edited it, so a renamed column or a new table is a change someone has to make.
What does a semantic view leave out?
A semantic view covers the subjects someone modelled. Questions that cross into tables outside it, or into a second warehouse, fall back to guessing. It also sits inside Snowflake. Claude, Cursor or ChatGPT connected to the warehouse do not inherit it, and an agent still needs a role, a call limit and a log of its own.
Why do AI agents need more than a semantic view?
An agent asks the question nobody wrote a metric for: which accounts churned after a price change, or where an order went between three systems. To answer that it needs table and column meanings, join paths and business terms across the warehouse, which is the context layer. It also needs governed access, so it reads only what its role allows and every query is traceable.
How Rig does it
Rig Map holds the ontology, the semantic layer and the context layer for your warehouse, and keeps them current as schemas change. It drafts table descriptions, joins and metrics from the schema and the queries your team runs, and your data owners certify them. Existing LookML and Lightdash metrics can be imported, and Rig stores metrics in the Open Semantic Interchange format so you can export them. Gateway then serves those definitions to Claude, Cursor and your agents over MCP, with a role per person and per agent and a named audit trail.
Common questions
How do I create a semantic view in Snowflake?
Use the CREATE SEMANTIC VIEW SQL command, Semantic View Autopilot in Snowsight, or Semantic Studio in Workspaces. Autopilot can also start from a YAML specification, a Tableau file or a Power BI file.
What do Snowflake semantic views cover?
Logical tables, relationships, facts, dimensions and metrics for the subjects you model. Cortex Analyst, Cortex Agents and BI tools read them.
Do I still need a semantic layer if I use Rig?
Rig Map includes one. You can keep your Snowflake semantic views for Cortex and use Rig to serve certified metrics, the wider context and governed access to the other AI tools your team uses.
Does Rig work with Snowflake semantic views?
Rig reads your Snowflake schema and the queries your team runs, and drafts metrics for your data owners to certify. Ask us about bringing existing semantic view definitions across on a call, since the route depends on how they were built.
Sources and product details
Vendor information checked on 20 November 2026. Confirm current editions, licence terms and prices with the vendor before buying.