Enable Snowflake Cortex AI for Solid

Overview

Snowflake Cortex AI is a suite of serverless AI functions built into Snowflake — including LLMs, semantic search, and sentiment analysis — that run inside Snowflake's compute boundary and bill against your Snowflake credits.

Enabling Cortex AI for Solid unlocks two specific integration scenarios. It is not required for Solid's core text-to-SQL and semantic modeling functionality, which runs on Solid's own engine independently of Cortex.

When to Enable Cortex AI

Scenario 1: Export to Snowflake Semantic Views for Cortex Analyst

If you use Snowflake Cortex Analyst — Snowflake's own natural-language query interface — you can export Solid's certified semantic models as Snowflake Semantic Views. This lets Cortex Analyst benefit from the definitions, metrics, and relationships Solid has already built and tested, instead of starting from scratch.

With both enabled, Solid can also benchmark Solid's text-to-SQL engine against Cortex Analyst on the same semantic model — running the same benchmark questions through both systems so you can compare accuracy and understand where each performs better.

Scenario 2: Cortex AI functions within Solid workflows

Some Solid features can optionally invoke Cortex AI functions (e.g., LLM-powered enrichment tasks) that run inside Snowflake's compute rather than Solid's. This keeps compute and data inside Snowflake's boundary for those operations.

How Solid's Text2SQL Engine Compares to Cortex Analyst

Solid and Cortex Analyst approach text-to-SQL from different directions — they can be complementary rather than competing:

	Solid's Text2SQL	Snowflake Cortex Analyst
Works with	Any warehouse (Snowflake, Redshift, BigQuery, Databricks, and more)	Snowflake only
Grounded in	Solid's certified semantic model (tested, governed definitions)	Snowflake Semantic Views
Accuracy mechanism	Iterative correction loop with DB Agent validation; benchmarked against certified ground-truth SQLs	Snowflake LLM inference on the Semantic View
Governance layer	Full lifecycle: auto-create, benchmark, certify, maintain, version	Semantic Views defined and maintained manually
Correction loop	Yes — generated SQL is validated and corrected before returning	No built-in correction loop
Output	SQL + explanation, returned to the calling agent or used by Solid Analyze	Natural-language answer via Cortex Analyst interface

If your primary warehouse is Snowflake and you also use Cortex Analyst, enabling this integration lets you manage your semantic layer in Solid (with its full governance and benchmarking workflow) and consume it in both Solid's MCP server and Cortex Analyst simultaneously.

How to Enable

Run this single SQL command as ACCOUNTADMIN:

USE ROLE ACCOUNTADMIN;

SET solid_role_name = 'SOLID_DATA_ROLE'; -- Your Solid role name

GRANT DATABASE ROLE SNOWFLAKE.CORTEX_USER 
TO ROLE identifier($solid_role_name);

This grants Solid's service role access to Snowflake's built-in Cortex AI capabilities.

Need Help? Contact your Solid support team.