Snowflake Cortex

Snowflake

LLMs as SQL functions inside Snowflake: complete, embed, search and extract without moving the table — AI for teams whose governance story is already the warehouse.

Product preview

Overview

Cortex is the bet that the warehouse should not send rows to a chatbot: you `SELECT SNOWFLAKE.CORTEX.COMPLETE(...)` over the table you already govern, embed in place, search in place. Databricks has a parallel story; Hex is the notebook on top. Cortex is the function.

It will not beat a purpose-built agent loop. It will beat a security review that starts with “we copied PII to OpenAI.” Model choice is Snowflake's catalog, not yours. If the data already cannot leave Snowflake, this is the least-friction LLM.

Pros / Cons

Pros

LLM where the governed table already sits
SQL-shaped, so analysts can call it
Search and embed without a second database
Shortest path through a warehouse security review

Cons

Model catalog is Snowflake's, not yours
Not an agent platform
Useless if the data is not already in Snowflake

Who it's for

Summarizing warehouse rows in SQL

RAG over Snowflake tables without export

Extracting fields from unstructured columns

Analyst-accessible LLM inside existing roles

Bottom line

LLMs as warehouse functions — the path of least governance friction when the table cannot leave Snowflake.

Key Features

LLM complete / extract as SQL

Cortex Search for RAG over tables

Embeddings without an extra store

Governed access via Snowflake roles

No data copy out of the warehouse

Compare Snowflake Cortex with alternatives

Snowflake Cortex vs Hugging Face Snowflake Cortex vs Replicate Snowflake Cortex vs Ollama 4 products · AI Platforms

Frequently asked questions

What is Snowflake Cortex?

Snowflake Cortex is a AI Platforms tool developed by Snowflake. The company was founded in 2023 and is headquartered in Bozeman, USA. LLMs as SQL functions inside Snowflake: complete, embed, search and extract without moving the table — AI for teams whose governance story is already the warehouse.