Snowflake Cortex

Overview

Snowflake Cortex provides a set of managed AI functions and services designed to integrate large language models (LLMs) and vector search capabilities directly within the Snowflake Data Cloud. It aims to enable developers and data practitioners to build AI-powered applications, extract insights, and perform advanced analytics on their enterprise data using familiar SQL constructs (Snowflake Cortex documentation). This approach allows users to leverage AI without needing extensive machine learning operations (MLOps) expertise or managing separate infrastructure for model deployment and inference.

The platform is engineered for scenarios where data resides within Snowflake, minimizing data movement and maintaining data governance and security (Snowflake Cortex homepage). Key use cases include augmenting business intelligence (BI) dashboards with natural language explanations, enabling semantic search over internal documentation, summarizing large datasets, and generating content based on proprietary information. By exposing LLM functionalities as SQL functions, Snowflake Cortex simplifies the development process for those already proficient in SQL, allowing them to embed AI directly into their data pipelines and applications.

Snowflake Cortex is built atop a unified data platform, which includes capabilities for data warehousing, data lakes, and data engineering. This architecture supports a range of workloads from data ingestion and transformation to AI/ML model inference. The integration of LLMs within this environment allows for the processing of both structured and unstructured data, facilitating tasks like text summarization, entity extraction, and question answering directly on data stored within Snowflake tables. This can reduce the operational overhead associated with integrating external AI services and help maintain data locality for compliance and performance considerations.

Key features

• LLM Functions: Provides SQL functions to interact with various LLMs, enabling tasks like text summarization, sentiment analysis, entity extraction, and question answering directly within SQL queries (Snowflake Cortex LLM Functions).
• Snowflake Cortex Search: Offers capabilities for semantic search and retrieval-augmented generation (RAG) by converting text into vector embeddings and performing similarity searches over enterprise knowledge bases stored in Snowflake (Snowflake Cortex Search overview).
• Snowflake Cortex Analyst: Designed to automate and assist with data analysis tasks, potentially through natural language querying and automated insight generation, although specific details are evolving (Snowflake Cortex Analyst description).
• Vector Embeddings: Supports the generation and storage of vector embeddings for text, images, and other data types, facilitating similarity searches and other AI applications.
• Managed Infrastructure: All AI services are fully managed by Snowflake, abstracting away the underlying infrastructure management, scaling, and model deployment complexities.
• Data Governance and Security: Integrates with Snowflake's existing security and governance features, ensuring data privacy and compliance when working with sensitive enterprise data.

Pricing

Snowflake Cortex operates on a usage-based pricing model, consistent with the broader Snowflake Data Cloud. Costs are incurred based on the compute resources consumed for running queries, executing LLM functions, and performing vector search operations. Specific pricing details are available on the official pricing page.

For detailed and up-to-date pricing information, refer to the Snowflake Cortex pricing page.

Common integrations

• Snowflake Data Cloud: Native integration with all aspects of the Snowflake platform, including data warehousing, data lakes, and data engineering pipelines.
• BI Tools: Can be integrated with business intelligence tools like Tableau, Power BI, or Looker by leveraging SQL queries that incorporate Cortex functions to enrich dashboards with AI insights.
• Custom Applications: Developers can integrate Cortex capabilities into custom applications using Snowflake's various client connectors (JDBC, ODBC, Python, Node.js, etc.) to execute SQL queries with LLM functions.
• ETL/ELT Tools: Works with data integration platforms that connect to Snowflake, enabling AI-powered transformations during ETL/ELT processes.

Alternatives

• Databricks Mosaic AI: Offers a data intelligence platform with integrated AI capabilities, including LLM fine-tuning and deployment, on a Lakehouse architecture.
• Google Cloud Vertex AI: Provides a managed machine learning platform for building, deploying, and scaling ML models, including access to Google's foundational models like Gemini.
• Amazon Bedrock: A fully managed service that offers access to foundation models (FMs) from Amazon and leading AI startups via a single API, along with tools for building and scaling generative AI applications.
• Cohere Platform: Focuses on enterprise-grade LLMs for various NLP tasks, providing APIs for generation, embeddings, and RAG.
• OpenAI Platform: Offers access to powerful LLMs like GPT-4 and GPT-3.5 via API, commonly used for a wide range of generative AI applications.

Getting started

To get started with Snowflake Cortex, you can use SQL to call the provided LLM functions. The following example demonstrates how to use an LLM function to summarize a piece of text directly within a Snowflake worksheet or application.

-- Example: Summarizing text using an LLM function in Snowflake Cortex

-- Ensure you have an active warehouse and database selected
USE DATABASE your_database;
USE SCHEMA your_schema;

-- Create a sample table with text data
CREATE OR REPLACE TABLE articles (
    article_id INT,
    article_text VARCHAR
);

INSERT INTO articles (article_id, article_text) VALUES
(1, 'The quick brown fox jumps over the lazy dog. This sentence is often used to test typefaces and keyboard layouts because it contains all letters of the English alphabet. It is a pangram.'),
(2, 'Artificial intelligence (AI) is intelligence—perceiving, synthesizing, and inferring information—demonstrated by machines, as opposed to intelligence displayed by animals or humans. AI research has been defined as the field of study that examines how to create computer programs that are capable of intelligent behavior.');

-- Use the COMPLETE_FLAT_TEXT LLM function to summarize the article text
SELECT
    article_id,
    article_text,
    SNOWFLAKE.CORTEX.COMPLETE_FLAT_TEXT(
        'llama3-8b',
        'Summarize the following text in one sentence: ' || article_text
    ) AS summary
FROM
    articles;

This SQL query first creates a table and inserts sample text. It then uses the SNOWFLAKE.CORTEX.COMPLETE_FLAT_TEXT function, specifying a model (e.g., llama3-8b) and a prompt, to generate a one-sentence summary for each article. This demonstrates how LLM capabilities can be integrated directly into standard SQL queries for data processing and analysis (Snowflake LLM Functions documentation).