Getting Started with GPT‑5.6 on Snowflake Cortex AI

Imagine you’ve just saved a spreadsheet of your grocery spendings, or you need a quick draft for a client email. What if a single AI could read that data, spot patterns, and give you a clear summary in seconds? That’s exactly what the combination of OpenAI’s newest large language model (LLM — think of it as the brain behind ChatGPT) and Snowflake’s Cortex AI (the AI‑powered layer on Snowflake’s cloud data platform) lets you do. The good news is you don’t need a PhD in data science to make it work.

1. Set up a Snowflake account (free trial works)

1. Sign up on Snowflake’s website and choose the “Free trial” option.
2. Create a new database – give it a friendly name like my_ai_demo.
3. Inside that database, create a schema (a logical folder) called cortex_demo.

Snowflake (a cloud data‑warehouse service) stores data in tables, similar to Excel sheets, but it can handle far larger volumes and does so securely.

2. Enable the Cortex AI extension

1. In the Snowflake UI, go to Marketplace and find the “Cortex AI” add‑on.
2. Click Add to account – Snowflake will automatically provision the necessary compute resources.

Cortex AI is Snowflake’s built‑in tool that lets you call external AI models directly from SQL (the language you use to ask questions of a database).

3. Connect OpenAI’s GPT‑5.6 model

1. Within the cortex_demo schema, run this simple command (replace YOUR_OPENAI_KEY with the API key you obtain from your OpenAI account):

CREATE OR REPLACE MODEL gpt56
    TYPE = 'OPENAI'
    VERSION = '5.6'
    API_KEY = 'YOUR_OPENAI_KEY';

API (application programming interface) is just a way for two programmes to talk to each other – here Snowflake talks to OpenAI.

2. After the model is registered, you can start sending prompts (the instructions you give to the AI) straight from SQL.

4. Two quick examples you can try today

a) Summarise a personal expense sheet

Assume you have a table expenses with columns date, category, amount. To get a short paragraph that tells you where you spend the most, run:

SELECT CORTEX_COMPLETION(
    model => 'gpt56',
    prompt => 'Summarise the spending trends in this table and suggest one cost‑saving tip.',
    input => OBJECT_CONSTRUCT(*)
) AS summary
FROM expenses;

The result will be a natural‑language summary, e.g., “You spent the most on groceries… Consider buying in bulk…”

b) Draft a friendly reply to a client email

Create a table incoming_emails that holds the raw text of an email. Then ask GPT‑5.6 to draft a reply:

SELECT CORTEX_COMPLETION(
    model => 'gpt56',
    prompt => 'Write a polite, concise response confirming the meeting on Thursday at 10 am.',
    input => OBJECT_CONSTRUCT(*)
) AS draft_reply
FROM incoming_emails
WHERE email_id = 42;

You get a ready‑to‑copy reply that you can paste into Outlook or Gmail.

Wrap‑up

Snowflake’s Cortex AI gives you a secure, governed environment to harness OpenAI’s GPT‑5.6 without needing any specialised coding skills. By linking your own data tables to the model, you can turn raw numbers and text into useful, human‑readable insights in minutes. Today, sign up for a Snowflake trial, load a small CSV (your grocery list or a few email snippets), and run one of the sample queries. In just a handful of clicks you’ll see how a powerful LLM can become a practical assistant for both personal and work‑related tasks. Happy prompting!