Snowflake for Gen AI · Topic 5 of 13
Copilot Inline, Snowflake Intelligence and the Gen AI Interfaces
Copilot inline, Snowflake Intelligence and the interfaces to Cortex: AI Studio in Snowsight, SQL functions and the REST API, and when to use each.
- 4 min
- Copilot inline: Workspaces, diff, no table data, needs
COPILOT_USERandCORTEX_USER. - Snowflake Intelligence runs on Cortex Agents and respects row access and column security.
- Interfaces: Snowsight and AI Studio (no code), SQL (batch), REST (apps), Python.
What it is
Snowflake offers Gen AI to three kinds of people: developers who write SQL, business users who ask questions, and application builders who call services. Each has its own door. This page connects the products to the people and to the interface they use.
Snowflake Copilot inline helps a developer inside a SQL file. Snowflake Intelligence is a conversational application for business users, built on Cortex Agents. Interfaces are the ways to reach the Cortex features: Snowsight (including AI Studio), SQL and the REST API.
How it works
Snowflake Copilot inline
Copilot inline extends Snowflake Copilot so you can ask for help from within your SQL code. The syllabus lists it as Public Preview.
- It generates SQL from a natural language request, explains and optimizes queries, fixes syntax errors and helps you explore data.
- It works in Workspaces. You open it with the CMD+I shortcut and review the proposed change as a diff, then accept or reject.
- It has no access to your table data, so you supply filter values yourself.
- Users need the
COPILOT_USERandCORTEX_USERdatabase roles in the SNOWFLAKE database. Both are granted to PUBLIC by default; revoking one removes access. - It also needs an allowed model (the docs name claude-3.5-sonnet or openai-gpt-4.1) that is on the model allowlist and not blocked by RBAC.
- In regions without native support it needs the
CORTEX_ENABLED_CROSS_REGIONparameter.
Copilot inline helps write SQL. It is not Cortex Analyst, which answers business questions for end users from a semantic view.
Snowflake Intelligence
Snowflake Intelligence is a ready to use agentic application where people ask questions about structured and unstructured enterprise data in natural language. Its agents interpret the question, pick tools, analyze data and can take action. Newer Snowflake documentation presents this application under the name Snowflake CoWork.
It runs on the Cortex Agent API. Cortex Analyst produces SQL over semantic views, Cortex Search retrieves from documents, and custom tools run UDFs or stored procedures. It inherits existing governance, so row access policies and column level security still apply, and uploaded files are treated as customer data under the same controls. A user can upload files (the documented limit is 50 MB per file and 5 files at a time), and they are kept in the user's stage for that thread.
Different interfaces
| Interface | Best for |
|---|---|
| Snowsight, including AI Studio | Trying features without code, configuring and testing |
| SQL | Batch work over table rows with AI functions |
| REST API | Applications, low latency interactive use |
Python (snowflake.core, Snowpark) | Notebooks and pipelines |
AI Studio
AI Studio is the Snowsight, no code entry point to Cortex features. Use it when the scenario describes someone who wants to experiment or configure without writing code, and move to SQL or REST for production.
SQL
The AI functions are called as ordinary SQL, so they work inside SELECT, views, tasks and dynamic tables. Snowflake advises batch processing in SQL for large tables.
REST API
The REST endpoints suit latency sensitive applications. For example POST /api/v2/cortex/inference/complete runs completions, POST /api/v2/cortex/analyst/message runs Cortex Analyst, and an agent is run with agent:run. They are covered in detail on a later page.
SQL is best for large table batches, and REST for interactive latency. A scenario that says "thousands of rows overnight" points to SQL.
Common mistakes
Mixing Copilot with Analyst. Copilot helps developers write SQL; Analyst serves business questions.
Thinking Copilot can see table rows. It cannot, so filter values must be supplied.
Treating Snowflake Intelligence as separate from Cortex. It is built on Cortex Agents, Analyst and Search.
Choosing REST for a nightly batch. Use SQL functions over the table.
Assuming file upload has no limits. The documented limits are 50 MB per file and 5 files at a time.
What to remember for the exam
- Copilot inline: Workspaces, diff, no table data, needs
COPILOT_USERandCORTEX_USER. - Snowflake Intelligence runs on Cortex Agents and respects row access and column security.
- Interfaces: Snowsight and AI Studio (no code), SQL (batch), REST (apps), Python.
Check yourself
Who is Snowflake Intelligence for?
Business users who ask natural language questions over structured and unstructured data.
Why must a Copilot inline user supply filter values?
Copilot has no access to the table data.
Which interface suits a nightly job over millions of rows?
SQL AI functions, because batch processing is recommended for large tables.
Topic 5 of 13 in Snowflake for Gen AI
Test yourself on Snowflake for Gen AI
The full Snowflake GES-C02 question bank has 585 questions, with timed practice exams and an explanation for every answer.
This study guide is independent and is not affiliated with, endorsed by, or sponsored by Snowflake. Vendor names, certification names and exam codes are trademarks of their respective owners and are used for identification only.