Cortex Code is Snowflake's AI coding agent for data work. It reads your schema, writes and explains SQL, builds pipelines, and runs multi-step tasks for you. It works with Snowflake Postgres and with Postgres databases running somewhere else.

There are three ways to run it:

Cortex Code Desktop is a standalone IDE built on VS Code. It connects to Snowflake and to Postgres, and the agent can read your schema, write and run queries, edit files, and work through multi-step tasks.

Cortex Code Desktop manages Snowflake Postgres instances through your Snowflake connection. For Postgres outside Snowflake (RDS, Aurora, Cloud SQL, self-hosted), you connect with the usual Postgres connection details, and the agent works with either.

Download Cortex Code Desktop

See the docs for supported operating systems.

Cortex Code is also built into Snowsight, so you can use it from the browser without installing anything.

In Snowsight, Cortex Code works through your Snowflake session and role, so there's no separate client to configure. What it can see and do with your Snowflake Postgres instances depends on the privileges your role has.

Open Snowsight

Log in to your Snowflake account and open Cortex Code from Snowsight.

The Cortex Code CLI puts the agent in your terminal. It's handy for quick queries, scripts, automation, and fitting AI-assisted data work into the command-line habits you already have.

The CLI manages Snowflake Postgres through your configured Snowflake connection. For other Postgres instances, use the same connection parameters (host, port, database, user) you'd give any Postgres client.

Cortex Code CLI Documentation

The docs cover installation options for your platform.

Cortex Code Desktop and the CLI include a snowflake-postgres skill that kicks in when you're working with Postgres. You don't have to set anything up to use it.

It handles instance management, so you can create, suspend, resume, and resize Snowflake Postgres instances by asking in plain English.

It answers questions about your data. Ask in English, and Cortex Code writes and runs the SQL against your Postgres instance.

It runs health checks with pg_doctor. These are read-only diagnostics covering cache hit rates, table bloat, vacuum status, connections, locks, blocking queries, long-running queries, unused indexes, and table sizes. It's the same list of things we'd check by hand on a struggling database.

It walks you through data movement. Connecting Postgres to Snowflake with pg_lake touches Snowflake SQL, cloud storage permissions (for customer-managed S3 or Azure), and Postgres SQL, and the skill takes you through each piece.

It manages connections in the standard ~/.pg_service.conf and ~/.pgpass files, so psql and every other Postgres client can use them too.

Cortex Code asks for confirmation before write operations like INSERT, UPDATE, DELETE, and DROP. Read-only work such as SELECT, DESCRIBE, and health checks can run without stopping to ask.

Billable or security-sensitive operations also need confirmation. That includes creating instances, changing network policies, and changing storage integrations. Your approval settings in Cortex Code control how much it asks.

Cortex Code Desktop and the CLI can connect to Postgres outside Snowflake: RDS, Aurora, Cloud SQL, Azure Database for PostgreSQL, or your own servers.

When you're connected to a Snowflake Postgres instance, Cortex Code can also help with the Snowflake side of things.

Cortex Code knows how pg_lake connects your Postgres tables to Snowflake, so it can help you build pipelines and queries that span both.

It can write and test Cortex AI functions like AI_EMBED and AI_COMPLETE against your Postgres data. See Snowflake Postgres + Cortex AI for the patterns.

Instances are Snowflake objects, so Snowflake roles and privileges govern who can manage them. Cortex Code can create, resize, suspend, and monitor instances for you.

1. Connect Cortex Code to your Snowflake account.
2. Pick the Postgres instance you want to work on.
3. Ask Cortex Code to explore your schema, write queries, or build data pipelines.
4. Use pg_lake to get Postgres data into Snowflake for analytics at warehouse scale.
5. Use Cortex AI functions to enrich or summarize that data.

Here's how the three compare at a glance. Features change quickly, so check the docs for the current details.