dbt integration in Cortex Code Desktop¶

Cortex Code Desktop includes a built-in dbt panel that provides a full development environment for dbt projects. When you open a workspace containing a dbt_project.yml, the dbt panel activates automatically and gives you interactive lineage visualization, compiled SQL preview, data preview, documentation, and a command bar — all without leaving the editor.

The dbt panel with the Lineage tab active, showing upstream/downstream model relationships and the command bar.

Getting started¶

The dbt panel appears in the bottom panel area (next to Problems, Output, Terminal, etc.) when Cortex Code detects a dbt_project.yml in your workspace. Click the DBT tab to open it.

The panel has five tabs and a command bar:

Execution mode¶

Cortex Code Desktop can run dbt commands two ways:

In Snowflake-managed mode, your dbt project files are uploaded to a Snowflake workspace under your account, dbt executes server-side, and the generated artifacts (target/, dbt_packages/, package-lock.yml, logs) are downloaded back to your local project directory after each run. The Lineage, Output, Docs, and Compiled tabs work the same way in both modes — the difference is where the work happens.

The dbt panel in Snowflake-managed mode. A Snowflake Managed badge in the panel header confirms which mode is active. The project picker selects which dbt project to run; the profile picker selects which target (within the profile defined in your project’s dbt_project.yml) the next run will use.

For the underlying Snowflake feature, see Workspaces for dbt projects on Snowflake.

Switching between modes¶

Change the active mode in any of three ways:

• Settings UI — Search for dbt: Execution Mode and pick local or snowflake-managed.
• settings.json — Set "dbt.executionMode": "snowflake-managed".
• Command Palette — Run Toggle dbt Execution Mode (Snowflake-Managed/Local). A notification confirms the new mode.

The Command Palette entry for switching dbt execution mode.

When Snowflake-managed mode is active, the panel header shows a Snowflake Managed badge on every tab.

What happens when you run a command¶

When you run a dbt command in Snowflake-managed mode, Cortex Code Desktop:

1. Ensures a workspace exists in your account at USER$.PUBLIC named coco_<project>_<hash>. The workspace is created on first use and reused on subsequent runs.
2. Syncs your project files to the workspace. Only files that changed since the last sync are uploaded, so subsequent runs are fast.
3. Executes the dbt command on Snowflake via EXECUTE DBT PROJECT FROM WORKSPACE. The warehouse, database, schema, and role specified in the active target (selected in the profile picker, defined in profiles.yml under the profile: named in dbt_project.yml) are used for the run.
4. Downloads the generated artifacts back to your local project directory so the Compiled and Lineage tabs reflect the server-side run.

The Output tab shows the SQL that was executed and streams the dbt output. Click Stop to cancel a running command.

In Snowflake-managed mode, the Command bar and the Build / Compile / Show quick-action buttons (described later in this page) are hidden. Run dbt commands through the Run button and the execution popover instead.

Customizing a run¶

Click the chevron next to Run to open the execution popover and customize the next run:

The execution popover with per-run controls.

Click Execute to run with the selected options.

Prerequisites for Snowflake-managed mode¶

• An active Snowflake connection signed in via the connection picker.

• A profiles.yml in each dbt project folder. The type must be snowflake and the warehouse, database, schema, and role fields must be set. Unlike dbt Core, the account and user fields can be left empty or with arbitrary values — the workspace runs under your current Snowflake user and account context. See Workspaces for dbt projects on Snowflake for the full requirements.

• Personal databases enabled on your account. Snowflake workspaces are created in a personal database, which an account administrator must enable. See the personal database requirement.

• An external access integration, only if your packages.yml references remote packages (for example, packages from the dbt Packages hub or a Git repository). The EAI is selected per-run in the execution popover; local-only dependencies don’t require one. See Workspaces for dbt projects on Snowflake — dependencies for setup details.

Project size limit¶

A dbt project synced to a Snowflake workspace can contain at most 20,000 files across all folders, including generated target/ and dbt_packages/ artifacts. See Limitations, requirements, and considerations.

Inspecting your workspaces¶

Each project gets a dedicated Snowflake workspace. The name has the shape coco_<project>_<hash>, where <project> is the lowercased project name (non-alphanumeric characters replaced with _) and <hash> is a 6-character hex tag. To list the workspaces Cortex Code Desktop has created in your account:

SHOW WORKSPACES LIKE 'coco_%' IN SCHEMA USER$.PUBLIC;

Workspaces persist between sessions, so the next sync only uploads what’s changed. If a workspace is dropped externally — for example by an administrator — Cortex Code Desktop re-creates it and performs a full sync on the next run.

Lineage¶

The Lineage tab renders an interactive directed acyclic graph (DAG) of your dbt project. It shows how models, sources, and other resources relate to each other.

• Upstream / Downstream — Toggle between viewing ancestors (upstream) or dependents (downstream) of the selected node.
• Depth — Control how many levels of dependencies to display (default: 5).
• Navigation — Use the back/forward arrows to navigate through previously viewed nodes.
• Refresh — Re-parse the project and update the graph.
• Selection — Click nodes to select them. Use ⌘+Click to add/remove from selection, or Shift+Drag to select multiple.

Output¶

The Output tab displays real-time results from dbt commands. Each command execution appears as a collapsible entry showing:

• The command that was run
• Timestamp and duration
• Exit code and status (running, success, or error)
• Per-model progress during dbt run or dbt build

Click Clear All to remove the output history.

Docs¶

The Docs tab shows the documentation for the currently selected dbt resource. It displays model descriptions, column metadata, and relationships. From the docs view you can:

• Navigate to related nodes by clicking their names.
• Open the source file for any referenced model.
• Trigger a compile to update the documentation.

Compiled¶

The Compiled tab shows the fully resolved SQL for the current model — all Jinja macros, refs, and sources are rendered into their final SQL form. This is the exact query that dbt would execute against your warehouse.

• Source / Compiled toggle — Switch between viewing the raw source and the compiled output.
• Syntax highlighting — Full SQL highlighting with line numbers and minimap.
• Search — Use Ctrl+F (or ⌘+F) to search within the compiled SQL.
• Copy — Copy the compiled SQL to your clipboard.

The status bar shows the model name and compilation duration.

Preview¶

The Preview tab executes dbt show and displays the results in an interactive data grid. This lets you quickly inspect the output of a model without running a full build.

• Click the Preview button (or use the Show action in the control bar) to fetch data.
• The default row limit is 10; adjust it using the limit control in the grid header.
• Results are cached per model until you re-run the preview or switch models.

Command bar¶

At the bottom of the dbt panel is a command bar where you can type and execute any dbt command directly. Type a command (for example, dbt run, dbt test, dbt compile) and press Run or hit Enter.

• Use ↑ and ↓ arrow keys to navigate through command history.
• Output from commands appears in the Output tab.

Quick actions¶

The control bar also provides one-click buttons for common operations:

Project detection¶

Cortex Code automatically detects dbt projects by scanning your workspace for dbt_project.yml files. When a dbt file is open in the editor, the panel automatically focuses on the corresponding model in the lineage graph and updates the Docs, Compiled, and Preview tabs.