Duration: 3

Dataiku Cobuild on Snowflake is an AI agent that turns natural-language requests into production-ready AI workflows, agents, and applications running on Snowflake. Powered by Snowflake Cortex, Cobuild lets teams describe a goal in plain language and get back an inspectable visual workflow they can review and govern before it reaches production, putting AI building within reach of analysts, operations teams, and business users.

This guide walks through how to set up Dataiku Cobuild with Snowflake Cortex to give your team model flexibility and AI governance while building production-grade data and ML pipelines. Cobuild is particularly useful for:

This guide is designed for users who already have a Dataiku instance connected to Snowflake. If you are new to Dataiku and Snowflake, visit the Snowflake Developers Guides and search for Dataiku to find guides that cover the initial setup and integration.

Duration: 2

Dataiku calls Snowflake Cortex via the REST API, which requires the user's default role to have either SNOWFLAKE.CORTEX_USER (all Cortex features, granted to PUBLIC by default) or SNOWFLAKE.CORTEX_REST_API_USER (REST API only). In most accounts no action is needed. If access has been revoked, run as ACCOUNTADMIN:

Also confirm the Dataiku service user's default role is set correctly:

See the Snowflake Cortex REST API documentation for more detail.

Enabling cross-region inference gives Dataiku access to a broader set of foundation models and increases rate limits by routing requests to available inference capacity across regions.

Run the following as ACCOUNTADMIN:

Duration: 5

In Dataiku, navigate to Administration > Connections, click + New Connection, and select Snowflake Cortex from the connection type list.

Select your existing Snowflake connection and configure your model options. From this screen you can control which models are available to your team. Under Usage Control, you can enable Prompt Caching to reduce redundant token usage and lower inference costs.

Once saved, this connection is available across LLM Mesh, Agents, LLM recipes, and Cobuild.

With the Cortex connection in place, you need to point Dataiku's AI Services at it so Cobuild can use it. From the grid menu in the top navigation, go to Administration.

Click the AI Services tab, set the Operation mode to Bring your own LLM, and click Enable. You will be prompted to accept Dataiku's AI Services Terms of Use before proceeding.

Under LLMs to use, enter your Reference project key (this is used for LLM Mesh requests that run outside a DSS project, such as AI Search). Then select a Main model from your Snowflake Cortex connection. You can also optionally assign a Fast / light model and a Response-format-aware model for different task types.

Fill in any additional usage permissions or global custom instructions, then click Save. Your Cortex configuration is now active and ready for Cobuild.

Duration: 5

Open any existing Dataiku project and click Cobuild from the project view.

Cobuild will inspect your flow automatically. You can also select specific items directly in the flow to give Cobuild more targeted context, or type @ in the prompt to reference a specific dataset, recipe, or model by name.

Before running any actions, you can choose how Cobuild executes: either prompt you to approve each action before it runs, or set it to automatic execution so Cobuild applies changes without waiting for confirmation.

To get started with Cobuild, type:

Cobuild will analyze your project and return a plain-language summary of what it contains, how it's structured, and what it does. Here's an example of what that looks like:

This is a good starting point to verify your Cortex connection is working and to get oriented before building or refactoring.

Once you're comfortable with how Cobuild reads your project, try starting fresh with a new dataset connected to Snowflake. Connect your data and then ask Cobuild to do the heavy lifting:

Cobuild will generate the complete flow as a visual, inspectable pipeline that your team can review and refine before anything runs in production.

Duration: 2

You now have Dataiku Cobuild connected to Snowflake Cortex and a production AI pipeline built entirely through natural language. From here you can apply the same workflow to new projects, refactor existing Dataiku flows, or use Cobuild to generate documentation for pipelines your team already owns.