Snowflake Data Clean Rooms Agent

Overview

The Snowflake Data Clean Room (DCR) Agent offers a conversational interface in Snowflake Intelligence, making data clean rooms intuitive and accessible for business and non-technical users. It facilitates key workflows, including the effortless execution of approved analyses, with results presented in an easy-to-understand format.

The agent currently supports the following Snowflake Data Clean Rooms templates:

  • Audience Overlap: A template that measures how many records overlap between two datasets.
  • Audience Overlap Activation: A template that generates a list of overlapping records with respective attributes, and exports it to an approved account.

See Overlap and activation for collaboration roles, sample data, and setup notebooks for this scenario. The notebooks generate sample datasets based on data from the public DCR sample dataset.

The agent uses Claude Sonnet 4.6 as the LLM.

Supported flows

The DCR Agent can currently perform the following actions in a collaboration:

  • Join a collaboration.
  • View information about a collaboration, including the status, templates, and data offerings in that collaboration.
  • Run an audience overlap analysis in a collaboration.
  • Activate data from an overlap analysis.
  • View the analysis history for any analyses run using the agent in this account.

Requirements

Agent setup

Run the following SQL in an account to enable agent functionality:

-- Enable cross-region inference.
USE ROLE ACCOUNTADMIN;
ALTER ACCOUNT SET CORTEX_ENABLED_CROSS_REGION = 'AWS_US';

-- Ensure DCR application has the appropriate grants via caller.
GRANT CALLER FULL MANAGEMENT ON ACCOUNT TO APPLICATION SAMOOHA_BY_SNOWFLAKE;

Register the DCR agent with Snowflake Intelligence so users can access it in the Snowflake Intelligence interface:

USE ROLE ACCOUNTADMIN;
CREATE SNOWFLAKE INTELLIGENCE IF NOT EXISTS SNOWFLAKE_INTELLIGENCE_OBJECT_DEFAULT;
ALTER SNOWFLAKE INTELLIGENCE SNOWFLAKE_INTELLIGENCE_OBJECT_DEFAULT
  ADD AGENT SAMOOHA_BY_SNOWFLAKE.AGENTS.SNOWFLAKE_DCR_AGENT;

Access requirements

DCR Agent and Data Clean Rooms permissions are managed separately. To use the agent, you need permissions on both the agent and the collaboration. Grant the proper permissions as follows:

  1. Grant the role permission to use the DCR Agent and access it in Snowflake Intelligence:

    USE ROLE ACCOUNTADMIN;
GRANT USAGE ON SNOWFLAKE INTELLIGENCE SNOWFLAKE_INTELLIGENCE_OBJECT_DEFAULT
  TO ROLE <your role name>;
CALL SAMOOHA_BY_SNOWFLAKE_LOCAL_DB.ADMIN.GRANT_PRIVILEGE_ON_ACCOUNT_TO_ROLE(
  'USE DCR AGENT',
  '<your role name>'
);

  2. Grant the role permission to run analyses in a collaboration by calling GRANT_PRIVILEGE_ON_OBJECT_TO_ROLE:

    USE ROLE ACCOUNTADMIN;
CALL SAMOOHA_BY_SNOWFLAKE_LOCAL_DB.ADMIN.GRANT_PRIVILEGE_ON_OBJECT_TO_ROLE(
  'RUN',
  'COLLABORATION',
  '<collaboration name>',
  '<your role name>'
);

Note

SAMOOHA_APP_ROLE must still have USE DCR AGENT granted separately to use the Snowflake Data Clean Room Agent.

Usage

After setting up your collaboration, run an analysis and activate results using the agent. There are two ways to access the agent:

Note

If a role has all required permissions to join a collaboration, you can also request the agent to join a current collaboration your account is invited to. If any data offerings need to be added, this will need to be done via the Collaboration API or Snowsight UI.

Run an analysis and activate results

From Snowflake Intelligence, ask the agent to run an analysis in the collaboration that you joined. You can also ask the agent what actions you can perform, what data is available, what templates are available, and other contextual questions.

If activation is enabled in your collaboration, you can ask the agent to activate analysis results to available destinations in the collaboration.

Save analysis results

Ask the agent to save an overlap analysis result by providing a name for the saved analysis. You can later list or fetch your saved analyses by name.

You can view only analyses that you have run yourself. Administrators using the SAMOOHA_APP_ROLE role can see analysis results for all analyses.