Configure warehouse AI model integration (Beta)

🚩

This documentation describes a public beta feature and is under construction. This page should not be considered part of our published documentation until this notice, and the corresponding Beta flag on the feature in the Sigma service, are removed. As with any beta feature, the feature discussed below is subject to quick, iterative changes. The latest experience in the Sigma service might differ from the contents of this document.

Beta features are subject to the Beta features disclaimer.

You can elect to use AI models in your cloud data warehouse to power Sigma AI features.

Limitations

• Only Snowflake and Databricks connections can be used for warehouse AI models at this time.
• Configuring a warehouse model as your AI provider is not currently supported for the Explain charts with AI feature. Both Ask Sigma and formula assistant are supported with warehouse models.

System and user requirements

• You must be assigned the Admin account type.

• If you are using a Snowflake warehouse as your AI provider:

  • You must have a Snowflake connection authenticated with either Key Pair auth or OAuth. Basic Auth connections are not supported.

  • Verify the user defined on the connection has the snowflake.cortex_user database role in their default role. By default, this role is inherited from PUBLIC. In case it has been revoked from PUBLIC, follow Setting up authorization in the Snowflake documentation to grant the role.

    • If you are using a connection with Key Pair authentication, the Key Pair service account user on the connection must have snowflake.cortex_user in their default role.
    • If you are using a connection with OAuth authentication, all users with access to the connection must have snowflake.cortex_user in their default role. If you are using a service account, the service account user must have snowflake.cortex_user in their default role.
    • The default role must have REST API access.

  • Your Snowflake warehouse must be in the US West 2 region, or, if your warehouse is hosted in another region, you must enable cross-region inference. See Cross-region inference in the Snowflake documentation.

• If you are using a Databricks warehouse as your AI provider:

  • You must have an active Databricks workspace within a supported model serving region with AI Model Serving enabled (currently in public preview).
  • You must have a running Databricks Serverless or Pro SQL warehouse (not available with Classic SQL warehouses) with access to the ai_query function. For detailed requirements related to the ai_query function for foundational models, refer to Example: Query a foundation model in the Databricks documentation.
  • The Databricks connection must have read access to the data sources you intend to query. This includes: USE, BROWSE, EXECUTE, and SELECT privileges on the catalog or schema containing the data sources.

Configure data warehouse hosted model as your AI provider

1. Go to Administration > AI Settings:
   1. In the Sigma header, click your user avatar to open the user menu.
   2. Select Administration to open the Administration portal.
   3. In the side panel, select AI Settings.
2. If this is the first time you are configuring an AI provider integration, skip this step.
   
   If External models is selected in the AI provider integration section and the available buttons read Remove and Update, you have an OpenAI or Azure OpenAI key already provisioned. If you want to use a warehouse model instead, click Remove to clear your credentials.

   📘

   Removing your OpenAI or Azure OpenAI credentials clears out all previously indexed data sources. You will need to re-index all data sources after configuring a new AI provider integration.

3. Select Data warehouse hosted model (recommended).
4. In the Connection dropdown menu, select the connection to use to access models in your data warehouse.
5. Click Save. If the connection is successful at reaching the AI models in your data warehouse, the Ask Sigma data sources section appears.
6. Next, Select the data sources that should be available to Ask Sigma.