What's New

October 2, 2026

What's new in Sigma

AI

Assistant in build mode: reuse data model metrics

In workbooks, Sigma Assistant in build mode now preserves data model sources you attach and reuses their metrics instead of recreating the underlying tables and calculations. When you ask Assistant to build a dashboard from a data model, KPIs and charts use the metrics defined in that data model, keeping values consistent with your governed definitions.

For more information about Assistant in build mode, see Use Sigma Assistant to build dashboards and apps.

Call Sigma agents from the Sigma MCP server

Interact with your workbook agents from any AI tool connected to the Sigma MCP server. The server includes two new tools:

  • List the agents you can access
  • Call an agent

These tools are only available in the Sigma MCP server, not in the plugin or connector. For more information, see Use the Sigma MCP server.

Chat history with customer-owned storage

Admins can now choose where chat history is stored:

  • If your organization grants storage rights to Sigma, you can configure chat history to be stored by Sigma in a Sigma-owned bucket.
  • If your organization requires more control over storage to meet data governance and compliance requirements, you can instead store chat history in a customer-owned bucket through an external storage integration.

For more information, see Configure chat history.

MCP connectors for Sigma agents (GA)

Admins can add MCP servers to Sigma as MCP connectors so that Sigma agents can retrieve context, fetch ad hoc data, and take actions in third-party tools. For details, see Configure MCP connectors.

New models used for AI providers

The following LLM models have been updated for the following AI providers:

  • If you use OpenAI, Sigma now uses GPT 5.6 as the LLM.
  • If you use Databricks, Sigma now uses Claude Sonnet 5 as the LLM.
  • If you use Anthropic, Sigma now uses Claude Sonnet 5 as the LLM.

For more details, see Supported AI models.

Sigma Assistant in the workbook (GA)

Sigma Assistant in the workbook is now generally available.

In workbooks, Assistant lets you use natural-language prompts to explore workbook data, analyze insights, and build or edit dashboards and apps.

This release also includes the following updates:

  • Chat history: View, resume, rename, and delete your own recent chats with Sigma Assistant. See Configure chat history.
  • Code-first architecture: Assistant builds by writing code, which improves accuracy, speeds up builds, and lowers the cost to run.
  • Broader element support: Assistant can build and customize more workbook element types, including maps, combo charts, navigation, drawers, and single row containers.
  • More in-chat chart capabilities: Assistant can include more chart types and formatting (like reference lines and trend lines) directly in the chat, and you can immediately add them to the workbook canvas.
  • Improved look and feel: Assistant builds with better design defaults and supports more theming options.

For more information, see Use Sigma Assistant to explore and analyze workbook data and Use Sigma Assistant to build dashboards and apps.

Sigma agents (GA)

Build agents in workbooks that use instructions, data sources, actions, warehouse agents, warehouse search services, and MCP connectors to help users with business tasks.

Chat with agents in a workbook, prompt them in an automated action run on a schedule, or call agents with the API or Sigma MCP server.

When you chat with agents, you can resume conversations stored in the chat history, approve tasks for the agent to perform, and start new chats. You can also embed a Sigma agent chat interface. For real-world use cases, see Example use cases for Sigma agents.

Admins can review agents across the organization in Administration, reviewing token usage, owners, data sources, access grants, user feedback, and an execution log for every conversation turn. See Manage Sigma agents for your organization.

For assistance evaluating agent quality, follow the guidelines in Build an evaluation suite for your agents.

Sigma agents are a premium feature. For billing details, see About billable usage events. For more details, see About Sigma agents.

API

New options for some workbook endpoints

To make it clear which tags on a workbook are inactive when using the API, the tags array of the response of the following endpoints now includes an isArchived option:

New options for the Create a deployment policy endpoint

The Create a deployment policy (POST /v2/deploymentPolicies) endpoint includes a new useDependenciesWorkspace option to specify a separate workspace when deploying dependent documents. For more details about dependency management in a deployment policy, see How dependencies are deployed.

New options for the List workbooks endpoint

The List workbooks (GET /v2/workbooks) endpoint includes a new query string parameter: includeTaggedSourceUrlId.

When that query string parameter is included, the taggedSourceUrlId is included in the response for version-tagged workbooks that were deployed from a parent or another tenant organization, making it possible to easily identify the source document for a deployed document.

Admin

Query variables support for Cortex Agents (Beta)

You can now support row-level security set in row access policies when using Snowflake Cortex Agents in Sigma. When you configure query variables on a Snowflake connection, Sigma sets immutable session attributes used for every Cortex Agent interaction, enforcing RLS. See Specify query variables for a Snowflake connection (Beta).

Data modeling

Revert migrated dataset references

When you migrate a dataset to a data model and update references to use the new data model as a source, you can now choose to revert the references so they use the old dataset as a source again. This is helpful when troubleshooting dataset migrations or testing different approaches to a migration.

For more information, see Migrate a dataset to a data model.

Workbook elements

Date support for slider and range slider controls

You can now select between Date and Number value types when configuring slider and range slider controls.

See Slider and Range slider in Intro to control elements for more information.

Bug fixes and improvements

  • If you used OpenAI GPT-5.4 or OpenAI GPT-5.1 as the model, calling an agent as part of an automated action failed to run.

  • Any user with Can edit access to a dataset can now migrate that dataset to a data model.

  • Previously, an error prevented selecting the Properties tab for an element after selecting another tab in the editor panel. Now, you can select the Properties tab as expected.

  • When trying to connect to OpenAI as an external AI provider, Sigma expected the GPT 4.0 model to be available instead of the GPT 5.4 model.

  • When using the Call agent action to automate a Sigma agent on a schedule, the action could return null values when the agent’s response was incomplete or had the wrong data type, and actual null values caused an “invalid action variable” error in later actions in the sequence. Missing or incorrect values now trigger a retry, and null values keep their type so later actions can use them successfully.

  • When using Azure OpenAI as the AI provider, chats with agents and Assistant ended partway through a response with the message “The assistant stopped responding.” Responses now finish normally.