What's New

September 25, 2026

What's new in Sigma

AI

Assistant in build mode: pivot table styling

In workbooks, Sigma Assistant in build mode now creates pivot tables with automatic heatmap and matrix styling based on the type of measures.

Measures that read against a threshold (e.g., profit, variance, margin, and growth) are styled as heatmaps with diverging color scales that help you spot patterns and outliers without reading every value. Measures that are straightforward totals (e.g., revenue and headcount) receive plain matrix styling.

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

Assistant in build mode: cohort retention tables

In workbooks, Sigma Assistant in build mode now generates styled cohort retention pivot tables. Assistant recognizes requests for cohort retention analysis and automatically builds the pivot table accordingly, including heatmap styling that makes retention patterns easy to identify.

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

Chat history for Assistant and agents (GA)

Chat history is now available for conversations with Sigma Assistant, warehouse agents, and Sigma agents. With chat history, users can now revisit and continue previous conversations instead of starting from scratch each time. Chat history is not yet available for conversations with Assistant in the workbook, or agents on the home page.

For organizations that meet certain conditions, an admin can configure chat history storage and retention. Considerations and limitations apply. See Configure chat history.

Improved Sigma Assistant

Sigma Assistant now includes improved reasoning, faster performance, and higher quality responses. Updates include:

  • Clarifying questions: Ask a broad business question and Assistant prompts you to narrow it down, sharing its assumptions and flagging caveats
  • All sources, or a curated shortlist: Admins can open Assistant up to every accessible data source. Role-based access controls still apply.
  • Visible reasoning and side by side detail inspection: Assistant describes its reasoning as it works, and you can inspect the chosen data sources alongside your chat conversation, including previous prompts in the same conversation.
  • Export conversations: You can now export a full conversation as an image.

For more details, see Ask natural language queries with Sigma Assistant.

Use warehouse agents with Assistant and agents (GA)

If you have Snowflake Cortex Agents or Databricks Genie Agents, you can use those agents in Sigma. Ask questions from Sigma Assistant, or set up a Sigma agent (Beta) to call one or more warehouse agents as tools.

Admins can turn off warehouse agents to use only Sigma Assistant from the home page. For more details about setup requirements and usage, see Use warehouse agents with Sigma.

API

New API endpoints for managing AI chat history settings (Beta)

The following endpoints are now available to configure chat history for an organization in Sigma:

Reports

Export and download reports as PowerPoint (GA)

The ability to download and export reports as PowerPoint (.pptx) files is now generally available.

See Share and export reports.

Workbook elements

Manually trigger an AI column or cell run (Beta)

When configuring an AI column, you can now use the Don’t run automatically option to configure the AI column to only run when manually prompted instead of automatically. You can re-run an AI column query for a single cell at any time.

See Create AI columns (Beta).

Bug fixes and improvements

  • Version-tagging or source-swapping a workbook with a broken/missing connection reference now returns a clear “connection no longer exists” error instead of a generic failure.

  • When calling the List workbooks for a tag (GET /v2/tags/:tagId/workbooks) endpoint using API credentials assigned to a non-admin user with access only to specific tagged versions of a workbook and not the published workbook, no results were returned.

  • When having a conversation with Sigma Assistant or a Sigma agent over a dozen messages when using an AI provider that uses Claude Sonnet 5 as the reasoning model, a 400 error “The final block in an assistant message cannot be thinking” displayed.