Sigma brings governed data analysis and workbook building into ChatGPT

Sigma has released a plugin that lets ChatGPT and Codex users search, explore and query governed data in a Sigma organization, then turn the result into a saved workbook. The feature appeared in Sigma’s September 11 changelog and is distributed through the ChatGPT plugin marketplace (Sigma changelog; plugin documentation).

The useful distinction is that this is not merely a chat answer layered over warehouse data. Sigma says the assistant can locate a source, compose a query and return results in chat, but it can also create a workbook or edit an existing one. A user can ask for monthly sales by region, a top-customer chart or a year-over-year KPI; the plugin creates the artifact in Sigma and returns a link for review and continued editing (plugin documentation).

The governance boundary travels with the user

The plugin connects to the same tools exposed by Sigma’s MCP server and authenticates with OAuth. It inherits the user’s Sigma permissions: access to connections, data models and workbooks depends on the user’s account type and document-level grants. Creating or editing workbooks requires the corresponding Sigma permissions, while querying a specific document requires at least view access and querying a connection requires use access (plugin documentation).

That design matters for enterprise adoption. The assistant is not granted a parallel, universal data identity; it operates inside the access envelope already assigned to the person. Organizations using IP allowlists must also permit the AI assistant’s outbound addresses with API scope, according to Sigma’s setup requirements (plugin documentation).

It is a builder, not an administrator

Sigma documents several deliberate limits. The plugin cannot create data models, create workbooks directly in shared folders, share a workbook, delete one, restore a prior version, configure exports or edit tagged versions. New workbooks land in the user’s Your documents folder. Those constraints keep the assistant focused on analysis and draft artifact creation rather than administrative lifecycle operations (plugin documentation).

For teams that want query access without workbook creation, Sigma offers a read-only ChatGPT connection through its MCP server instead of the full plugin. The plugin itself is published as open source under Apache License 2.0 and bundles Sigma-maintained skills alongside the MCP tools (plugin documentation).

The release is another step in analytics interfaces moving outside the BI application while preserving the BI system as the governed execution and artifact layer. Here, the chat client becomes the front door, but Sigma still controls what the user can see and where the resulting analysis lives.