Sigma’s upgraded Assistant asks before it guesses

Sigma Computing has made an upgraded version of Sigma Assistant generally available to all customers, shifting its natural-language analytics interface toward multi-turn exploration rather than one-shot answers. The release adds clarifying questions, broader source search, a visible analysis trail and persistent conversation history, according to Sigma’s Sept. 22 announcement.

What shipped

When a request is ambiguous, Sigma Assistant can now ask follow-up questions, state assumptions and flag caveats instead of immediately choosing one interpretation. Sigma gives “how did bookings perform last quarter?” as an example of a question whose meaning can vary between organizations. The company says the upgraded Assistant is also designed to answer open-ended questions more deeply and return follow-ups within a thread faster, though the announcement provides no benchmark or latency figures for those claims (Sigma).

Administrators can configure the Assistant to search all data sources a user can access or restrict it to a curated shortlist. Sigma says existing row-level security, column masking and connection permissions continue to apply, so widening discovery does not grant the Assistant access beyond the user’s own query rights (Sigma).

The answer now carries a receipt

Users can expand an answer to inspect the sources, underlying tables, generated SQL and step-by-step analysis. They can then move the result into a Sigma workbook for deeper inspection or turn it into a reusable artifact. That matters operationally because the conversational answer is no longer the only record of what the system did; the query and source path remain available for review (Sigma).

Chat history is also generally available, although an administrator must enable it. Sigma says past conversations retain snapshots of their insights, can be exported as images and can be promoted into reusable workbooks. This creates a distinction between a preserved conversational result and a workbook that colleagues can rerun against current warehouse data (Sigma).

What is not GA

Sigma is separately previewing a redesigned, AI-first homepage, but that experience remains in private beta. The company also says full support for Snowflake Semantic Views and Databricks Unity Catalog Metric Views is still coming. The current Assistant can search warehouse tables and data models, while administrators decide whether the default search scope is broad or curated (Sigma).

The release draws a clear product boundary: Sigma Assistant is the out-of-the-box experience for asking and exploring, while separately configured Sigma Agents are intended for curated workflows with their own instructions, sources and actions. For enterprise analytics teams, the meaningful upgrade is therefore not autonomous action yet; it is a more inspectable way to resolve business ambiguity before generated SQL becomes an answer (Sigma).