What the model sees
The privacy question, answered fully rather than compressed into a landing-page bullet.
Aggregates and schema, not rows
The engine computes over pre-aggregated cube cells. What reaches the language model is ranked segments and your schema's column names and types — never raw fact rows. The model never queries your source data directly.
The one non-deterministic step
The language model's only job is turning your sentence into a structured request — measure, window, filters — from schema alone. That parsed request is shown back to you before the deterministic analysis engine runs against the cube.
What this does not claim
This is an architectural description, not a certification. If your compliance review needs a PII guarantee in writing, ask your account team before relying on this page alone.