Plain English in, grounded answers out — every number computed from the underlying dataset.
How this works in production: the agent runs server-side on a schedule, reads the client's
live semantic model (Power BI / Microsoft Fabric), writes the brief, and delivers it to email or
Slack — with row-level security, so each leader sees only their org. In this public demo,
briefs are pre-generated at build time and Q&A runs against the same aggregates locally —
no data or API keys ever reach the browser. Same architecture, safe to put on the open internet.