One conversation, four scenes. Watch the chart-selector pick the right visual at every level (line → bars → table), follow-ups ship pre-validated SQL so the runtime skips nl-to-sql, and the conversation crosses Engines into Advisory without losing context.
Walkthrough · 1 of 4
Step 1 — Type a revenue question into the Reactor. The LLM router resolves the system (Salesforce + Snowflake), the nl-to-sql atom builds the query, sql-executor runs it, chart-selector picks a line chart, and the narrative streams in.
March revenue fell -18.3% MoM (Feb → Mar). The dip is concentrated in Enterprise; SMB and Mid-market are roughly flat. April recovered, suggesting a one-month event rather than a structural change.
▸ Inspect SQL
SELECT date_trunc('month', closed_at) AS month, SUM(amount) AS revenue
FROM salesforce.opportunity
WHERE stage = 'Closed Won'
AND closed_at >= date_trunc('year', now())
GROUP BY 1 ORDER BY 1;