BUSINESS INTELLIGENCE THAT SHOWS ITS WORK

Ask your warehouse anything. Get an answer you can defend.

Governed definitions, the exact SQL behind every number, and a version history you can query as of any past date. Read-only on your data, no migration.

Book a demo See how it works

You asked

Why did Northeast margin drop?

3.2 ptsdownvs May

Traced to

Freight surcharges · 2 distributors

Review the evidence

query 8f3c-42a1

gross margin %

Net of freight surcharges, excludes intercompany.

3 tables · 184,220 rowsread-only

Definition v3. The v2 wording is still queryable.

Approval before any action
Append-only audit log
Works with your MCP servers
Live in 4-6 weeks

One question, end to end

From a question to a number you can defend

A person asks. Every part of the answer opens - the read-only SQL, the rows behind the number, the approval before anything acts, and what the definition said at the time. Then ask for the whole brief instead of one number, or send it exploring on a scope you approve.

Ask in your own words

An analyst asks about the Northeast margin drop. xFalcon returns the plan it intends to run, and nothing runs until that plan is confirmed - it does not go looking for problems on its own.

Northeast margin is down. Where did it go?

Plan proposed

Read margin by region, Jan to Jun
Split the gap by freight contract and product mix
Size what is recoverable and by when

Confirmation gate

The plan runs only after the analyst confirms it. No run starts on a clock.

It writes the SQL. It never writes to your data.

Confirmed, it queries live warehouse data - read-only, enforced before a query reaches the database. It puts 80% of the 3.2 point drop on two freight contracts.

SELECT onlyRead-only on every backend

estimate_rows → 184,000 rows · confidence: high

Freight surcharges · 2 contracts80%
Product mix shift13%
Everything else7%

3.2 pts total decline vs May

Recommendation

Renegotiate both freight contracts. An estimated 2.4 points recoverable by Q4.

Follow any number back to the row

Every answer carries a query ID. Open it for the governed definition, the exact SQL, and the tables and columns it read - so the number can be checked rather than trusted.

query 8f3c-42a1Governed definition

gross margin % - net of freight surcharges, excludes intercompany

SELECT region, month, gross_margin_pctFROM   fct_margin_daily mJOIN   dim_freight_contract c  ON   c.contract_id = m.contract_idWHERE  region = 'Northeast'  AND  month >= '2026-01'  AND  month <= '2026-06'

fct_margin_daily184,220 rows · 6 columns read
dim_freight_contract14 rows · 2 matched
dim_region9 rows · 2 columns read

Recorded for this answer

12 tool rounds · 34.2 s · $0.041

Anything that would act stops here

Analysis is read-only, so acting on it is a separate decision. A recommended action stops at awaiting_approval for a named person, delivery runs on an MCP server your team connected, and both the approval and the invocation are appended to a log with no edit path.

awaiting_approvalDefault posture: ask

tooldraft_supplier_brief

serveryour-procurement-mcp

bound toarg digest 5b1f9c7a204e88d3

The approval covers those arguments and no others. Change one and it asks again.

Approved

D. Whitfield, VP Finance · 09:41:22Z

Append-only audit

mcp.tool.invokeapproval_required09:40:58Z

mcp.tool.approvalapproved09:41:22Z

mcp.tool.invokesuccess09:41:24Z

Your admins can read it. The product offers no way to edit or delete an entry.

Ask what the number meant in January

Correct a definition and xFalcon ends the old version and opens a new one, keeping the original wording for the life of the chain. A question dated to January answers with January's definition, not with today's.

gross margin %3 versions

v1to 4 Nov
v24 Nov - 12 Mar
v312 Mar - nowcurrent

as_of = '2026-01-31' → returns v2, verbatim

fact_check → conflict | redundant | novel | consistent

Nothing joins the chain unreviewed

A correction from negative feedback is drafted, not applied - it arrives inactive and waits for review. A personal correction stays personal until an admin promotes it org-wide.

Ask for the brief, not the number

The same session assembles the whole executive view - KPIs, charts, filters. Every tile is a saved query rather than a picture of one: re-open the report and it re-runs each tile's SQL against live data and redraws it from the same spec, with no model in the loop. You open it. Nothing refreshes on a clock.

Executive briefSaved as a report

Gross margin %
28.9%
down 3.2 pts vs May
Freight per order
$18.40
up 22% vs May
Orders
41,180
down 1.4% vs May

Gross margin % by region, Jan to Jun

Northeast28.9%
Midwest33.1%
West31.4%

Share of revenue

Northeast 34%Midwest 29%West 22%other 15%

What a saved report is

Not a picture of a chart - the SQL, the columns and the chart spec behind every tile. Re-open it and it re-queries live data and redraws itself, no model call in the loop. You open it: nothing here runs on a schedule.

Send it after the question you did not think to ask

AutoExplore is the one thing you invoke by name. It proposes a scope and a budget and waits; once you confirm, it runs 20 or more iterations, forms its own hypotheses and keeps only what a second query from a different angle confirms.

/ida:autoexploreInvoked by name

Scope confirmation

Directed exploration: Northeast margin. 20 iterations, 200 query max, 2 fact-check cycles. Proceed?

70%On the themefreight terms, contract by contract
30%Beside itthe dimensions nobody framed the question around

Standard of proof

One query is a hypothesis. Two that agree from different angles is a finding. A premise the data overturns is a finding too, written up like any other.

What it leaves behind

journal.mdevery hypothesis, query and dead end, written as it goes
briefing.mdthe findings ranked, for whoever will not read the report
report.docxfact-checked - every number cites the query behind it

No scheduler exists that could start this for you.

Illustrative scenario. The figures are an example, not a named customer.

WHAT YOU GET

Same question, same answer. And you can prove where it came from.

01
One definition, so one answer
Ask in your own words. The metric resolves against your company's approved definition rather than whatever the query happened to say, so two people asking the same question get the same number - and a critical annotation overrides anything that conflicts with it.
02
Traceable, and it learns what you correct
Every answer opens to its query ID, the exact SQL, and the tables and columns it read. Correct it once and the correction is versioned and reviewed before it counts, then reused - so nobody re-explains the same thing next quarter.
03
Results in the shape you asked for
A table, a chart, or a saved report that re-queries live data and comes back in the format you captured. Personal work stays personal: nothing you save becomes the company's answer until an admin promotes it.

Explore all features See pricing

LIVE DEMOS