Now in private preview · cross-vertical

The agentic data analyst
for every question your
business asks.

Connect your data. We catalog it, model it, and answer business questions across all of it — with sources, SQL, and citations.

Get a demo See how it works
Ask · Live workspace
01 · Question
Routes through
Customer Subscription Plan
02 · Answer Generated · 2.4s

MRR fell −$42.1K driven mostly by Pro-tier churn in the EMEA region — voluntary cancellations on annual renewals, not gross adds.

MRR Δ vs. prior month · $K by movement
New MRR
+$28.4
Expansion
+$12.7
Reactivation
+$3.1
Contraction
−$8.9
Voluntary churn
−$52.4
Involuntary
−$25.0
Sources Synced 14m ago
QuickBooks Postgres · subs GA4 Shopify
View SQL Inspect canonical
01 · The problem Why this exists

Modern teams have more data and more tools than ever. And still need a human in the loop for every novel question.

01

Every new question becomes a ticket.

Your data team is the bottleneck for every question that isn't already on a dashboard. The CEO asks at 9pm. The answer arrives Thursday.

02

Dashboards answer yesterday's questions.

Last quarter's KPIs, rendered beautifully. But the question that matters today is the one nobody built a tile for.

03

"ChatGPT over your warehouse" gets it wrong.

Generic LLMs hallucinate joins, miss your business logic, and can't show their work. You can't forward an answer you can't verify.

02 · Watch it answer Live demo

Four questions buyers actually ask.
Answered with receipts.

Every answer shows the canonical concepts the agent reasoned over, the sources it cited, and the SQL it generated. Click any card to see its receipts — this is the wedge against generic AI over your warehouse.

Finance · Variance 01 / 04
"Why was revenue down last month?"

Down −$118K vs prior month. Paid social attributable revenue fell 38%, while organic & email rose. The shortfall is acquisition-side, not retention.

Paid social
−38%
SEO / organic
−22%
Direct
−4%
Email
+8%
Affiliate
+11%
Revenue Channel Period QuickBooks GA4 Postgres
Growth · Segmentation 02 / 04
"Which customers should we target for the Q3 launch?"

1,247 accounts match the activation profile from your last successful launch — Pro/Growth plan, ≥$24K ARR, active in the last 14 days.

Q3 launch · Activation-ready
Persona · Approved
Plan tierPro · Growth
Last login≤ 14 days
ARR≥ $24K
IndustryB2B SaaS
Trigger eventActivated v3 feature
1,247 customers · live count
Customer Subscription Plan Activity Postgres QuickBooks Knowledge: launch-playbook.pdf
Marketing · Performance 03 / 04
"How is paid social performing vs. last quarter?"

Total paid social spend +14%, CAC +27%, attributable revenue +6%. Meta Reels efficiency improved; Meta Feed degraded — note: iOS attribution caveat applies.

Meta FeedMeta ReelsGoogle SrchGoogle YTGoogle Disp
Channel Spend CAC Revenue Meta Ads Google Ads GA4 QuickBooks
Executive · Cadence 04 / 04
"Send the board this report every Monday at 8am."

Scheduled. Snapshot will refresh from the canonical model on each run, and arrive as a PDF + PPTX to 4 recipients.

Scheduled
CadenceMondays · 8:00 AM PT Recipientsboard@ · execs@ · 4 people FormatPDF + PPTX · live snapshot Next runMon · Jun 02 · in 3 days
Auto-refreshes from canonical model on each run
Report Schedule Recipient Workspace · canonical report
03 · How it works The architecture

Five layers between your warehouse and a correct answer.

The data side discovers and proposes. Your team approves a canonical model of the business. The agent reasons over that model — never raw schemas — and learns from what it ships.

Data side
Agent side
01 · What exists

Source catalog

Connectors discover every table, column, dimension, and API endpoint across your stack — and stay in sync.

02 · What it probably is

Inference

Statistical + LLM-assisted typing proposes entities, keys, and relationships — surfaced for human review.

03 · What it MEANS

Canonical model

A human-approved layer of business concepts — Customer, Order, Channel — that the agent reasons over.

04 · What worked

Runtime feedback

Every answer, citation, and follow-up updates a corpus of grounded reasoning the planner can learn from.

05 · Where to route now

Planner

Decides which canonical concepts, joins, and tools a new question needs — never raw schemas.

The agent never sees raw schemas — only the canonical layer your team approved.
This is the wedge. Every defensibility claim flows from it.
04 · Connectors Plug in any source, ask any question

The agent works across all of these in one conversation.

We're not an ETL vendor — we consume the world your data teams already built. Connectors are growing every release; what's live today and what's next, in one list.

Warehouses & DBs
BigQueryLive
D
DatabricksComing soon
Google SheetsLive
MySQLLive
PostgresLive
RedshiftComing soon
SnowflakeComing soon
Analytics APIs
GA4Live
GA
Google AdsComing soon
MA
Meta AdsComing soon
Financial & Ops
PlaidPrivate alpha
QuickBooks OnlineLive
Commerce
ShopifyLive
CDP
A
AmplitudeComing soon
M
MixpanelComing soon
M
mParticleComing soon
S
SegmentComing soon
T
TealiumComing soon
Documents Upload + persona grounding live. Agent-time retrieval is on the roadmap.
Docs · KBLive
PDFs · KBLive
Today · 10 connectors live Coming soon · cloud warehouses, ads APIs + CDPs Custom · request a connector
05 · Surfaces What you actually get

Six surfaces. One semantic layer underneath.

Surface · 01Live

Ask

Conversational analysis across every connected source, with citations.

Surface · 02Live
$2.4MMRR

Dashboards

Persistent views of the canonical model your team aligned on — composed from answered questions or built directly.

Surface · 03Live

Reports

Scheduled PDFs and PPTX delivered to inboxes — board decks on autopilot.

Surface · 04Live
1,247
842
318
87

Personas

Generate target cohorts grounded in your knowledge base; query them like first-class segments.

Surface · 05Live
PDF · 12p
PDF · 16p
PDF · 20p
PDF · 24p

Knowledge

Upload PDFs, briefs, and documents — the agent grounds answers in your context.

Surface · 06Live
Customerapproved
Subscriptionapproved
Channelreview
Plan tierapproved

Semantic

Review and approve the canonical model the agent reasons over. The buyer keeps the keys.

06 · Defensibility Why answers are correct

Receipts aren't microcopy. They're the product.

Every answer can be inspected, cited, and traced back to the canonical concepts and source rows that produced it. Forward the answer to your CFO; the receipts go with it.

Inspector · receipts for one answer
churn-by-region · run #4218
Canonical concepts used
Customerapproved Mar 14
Subscriptionapproved Apr 02
Plan tierapproved Feb 11
Regionapproved May 06
Source citations
Postgressubscriptions2.4M rows · 14m ago
Postgrescustomers482K rows · 2m ago
GA4sessions11.3M rows · 1h ago
QuickBooksinvoices38K rows · 6m ago
Generated SQL · canonical layer 
SELECT  region,
        SUM(canceled_mrr) AS churn_mrr
FROM    canonical.subscription_movement
WHERE   plan_tier IN ('Pro','Growth')
  AND   event_type = 'voluntary_cancel'
  AND   period = DATE_TRUNC('month', CURRENT_DATE - INTERVAL '1 month')
GROUP   BY region
ORDER   BY churn_mrr DESC;
01

Canonical-first

The agent never queries raw tables. Every question routes through an approved, human-reviewed model of your business — the same Customer, Order, and Channel your team has already aligned on.

02

Provenance

Every result carries refs back to source, sync batch, and freshness. Revoke a source, and impacted answers update — automatically, with a record of what changed.

03

Citations

Every claim links to the SQL we ran, the canonical concepts we used, and the underlying source rows. Forward the answer; the receipts go with it.

04

Reviewable proposals

New attributes, metrics, segments, and relationships enter a review queue — not the live model. The buyer keeps the keys to what "Customer" means.

Stop building dashboards. Start answering questions.

A 30-minute walkthrough on your data shape, your top four questions, and what the first canonical model would look like.

Thanks — message received.

We'll be in touch within a business day.

Typical setup2 weeks
Canonical modelYour team approves
Vendor lock-inNone — your warehouse, your SQL