The AI analyst that refuses to be wrong.

Watch it refuse the question it can't answer honestly.

Ask it something the data can't support and it won't guess — it declines, shows you exactly why, and points to what it can prove. Try both questions:

This is the whole philosophy in one interaction: it would rather tell you the truth than tell you a number.

Everyone else gives you a template. Columnly picks a form — and shows its working.

It reads your data — the dimensions, the slices, whether there's a date axis or a geography or a real sequence of stages — and works out which of seventeen distinct layouts (7 for briefs, 10 for dashboards) actually fits. Then it shows you the options, recommends one with a reason, shows the ones it ruled out and why, and lets you overrule any of it.

The plates show where things go, not how they will look. Every shape uses the same palette, the same type and the same traceable figures — a number is one click from its source in all of them. What changes is the order a reader meets things in, and how much they have to hold in their head at once.

On a real board this popup is anchored to the number itself — because the tie to that figure is the point.

Point Columnly at your live business data.

Two ways in, from one "Add source" entry. Connectors are read-only, forever — so Columnly can never write back to your CRM.

Most AI needs clean data to be right. Columnly cleans it first.

It detects the mess, quotes your own values back as a yes/no, and shows every fix in a diff you can undo — before it computes a single number.

Your data's already clean and verified here — why export it to a chatbot that'll guess at the numbers?

Two models. One answer. It ships only when they agree.

Every data question is solved twice, by two different-class models. One decomposes and computes it; an independent second re-derives the whole thing from scratch — it never sees the first one's work. The answer ships only when the two agree within a strict tolerance. A genuine second opinion, not a rubber stamp.

Things no other AI product does.

Not a feature race — a different standard. Every one of these is live in Columnly today, and we haven't seen any other AI data tool do them.

A dashboard is an instrument. A brief is an argument.

Because the reader's job differs, the output does too. A dashboard is a dense console you monitor. A brief is a single reading column you read once — thesis, evidence, tension, recommendation.

app.columnly.ai / dashboard · booked against banked

Dashboard — an instrument you monitor. A real board over 10,000 rows, all figures verified — here with one figure's Source → Computation → Reconciliation open.

app.columnly.ai / brief

Brief — an argument you read once

The same verification engine, in a warm editorial register — thesis, evidence, tension, recommendation, read top to bottom. Every figure is still one tap from its source; the identity changes, the guarantee never does.

These are real Columnly outputs — the click-to-trace popup is open in each.

Your brand on every artifact — without ever lying with colour.

Set the company's identity once, and every dashboard and brief anyone builds comes out in your colours and logo. Choose a palette and a register for each kind — briefs default to light (they're read and printed), dashboards to dark (they're watched for hours).

Give each person a budget. Never get a surprise bill.

An owner splits one org credit pool into per-member budgets — power users get more, occasional users get less — set once a year. Roles decide who can build, finalize, and manage connectors.

app.columnly.ai / settings / members

Free viewers

Only pay for the people who build.

Share a live dashboard or brief with any number of colleagues as free view-only viewers — and every viewer can click any number and watch it reconcile. Only builders pay.

Roles & credits

Owner / Admin / Member, enforced server-side. Removing a member returns their credits to the pool — and credits run annually, not monthly, so there's no monthly surprise.

A real workspace's members panel — the org credit pool split into per-member budgets, set once a year.