📄 Document Intelligence

Somebody in your office is
reading by hand right now.

Statements, invoices, receipts, delivery notes, contracts, applications — read, checked and summarised in seconds instead of hours. Every figure is calculated in code, never guessed by a model. And the file is processed in your browser, never stored on our servers.

See what it reads ↓
Pay as you go
₦2,000/document
Volume plans
from ₦150,000/month
Lenders & high volume
Book a demo
The problem

Reading documents is the most expensive thing your smartest person does.

Somewhere in your business, a capable, well-paid human is holding a highlighter over a six-page bank statement, adding up credits on a phone calculator, and squinting to work out whether the closing balance actually follows from the lines above it.

They do this on Monday. They do it again on Tuesday. By Friday they've read forty documents, they're tired, and tired people miss things — a duplicated invoice, a delivery note that doesn't match the order, a renewal clause that quietly auto-renews for another year.

It isn't a skill problem. Nobody stays sharp on page five of the fourth statement of the afternoon. It's the wrong job for a person.

What it's costing
Careful read of one bank statement, by hand20–40 min
Same statement, extracted and checkedSeconds
Invoice checked line by line, by hand5–15 min
Contract read for dates and obligations30–60 min
Chance of a missed detail at 4pm on a FridayHigh

These are the ranges we hear when we ask. The first thing we do on a demo is time your real process on your real document, so you're deciding on your numbers, not ours.

Six document types, one engine

Pick the one that's eating your week.

Each type has its own extraction rules, its own checks, and its own red flags. Tap through them — the detail is the product.

◆ Live tool — on this page

Try it on your own document.

Drop a file in, pick its type, read the verdict. It is processed in your browser, never stored on our servers — the file is never uploaded. Extracted text is sent to our API so the plain-language summary can be written.

Rather we ran it with you? Book a demo and bring the messiest document you have.

How it works

Upload. Read. Check. Verdict. Discard.

Five stages, and the order matters — especially the last one.

1

Upload

Drop in a PDF or a photo. It's opened and read inside your own browser — it isn't uploaded to a document store somewhere and left there.

2

Extract

The text layer is pulled out directly. Scans and photos fall back to on-device OCR, then table structure is rebuilt so columns stay columns.

3

Check in code

Every figure — totals, VAT, running balances, quantities — is computed and reconciled by ordinary arithmetic in code. Not by a language model.

4

Verdict

A plain-language summary: what the document says, what checks out, what doesn't, and exactly which line to look at yourself.

5

Discard

You take the output. The document isn't kept — there's no archive of your customers' statements sitting on our side waiting to be breached.

The one rule we won't bend: all arithmetic happens in code.

A language model is excellent at reading messy narration — “TRF/FRM ADEBAYO O/RENT JUL” — and terrible at being trusted with a sum. So we split the job. The model classifies and writes prose. Code does every single calculation and comparison. A wrong total on a financial document isn't a bug you apologise for; it's the end of the product's credibility. This is why we built it the slow way.

Privacy

Your documents are not our asset.

You're handing over other people's financial lives. Here is exactly what happens to them — including the part most companies leave out.

🖥️

Read on your machine

The file is processed in your browser, never stored on our servers. The PDF or image itself does not get uploaded to us and does not sit in a bucket with your client's name on it.

🗑️

Nothing retained

We keep no copy of your document and no archive of what was in it. When you close the tab, the working copy is gone. There is nothing on our side to leak, subpoena or sell.

🧮

The numbers never leave code

Balances, totals and reconciliations are computed locally in code. No figure in your verdict was produced by a model being asked to do maths.

The part we won't hide from you

Some products claim your document “never leaves your device.” We won't say that, because for us it isn't strictly true. After the file is read locally, extracted text — narration lines, descriptions, clause wording — is sent to an AI API so it can be classified and summarised in readable English. The file isn't sent. The text is.

That distinction matters if you handle other people's financial data, so you should hear it from us before you buy, not discover it in a policy document afterwards. If your compliance rules can't allow that, say so on the demo — for some workflows we can keep the whole thing local, and for some we honestly can't. We'll tell you which one you're in.

Pricing

Pay per document, or per month.

Start with one document. Move to a plan when reading them stops being an event and starts being a routine.

Per document

₦2,000/document
For occasional documents and first tests.
  • Any of the six document types
  • Full extraction + code-checked figures
  • Plain-language verdict you can act on
  • Flagged lines pointed to by page
  • Nothing stored afterwards
Send one document
🔥 MOST USED

Volume plan

from ₦150,000/month
For teams reading documents every week.
  • Everything in per-document
  • Monthly document allowance
  • Your document types set up and tuned
  • Consistent output format for your team
  • Priority support

Lenders & large volume

Book a demo
Lending, credit, audit and anyone at real scale.
  • Statement analysis at review volume
  • Tamper and consistency checks
  • Accuracy verified per bank format first
  • Output shaped to your credit process
  • Scoped and priced after the demo

Volume plans are 50% upfront / 50% on delivery for the setup portion. We won't quote a lender price before we've seen your statement formats — anyone who does is guessing.

Who it's for

If someone reads for hours, this is for them.

Lenders & credit teams

Someone reads a statement before every decision. Speed matters, and so does spotting the one that's been edited.

Accountants & bookkeepers

Invoices and statements in volume, from clients who send them as blurry photos on WhatsApp.

Logistics & distribution

Delivery notes against orders, week after week. The mismatches are small, frequent, and expensive when nobody catches them.

Property & facilities

Tenancy and service contracts with renewal dates and notice periods buried on page nine.

Admissions & HR

Application packs where the real work is checking what's missing and what contradicts what, not reading the prose.

Any owner doing it themselves

If reading documents is what's still on your desk at 9pm, that's the clearest case of all.

When we'd tell you not to bother

If you read fewer than a handful of documents a month, a careful pair of eyes is cheaper than any software — including ours. And if your documents are handwritten, this is the wrong tool: extraction on handwriting is unreliable, and unreliable is worse than slow when there's money on the line.

Questions

The things people actually ask

Why are bank statements marked BETA?

Because every bank lays its statement out differently, and a statement reader that works on one bank's PDF can misread another's badly. We verify accuracy per bank format before we let anyone rely on it, and until a format is verified it stays labelled BETA. We'd rather carry an unglamorous badge than have you make a lending decision on a number we hadn't checked.

Can it tell if a statement has been edited?

It runs the checks that catch the common cases: reconciling the running balance line by line so the arithmetic has to hold all the way down, inspecting the PDF's own metadata and producer, and looking at whether the text layer looks bank-generated or word-processor-generated. Those catch a lot. They are not a forensic guarantee, and we won't sell them as one — the output flags what looks wrong and tells you which line to check yourself.

Does the model do the maths?

No, and this is deliberate. The model reads narration and writes the summary. Every total, balance, VAT figure and comparison is computed in code. A model that is right 99% of the time about a number is a liability on a financial document, because you cannot tell which 1% you got.

What actually gets sent to an AI?

The file itself is processed in your browser, never stored on our servers. Extracted text — the narration lines and wording — is sent to an AI API to be classified and summarised. If that's a problem for your compliance rules, tell us on the demo and we'll be straight with you about whether your workflow can be kept fully local.

What if it gets something wrong?

Everything it flags is pointed to a specific line and page so you can check it in seconds. It's built to surface things for a human to confirm, not to make the decision for you. It replaces the reading, not the judgement.

What formats does it take?

PDFs — text-based ones work best — plus scans and phone photos, which go through OCR on your device. Handwritten documents are out of scope. Password-protected PDFs need to be unlocked before they're read.

Can I get the output into my own system?

Structured output is the whole point — the extraction produces fields, not just prose. Where it goes next depends on what you run, so we scope that on the demo rather than promise an integration we haven't looked at.

How do I start?

Send one document at ₦2,000 and judge it on your own paperwork, or book a demo and we'll run two or three with you live — ideally including one you already know has a problem in it.

Bring the document you dread.

The messy one. The six-page statement, the contract nobody has read to the end. We run it on the call, and you keep the output whether or not you buy anything.

Ask on WhatsApp