How AI Bookkeeper works

Sync QuickBooks Online, Xero or Zoho Books and your bank transactions start arriving on their own — no separate bank link to maintain, because we work through the connection your accounting platform already has.

Paperwork comes in alongside them. Upload bills, receipts, invoices and sales receipts as PDFs or images, straight from wherever they land — a scanner, a phone camera, an email attachment. A crumpled receipt photo and a clean PDF invoice are both fine.

Signing up takes an email and password, or a Google account. Everything is filed against a specific business, so if you keep books for several clients, each one's transactions and paperwork go into their own isolated set from the moment they arrive.

AI Bookkeeper does the first pass on every document, and scores its own certainty as it goes. OCR reads the file into text, then a language model interprets it the way a bookkeeper would: it extracts the vendor, totals, tax, dates and individual line items as discrete fields, not one undifferentiated blob of text.

The same pass works out what kind of document it is — a bill, a receipt, an invoice or a sales receipt — checks it against what you have already filed so the same invoice does not get entered twice, and matches the party named on it to a vendor or customer you already have, including when the OCR reads a trading name or a slight misspelling.

Your bank transactions get the same treatment. Each one is coded against your own chart of accounts using your history, your vendors, your classes and your tax logic — not a generic model of what a business like yours usually does — and then matched to the bill or receipt it belongs to, so the document and the money it accounts for end up on the same record instead of in two separate queues.

Anything that does not add up is flagged rather than quietly applied: a duplicate, a coding error, an amount out of line with what that vendor normally charges, a transaction with no supporting document. You can set the thresholds and rules that decide what needs a human and what does not.

Every extracted value comes back with its own confidence score. The score is not decoration — it is what tells you, at a glance, which fields deserve a second look before you approve.

Every extracted value links back to the exact spot on the original document it was read from. Click the vendor name, the total or the tax amount, and the source document opens with that region highlighted — a bounding box drawn around the precise place the value came from.

This is the part most AI bookkeeping tools skip, and it is the reason we describe the product as the AI bookkeeper that shows its work. An extraction you cannot trace is an extraction you have to either trust blindly or re-check by hand, which defeats the point of automating it. With click-to-highlight provenance, verifying a field takes a glance instead of a hunt through the PDF.

It matters most when something looks off. If a total seems wrong, you do not debug the AI — you click the field, look at the highlighted region, and see in seconds whether the document really says that or the extraction missed.

Every document reaches your review queue prepared rather than blank. The fields are already filled in, each one carries a confidence score, and each one is a single click from the place on the document it was read from. Your job is to check and approve, not to retype.

That is the actual saving. The slow part of bookkeeping is not deciding whether an invoice is right — it is typing it in and then hunting through a PDF to confirm what you typed. Both of those are gone.

You approve or reject each item and can attach a memo explaining the call. If your organization requires it, approval can be restricted so that the person who uploaded a document is not the person who signs it off.

Every action — every approval, every rejection, every memo, every edit to an extracted field — lands in an append-only audit trail. Entries are added, never rewritten or deleted, so there is always a permanent record of who decided what, and why.

Each client's records are fully isolated. Separate organizations and businesses each keep their own documents, contacts and history, with per-business access control on top, so a bookkeeper managing several clients never has one client's data anywhere near another's.

Combined, that is the whole promise: every number has a source you can click through to, and every change has a record you cannot quietly edit.

Reconciliation arrives prepared rather than started from scratch — coded, matched, and with the exceptions already surfaced — so closing a period is a review of what has been flagged instead of a hunt for what might be wrong.

Approved work syncs back to QuickBooks Online, Xero or Zoho Books, so your accounting platform stays the system of record and nobody is re-keying between two tools. When a transaction needs the client's input, you can ask them about it — or request the missing document — from inside the app, instead of starting an email thread you then have to chase.

Reports come out in a click and can be shared as they are. If you keep books for several clients, one dashboard covers all of them, with each client's records isolated from the rest.

We are early-stage, and we would rather say that plainly than pretend otherwise. Pricing will be published at launch — no sales calls required.