Every accounting system has AI now. Only one hands the work to yours.

Each of these systems automates most of its matching internally. Their APIs and MCP servers read reports or push transactions in. What is left over — the unmatched, the uncategorized, the questions — lands in a review queue inside their app for a person to clear.

FinStat hands those items to your AI, each with its evidence. Your AI clears what the documents settle and asks you only about the rest. The books come back finished, with the open questions named.

FinStat is there for your workflow; it does not define it. Rillet’s Aura Flows and the other systems’ apps give you their workflow to learn and follow. FinStat is built for your AI: you talk in goals — “rebuild last year from this export”, “what doesn’t tie?” — and your AI handles the details. There is almost nothing to learn.

Rillet, QuickBooks and Digits are built around their own apps. Your data is captured into their cloud, your books live there, and the monthly subscription pays for keeping them there.

FinStat is built around the accounting. Every document it checks is sealed and handed back to you, so your data can live on your own machine, where your AI can read it directly, as well as in your FinStat workspace. Pull it, store it, change the view, take it to another system. You pay only for the work FinStat does, never for holding your data.

Chart-centric systems record each transaction as a posting to an account, so the chart is the books. FinStat works far upstream of the chart: it uses mathematical structures that resolve every obligation and every money movement from the documents — who owed whom, what was paid, what cleared. That is the reality. A chart of accounts is an opinionated projection of it.

That is why FinStat can hold several charts over the same facts — a tax view, a management view, a lender’s view — and why the books stay accurate when the chart changes. Guessing a category from a transaction’s description is what rules engines and machine-learning categorizers do, and the hard cases are left for you to clean up. FinStat categorizes from what actually happened: the payment event, the counterparty, the obligation it settles, and the accounting decision behind it. Its AI goes where that work is hard: reading large and difficult statement PDFs, reconciling them to the bank’s own printed figures, identifying counterparties, with EIN where public records carry one, and making FAR-level accounting decisions.

Deterministic engines settle everything arithmetic can, exactly and repeatably. The frontier models get only what is left, split into bounded questions answered in parallel, so a job costs and takes what its hard parts need and no more. It is built the way high-performance trading and simulation systems are built, by people who built those.

With FinStat connected, your AI turns PDFs into data it can trust. FinStat reads fourteen document types in three mathematical forms, and checks each one against the law of its form before it is accepted:

• Registers — checking, savings, wallet, credit card, line of credit, loan and brokerage statements. Opening balance plus every transaction equals the closing balance (except brokerage, whose value moves with price).
• Obligations — invoices, bills, credit notes and debit notes. The line items plus tax equal the total.
• Itemizations — receipts, processor settlements and payroll statements. The detail equals the one payment it explains; gross less fees equals net.

A document that is missing required data, or does not satisfy its law, is refused with the reason. One that passes is sealed. The data stays in clear text for anyone to read; the seal means any later change is detectable, so every time it is used after that, it is known to be what FinStat checked.

From there your AI can produce import files for QuickBooks, Xero or Sage, render a clean PDF from data, and verify any document it is handed. Where your client has no feed, the PDFs are the books; where it has a custom feed, your AI maps it and hands it to FinStat.

Simply ask your agent, and FinStat will generate a viewer for your books: one standalone file you open in a browser, with the accounts, the charts, the documents and every number traced to its source. Your AI shapes it to your situation — the views a lender needs, the ones a tax preparer needs, the ones you check every Monday — and it works offline, from your own machine.

Because the chart is a projection of the facts, it can be as rich as the business needs without anyone re-keying a transaction.

• Dimensions — company, region, district, or division, location and project: each account sits at a level of the organization, and reports roll up at any level.
• Summary nodes — groups that total their accounts. A rollup is computed from what sits beneath it, never posted as an extra entry.
• Seven numbering styles — assigned, three-, four- and six-digit, dotted, segmented and alphanumeric. Switch the display and the account stays the same account.
• Descriptions and instructions — every account carries its purpose and what does and does not belong in it: the working instruction your AI and FinStat use to place new transactions.
• Statements from account type — the balance sheet and P&L are laid out by what an account is, not by its number, so any numbering style renders clean statements.

FinStat is fast. Analyze an existing QuickBooks Online export in a few minutes, and import it in a few more. Work that would take a person days, or an AI millions of tokens of reading, runs in FinStat’s engines.

FinStat makes your AI highly efficient. It takes a task from your AI, breaks it into portions, and answers each one with its own engines or with a stable of tuned frontier-model calls from US providers — saving time and tokens. The mathematical structure of the data makes a question far cheaper for FinStat to answer than for your AI to work out from raw rows.

About fifty tools let your AI stay on the high-level task — your goal — and get the job done efficiently.

Craig’s Design and Landscaping Services is the sample company Intuit ships with QuickBooks Online. We exported it, rebuilt the books, and ran the audit. From the zip alone, it caught all five of the company’s known errors:

Each one comes back as a question with the entries behind it. The whole run, from the zip to the finished audit, took about four and a half minutes and $1.61 of model time, and raised 33 issues in all.