AI Powered

AI Powered covers three capabilities that work differently from one another. Parsing reads a document and fills the entry form in for you. The Agent answers questions about your own figures in plain language. Statement Import reads a bank statement you exported yourself, and it uses no AI at all. They are grouped because that is how you meet them in the product. They are explained separately here because they are metered separately, and treating them as one thing is the quickest way to be surprised by a limit.

​

AI Parsing

Parsing turns a financial document into a pre-filled entry form. It never saves anything on its own: you see every extracted field, you change what is wrong, and you confirm. An entry created from a parse you approved is recorded as one you made, because the thing that matters is your confirmation and not the extraction. Both paths draw on one allowance. Upload and connected mail are two doors into the same pipeline and share a single daily counter, so eight uploads leave two email parses on Basic rather than a separate pool.

​

What AI Parsing reads

Two groups of formats, and the difference is in how each is handled rather than in what you may send. Up to 10MB per file. A converted file has a second ceiling on the amount of text it produces, and a file that exceeds it is refused with a clear error rather than quietly shortened. A truncated spreadsheet loses entries you believe were captured, which is worse than a refusal you can act on.

​

What comes back, and what you confirm

Seven fields: type, amount, currency, date, the vendor or source name, a suggested group, and whether it looks recurring. Each carries a confidence value between 0 and 1, and anything below 0.6 is flagged on the confirm form so the uncertain values are the ones you look at first. Every value is a suggestion. Change any of them before you save, and add groups, a description or a receipt exactly as you would on a manual entry.

​

Your daily parse allowance

The counter resets at midnight UTC, and uploaded files and parsed mail both come out of it.

​

What happens to your document

The file is held only for as long as the parse needs it and is deleted the moment you approve or dismiss. Before any of its text reaches the reader it is stripped of personal information, so the figures are read without your name, your email address or an account number travelling alongside them. Nothing you upload is used to train a model.

​

The myClerkBook Agent

The Agent is a financial assistant you talk to, in the product and over WhatsApp. It is an add-on to either plan rather than a third plan, so it stacks on whichever plan you hold. See Plans compared for what it costs. Every tool that writes is off until you turn it on. Permission is granted per tool and per book, and starts denied. Listing your entries and fetching a summary are available; creating, changing and deleting are not, until you allow them in your integration settings. A tool call that fails, because a permission is off or a value did not validate, says so in the conversation and tells you what to do next. Included usage, or your own key. The add-on comes with 10,000 credits of usage a month on the in-house model, resetting with your billing cycle, and you are told when you reach 80% and 95% of them. If it runs out, the chat stops accepting new messages and your history stays readable. Nothing is deleted and nothing is locked. You can wait for the next cycle, or connect your own Anthropic, OpenAI or Google key, which runs the Agent under your own billing and draws no allowance at all. Your own key applies to the Agent only and never to parsing. What the Agent is given. A summary of the book and your last 10 entries. Never your full history, your name, your email address or an internal identifier. The conversation lives for the length of the session and is not stored afterwards.

​

Statement Import

Statement Import reads a bank statement you exported yourself, in OFX, QFX or CSV, and turns it into entries you select, categorise and confirm. No bank connection is involved and there is no facility to create one. You export the file from your bank, and myClerkBook reads the file you chose to give it. One import per calendar month, per book. That single rule produces both plan limits, because what differs between the plans is how many books you hold. Premium allowances are not pooled. An unused allowance on one book cannot be spent on another. The allowance is spent when you commit, not when you upload. Opening the wrong file, reviewing it and dismissing it costs nothing. The screen states what you have left and when it resets before you start selecting. A file carries at most 1,000 rows, on both plans. That is a guard against a pathological file rather than a plan lever. Basic’s cap of 100 entries a month still applies and is counted per row, so a Basic book with 40 entries left this month can commit 40 rows, and the screen says so before you begin rather than after. Duplicates are the feature, not a refinement of it. Re-importing an overlapping date range is normal use.

• OFX and QFX carry your bank’s own identifier for each transaction, so an exact re-import creates nothing.
• CSV carries no such identifier, so its rows are matched against what is already in the book on date, amount and description.
• Every suspected duplicate arrives unticked, with the row it collided with shown. Nothing is dropped quietly. In a financial record, silent deduplication and data loss look identical from the outside.

Your statement is never stored. A bank statement is the most sensitive document myClerkBook will ever handle: an account holder, an account number, a balance and every party you transacted with, most of which the product has no use for. So the file is read while your request is being served and discarded when it ends. It is never written to storage, never sent to any model, and no third party sees it. What survives is a short record that an import happened, holding the file name, a fingerprint, the period covered and three row counts.

​

Related features