Knox Finance

• "What's in the bank?"Before: log in to the bank. Now: the home screen, current to the last SMS, and it says how many rows are still waiting for a statement.
• "How long does the cash last?"Before: a rough guess. Now: one sentence at the top of Forecast, with and without new income.
• "Has that client paid?"Before: scroll the statement looking for a name. Now: the payment is matched to its invoice, less the tax the client kept, and waits for me to approve it.
• "What was that debit?"Before: squint at a UPI string. Now: one tap to categorise, and a rule so it never asks again.
• "Is the book right?"Before: hope. Now: a list of checks, like missing statement lines, double payments and gaps in the invoice series.
• "What goes to the CA?"Before: rebuild the statement in Excel. Now: a download in the bank's own layout, with our categories added.

The bank's statement is the truth. Everything before it is a placeholder that knows it will be replaced.

An iPhone Shortcut forwards every Axis SMS to the app the moment it arrives, so the balance moves with the money. When I pay cash or spot something early, I log it on the phone: the amount, what it was for, and a one-tap category. Both are provisional.

When the month’s statement is uploaded, each SMS or logged row finds its statement line, by the UPI reference both carry or by amount within a day, and steps aside. My category and note carry over. Nothing is counted twice, and nothing the statement didn’t confirm is quietly deleted. It goes on a list for me to look at.

The app is good at guessing which invoice a payment settles. Clients deduct TDS at different rates, sometimes pay three invoices in one transfer, and their bank names rarely match their company names. So it proposes, and I decide.

• SureThe client is named and exactly one invoice, or set of invoices, fits the amount.
• Pick oneThe client is named, but more than one option fits. I choose.
• Amount onlyNo name, but only one invoice anywhere fits. Shown, never assumed.
• Not thisRejected matches are remembered, so the same wrong guess is never offered twice.

Forecast starts with the month we’re in. What the bank has already shown, by category, and what is still due: salaries not yet paid, invoices not yet settled. Then month by month for a year: coming in, going out, left at the end.

No jargon in the labels. GST is “GST to pay”. TDS is “tax the client keeps”. The plan underneath is a list of amounts I can change, and the whole page moves when I do.

• HostingCloudflare Pages for the app, Functions for the API, D1 (SQLite) for the data. All on the free plan.
• Front endOne plain JavaScript file, no framework and no build step. It installs to the home screen and works as an app on the phone.
• LogicThe rules (merging, matching, forecasting, audits) are small pure TypeScript modules, so they can be tested without a database.
• Data inAxis statements as .xlsx, Wise's CSV, Amazon Business order reports, the SMS webhook, and Zoho read through its API.
• LoginOne code per device. It fails closed: with no secret set, the site serves nothing.
• BudgetD1's free plan counts every row a query reads, so every response reports its own cost and the dashboard is kept under a fixed budget.

• Write the rules down before the code. Most of this app is decisions: what counts as income, which source wins, when a match is sure. Each one has a line saying who decided it and when.
• Let the bank be the source of truth. Every other source is allowed to be wrong for a while, as long as it gives way when the statement arrives.
• Suggest, don’t decide. A wrong match quietly corrupts the book. Approval costs a tap.
• Plain words beat accounting words. “Tax the client keeps” needs no explaining. “TDS receivable” does.