Xtracto roadmap and release notes

Done

“Done” means written, tested and verified on a real phone. It does not mean it is already in the version you have installed. The one in the closed test today is 1.0, and out of all this it carries four things: reading the alerts, entering by hand, choosing which apps it listens to, and the total with its breakdown —which there is the month’s only, without the other periods—. The rest of this box is waiting for the next version, and the release history at the end of the page says what each one carried.

Reading your bank’s alertsWhen your bank tells you about a card payment or a direct debit, Xtracto reads that alert and turns it into a transaction, with its amount, its merchant and its card. Without typing anything and without talking to any server.
Entering an expense by handFor what the bank never announces: cash, a store card, a payment someone else made. And if the alert for that same payment arrives later, it replaces your entry instead of counting it twice.
Deciding which apps it listens toThe catalogue is a closed list of finance applications, and within it you choose one by one. Turning one off is not hiding it: the service does not even open its notifications.
The period total, and something to compare it withDay, week, month, year or the whole history. The comparison with the previous period is trimmed to the same point: setting half a month against a whole one tells you nothing.
Naming your cards and accounts“••0634” means nothing to anybody and “Trade Republic” does. The name is presentation only: the four digits stay visible, because they are what lets you reconcile the breakdown against your bank statement.
Spending categoriesA closed list of fourteen, and what gets classified is the merchant, not the transaction: you classify the supermarket once and every purchase there is classified, the past ones and the ones still to come.
A summary you can read at a glanceThe total says how much, the heatmap when, the category ring on what and the merchant top where. It is drawn by hand, with no charting library: every dependency is weight, surface to audit and one more thing that might want to touch the network.
A monthly budgetWith a pace marker, because having spent 60 % on the 5th and having spent it on the 25th are not the same news, and a plain bar paints them identically.
Choosing which charts you seeEvery new chart makes the summary longer. The ones you do not care about are turned off from the summary itself, which is where you can see what is in the way, and not from a settings screen.
Money in and transfers, told apartMoney in shows green with a “+”; a transfer shows muted and unsigned, because moving money from one of your pockets to another is neither spending nor income. Totals still count outgoings only, so that the day the first credit lands the month does not inflate.
TabletThe summary in three columns, and the lists with the detail beside them instead of one narrow column in the middle of an empty screen.

In progress now

What is being worked on these days, ordered by what it costs most to leave unfinished.

Charges that do not say who they are fromSome alerts only say that an amount was charged to the account, with no merchant. The same sentence, word for word, covers an instant transfer you sent and the monthly settlement of a credit card, and nothing in the text tells them apart. Adding the rule blindly would count that card’s purchases twice: each purchase first, and then the receipt that charges them all together. The way out is for you to declare in Settings which card is a credit card and which account it settles against, then reconcile by exact amount. It is the only thing on this page that is counted wrong today, which is why it comes first.
Getting money in to read itselfThe rules for incoming money are already written —transfer received, interest, salary— and what is missing is checking them against real alerts. Written is not verified, and a salary shows up once a month: this moves at the pace of the calendar, not of the keyboard.
More banksThe list of recognised banks grows with the formats people send in. When an alert is not understood, it can be sent without sending any of your data: the app replaces amounts, dates, merchants and numbers with placeholders and shows you the exact text before it goes.

Decided and not started. Nearly all of it waits on the same thing: the app seeing the money that comes in as well.

Money in against money out, in barsWhat came in and what went out in each period, at a glance. Waiting on incoming money reading itself.
Net balance and savings rateComputing them is trivial, which is exactly why they are not there. Until the app sees every credit, the net figure for a month in which a salary landed would come out red, as if somebody were burning through savings. An exact, false number is worse than no number.
TrendHow spending is going against previous periods. Same wait.
Picking a custom date rangeUnder the hood it has worked from the start; what is missing is the screen to pick the two dates. It does not appear as an option until then: an option that cannot do anything is worse than no option.

Later on

Where the app is heading once the above is closed.

Spotting what repeatsSubscriptions and recurring bills deduced from your own history, without typing them in: same merchant, similar amount, monthly or yearly cadence. It cannot be done yet, and not for lack of wanting to: deducing a cadence needs roughly three months of history, and a freshly installed app does not have them. Writing it today would mean calibrating it against invented cases.
Alerts, without leaving the phone“You are at 90 % of your budget”, a subscription that went up in price. The usual way to do this is push notifications from a server; here there is no server and no network permission, so Android itself fires them locally. It changes the character of the app —from watching to interrupting— so it will be decided slowly.
A findings panelThe biggest expense of the month, the most frequent merchant, the change against last month. It is the honest substitute for an AI assistant: “how much did I spend on going out?” is not intelligence, it is a query.

What we are not going to do

A plan that only says what is coming cannot be judged. This is what has been ruled out on purpose, and why: nearly all of it is what expense apps do, and each one would cost exactly what makes this one different.

Linking to your bankNo credentials and no Open Banking. It is the line that defines the app: without it Xtracto would be one of the hundred that already exist, and with it the app cannot read your real balance or move a euro. The three ways of recording expenses, and what each one sees, are compared here.
A cloud AI assistantSending your spending to somebody’s API so it can summarise it contradicts everything else. A model running inside the phone itself is an open door, but only the day it can be shown that not even downloading the model breaks the “nothing leaves the phone”.
Reading receipts with the cameraIt would cover cash, and technically it can be done offline. But it costs the camera permission, and this app’s whole argument is that it asks for the minimum and not even network access.
Balances, debts, goals and net worthIt contradicts the premise: Xtracto watches, it does not put you to work doing bookkeeping. Entering by hand exists to cover the gaps the bank never announces, not so that you type out your life.
Splitting expenses with friendsThat is another product, and there are already a few good ones.

Release history

Versions are numbered when they reach Google Play, not when they are finished. That is why there are things under “Done” that do not appear down here yet.

1.0 September 2026

First version, on Google Play and in closed testing. It carries just enough for the app to stand on its own: it reads card payments and direct debits from the recognised banks and turns them into transactions; it lets you enter by hand what the bank never announces, and replaces that entry if the alert turns up later; and it shows all of it across four tabs —the month total with its breakdown by card, the searchable history, the archive of raw alerts and the permission settings—. It keeps the original alert behind every transaction, exports to CSV, hides itself from screenshots and the Recents view, and comes in Spanish and English. With no internet permission, so it cannot send anything anywhere.What it does not carry yet: periods other than the month, card names, categories, the summary charts, the budget, money in and transfers, and the tablet layout. It is all up above under “Done”, waiting for the next version.

What speeds all of this up most

Joining the closed test. It is the critical path, and the page asks you for nothing: the button opens your own mail app with the message written and you send it yourself.
Asking for your bank. The name is enough. Every one that arrives is one more bank recognised the day it ships.
Sending the format of an alert that is not understood. With no amounts, dates, merchants or numbers: only the shape of the sentence.

Join the closed test Ask for my bank

This page is written by hand from the app’s plan, so it can lag a little behind what is actually inside. If anything here disagrees with the app, the app wins.

Roadmap and release notes

Where it stands today