Skip to content
SageFin

What SageFin does with what it finds

Once an account is connected, a lot happens without being asked. This page is what, and — the part a finance app owes you an answer to — where your data goes while it happens.

It sorts your spending

Every transaction gets a category as it arrives, from one of four things: a rule you wrote, what SageFin has learned about that merchant, a model reading the merchant and amount, or you.

They run in that order and the first one to answer wins, which is why a rule always beats a guess. Your own choices are never overwritten by a later pass — the full explanation is here.

It notices what repeats

Subscriptions, bills and anything else arriving on a rhythm get picked out and offered for review. Confirmed ones become a forecast: what is due, and when.

They are offered rather than assumed, because a monthly transfer to a savings account and a monthly gym membership look identical to a pattern-matcher and are not the same thing to you.

A bill or a merchant that arrives as statement text, such as "SQ *ROBLOX 8823", may be offered a plainer name, here Roblox. Only merchant names go to the AI for this, and nothing is renamed until you accept: merchant names SageFin suggests.

It writes last month up, if you ask

The Monthly summary card on the Dashboard lays out the month that just ended in sections:

The figures are SageFin's, never the AI's. They come from the same reports and pages you can open yourself. Press Write a takeaway and the AI is given those figures, already written out, and asked for one or two sentences on what stands out, shown above the sections. What it writes is checked before you see it: every dollar amount, percentage, count and date must be one of the figures it was given, with the same sign, and it may not advise you. If it gets that wrong twice, you see the sections without a takeaway. With AI off, or not yet allowed, you always see the sections alone, and nothing is sent. A takeaway written on the 1st is checked again against the figures each time the card opens, and left off if they have changed since.

Turn it on to get it on the 1st. Under Settings → Notifications, the Monthly summary row is off until somebody turns it on. Then the summary is written once a month, shortly after midnight UTC on the 1st, and arrives in the bell, your email digest or as a push, as you set it. A private household's is written when it is next unlocked and kept in the household's file, and its email says only that it is ready.

It describes; it does not recommend. Like everything else here, it will not tell you what to cut, save or invest.

It keeps a history you can trust

Balances are recorded over time, not just today, so a chart of last year is a record rather than a reconstruction. SageFin also compares the balance your bank reports against the transactions it has — if the two disagree it says so, which is unusual and is the point.

It reads receipts, if you send them

Forward one and the merchant, date, total and line items are read off it and matched to the transaction — emailing receipts in.

Where your data goes

Your bank credentials: nowhere near us. They are typed into your data provider's own window. We receive permission to read an account, never the means to sign in as you.

Your transactions: to our servers, in a file that is your household's alone, and to a model when there is a reason. Categorizing a transaction and answering a question in chat both send context to an AI provider — typically the merchant name, amount, date, your category list, and for chat your question and the figures needed to answer it. Not a copy of your history, and not on a schedule: only when there is something to decide. Asked how one of your settings is set, chat reads that too — the switches, alert minimums and household name the settings pages show, never an email address.

Receipt images: to the vision model that reads them, and nowhere else. A receipt you email in passes through SageFin's receipts mailbox at Google on the way, and is removed from it once it is filed — at the latest 30 days later.

Which AI provider is your choice. By default it is one we choose and pay for. An owner can have the household's AI run on a provider and key of its own instead, and then the same things go there, under that company's terms.

Sage tells you which model is answering. Under the box where you type, beside the note that it can make mistakes, it names the model and whose it is: the one SageFin provides, or the one on your own key. That line is read from what is actually in use, so it stays right when the model changes.

One thing is shared, and it is worth naming precisely. When SageFin finds a merchant's website, which is where its logo comes from, it remembers that merchant-to-website match and reuses it for everyone. What is shared is "this merchant's site is example.com" — not your amounts, not your dates, not anything that says you shopped there.

What SageFin learns about which category a merchant belongs in is not shared. Each household has its own memory of that, taught by its own transactions and its own corrections.

You are asked first. Before anything of your household reaches a model, SageFin asks an owner, once, on whichever app they open: on the iPhone and Android app before the first sync, and on the web in a strip above every page. The model is run for us by Amazon Web Services (AWS), and under AWS's terms what we send isn't used to train it or shared with the company that made it. If AWS is down, a question you ask Sage can go to xAI instead, and what you type to Sage is checked for abuse by OpenAI. Until somebody answers, SageFin makes no AI call on your household's data at all. Members who are not owners are not asked; they see that AI is off until an owner answers.

You can change your answer. Two switches in Settings → Preferences, which an owner sets for the whole household:

The same page deletes your own chat history with Sage. That clears your conversations only, not anyone else's in the household, and it cannot be undone.

What Sage can do is your own setting on the same page, and it affects only your conversations:

Proposals need no other switch: Allow apps to make changes in Settings → Integrations is for AI apps you connect yourself, and Sage's proposals do not depend on it.

Between households: nothing. Every query is scoped to your household, and a household sees only its own money. Who can see what covers sharing inside one.

The privacy policy is the complete version, including how long things are kept and what survives deleting your account.

What it does not do

It does not change your transactions by itself. Sage can propose categories, tags, hiding, businesses and rules, and you approve them; it cannot approve anything, and saying "yes" in the chat approves nothing either.

It does not move money. SageFin reads; it cannot pay a bill, transfer between accounts or cancel a subscription. A connection that could do those things is a different kind of permission, and we do not ask for it.

It does not give financial advice. It will tell you what you spent and what is coming; it will not tell you what to buy or where to invest.

It does not decide anything is final. Every category, split, budget and recurring stream is yours to change, and changing one is not a correction of an error — it is how the thing is meant to work.