You open a cloud chatbot and type: credit card debt of twelve thousand at twenty-nine percent, take-home pay of so much, mortgage of so much — what should I tackle first? A few seconds later, a sensible plan arrives. Fast, convenient, free. And now, somewhere on the servers of a company you've never met, there's an impression of your financial life — down to the exact number, tied to your account, your IP, quite possibly your name. Not an abstract "interest in money topics." Specifically: how much you earn, who you owe, how much, and how you were panicking about it at two in the morning.
This isn't the same as asking a cloud AI to help draft an email. An email is just text. Your finances are you: income, debts, spending habits — exactly the raw material an insurer, a lender, or a plain data broker would happily turn into a risk profile if it ever reached them, through a breach, a resale, a subpoena. Asking a cloud model for financial advice means handing the most specific, most identifiable part of yourself to an entity with no obligation to guard it better than the minimum the law and the fine print require.
The good news: the same kind of model that gives you sound advice in the cloud can do it offline — on your own machine, with access to your real documents, and not a single byte leaving the disk. Here's how to set it up in one evening, step by step.
Why finance is a special case
An ordinary text leak is unpleasant. A financial one is irreversible in a different way. A password can be changed. Last year's income, your debt size, your portfolio structure cannot be "changed" — they're facts about you, fixed in time. Gathered over months of chatting with a bot, they add up to a profile sharper than most banks hold: not just a balance, but how you make decisions under pressure, what you're afraid of, where you're willing to take a risk.
That profile doesn't belong to anyone malicious — yet. But data outlives the goodwill of whoever holds it. Privacy policies get rewritten. Companies get acquired. Databases get breached. And what lands on someone else's server once cannot really be pulled back — you can only file a deletion request and hope it was honored.
A local model removes this risk not through policy but through construction: the question "where does my data go" has no answer of "to the company's server," because the company's server was never in the chain to begin with.
What you need
Less than you'd think, and almost certainly less than what you already own. A laptop or desktop from the last few years with eight gigabytes of RAM runs a compact model; sixteen or more runs one that reasons carefully step by step, which matters for financial calculations specifically. A dedicated graphics card speeds things up but isn't required — modern chips, including recent Apple ones, handle this fine on their own.
The tool for running it is a "runner" — a free application that installs like any normal program and gives you a chat window without a single terminal command. Inside it you pick a model from the open-weight families — Llama, Mistral, Qwen, and similar ones exist in sizes from very small to genuinely capable — and start the download. A model weighing a few gigabytes downloads once and stays on your disk for good.
For financial work, pick a mid-sized model rather than the smallest one: calculations like "how much interest will I overpay if I pay off card A before card B" need careful, step-by-step reasoning, not just fluent text. Small models sometimes fumble signs and order of operations; a larger one, within what your memory can handle, does noticeably better. If in doubt, check any answer against a calculator once — it takes a minute and it's worth it.
Load your own documents, not your secrets into someone else's profile
A bare chat is already private, but it becomes genuinely useful for your finances once it can see your real numbers — not because you retype them every time, but because many current runners can "read" a connected folder of files and answer based on it, without ever sending them past your own machine's memory.
Put together such a folder: account statements as PDF or CSV, an export from a budgeting app, a debt payoff plan, a list of subscriptions. Connect it through a "local documents" or "knowledge base" setting — different runners name this feature slightly differently, but the principle is the same: files are read from disk, indexed on disk, and answers are built on disk. Now you can ask specifically: "Based on the last three months of statements, where am I spending more than I planned?" — and get an answer built on your actual numbers, not on abstract advice to "keep a budget."
That's exactly what no cloud chat can offer without risk: either you paste real numbers into the company's input box, or you settle for advice given blind, with no numbers at all.
Prove to yourself that it's actually private
Don't take it on faith — check it physically. Turn off Wi-Fi, or pull the cable. Ask the model something drawing on your document folder: "At my current pace, how long until this debt is paid off?" It answers. The network is off, and the answer came anyway — which means it had nowhere to come from except your own disk. That moment is worth seeing with your own eyes at least once: not a privacy claim you're trusting, but a fact you just verified with your hands.
> Our record. The Ib — the heart weighed on the scales of Maat — belongs to no one but you; in the old judgment it wasn't put on public display before the weighing, and even then what mattered was the balance, not a transcript of every private worry. A financial conversation with an AI is the modern equivalent: talking to your own heart about fear, debt, the wish to get out from under it. Handing that to the cloud means putting the Ib on display before its time, for an audience that never sat at the scales and has no standing to judge. A local model keeps that conversation exactly where it belongs — inside your own house, not in someone else's archive.
Where this fits your bigger strategy
A local model for finance isn't a one-time trick; it's a permanent shift in habit. Every time you catch yourself about to type an exact debt figure, a precise salary, or a list of assets into a cloud chat, that's the signal: the same thought can be worked through locally, with no leak and no compromise in the quality of the advice. Over time the cloud becomes the tool for abstract questions with no numbers attached, while anything touching your real money stays behind a wall you built yourself and fully control.
This isn't paranoia or a gesture against technology — it's the same logic behind refusing to keep savings somewhere access to them is someone else's decision. The tool you think about your money with should be exactly as much yours as the money itself.
Do this today
You don't need to gather your entire archive of financial documents and set up indexing in one sitting. You need to take the first real step right now: install a runner. Pick one of the well-known free local-AI tools for your operating system and start the install — just the application, no model yet, a matter of minutes.
By evening it'll be sitting there ready. Then you download a mid-sized model, connect a folder with one or two statements as a test — and that same evening, ask it your first real question about your own money, knowing the answer stayed exactly where the question did: at home.