A free Google Colab notebook that runs a local AI on your bank and credit-card statements. No logins. No subscriptions. Nothing leaves the runtime.
Educational only — not financial advice. The tool's job is to surface patterns in your own statements so you can make better decisions; it's not a substitute for an accountant or a planner.
Almost every personal-finance app asks for your bank login, quietly sells your aggregated data behind the scenes, and charges you a subscription on top of it. This one inverts that model entirely.
Here's the math most of these apps don't want you running: a $10/month "money-saving" app costs $3,600 out of pocket over 30 years just in fees. But the real number is opportunity cost — that same $10 a month invested at an 8% average annual return grows to roughly $14,900. A "small" subscription doesn't cost $10. It quietly costs you fifteen thousand dollars of future net worth. The thing pitched as helping you save is almost always the thing taking the most.
Here's how this works instead: copy the notebook to your own Google Drive, upload up to twelve months of bank and credit-card statements, and a local AI does the categorizing, the velocity math, the subscription audit, and the what-if scenarios — all inside a sandbox only you can see. Nothing is uploaded, nothing is stored, nothing leaves the runtime. You stay in control of the file. The AI runs next to it. Then both go away.
And it's completely free, forever — Colab is free, the model is open-source, there's no account to create, no subscription to cancel, no card to put on file. A tool meant to help you save money shouldn't be the thing draining it.
Opens in Google Colab. Click File → Save a copy in Drive to get your own version. Then go to Runtime → Change runtime type → T4 GPU → Save for faster analysis. Hit the big ▶ Run button — first boot takes about three minutes while the model warms up, then every analysis pass is seconds.
If you've never heard of either of these, here's the 60-second version. You don't need to install anything or understand the code — just know what's happening behind the play button.
Colab is a free Google product that gives you a temporary computer inside your web browser. It looks like a document made of text and code blocks. When you click ▶ on a block, Google's server (not your computer) runs that code and shows you the output right there in the page. It's the same idea as Google Docs, except instead of editing words it's running programs. Free accounts even get access to a GPU — fast hardware normally rented for dollars an hour — for short sessions.
What that means for you: nothing installs on your laptop, nothing slows it down, and when you close the tab the temporary computer is destroyed.
Ollama is a small piece of software that lets you run open-source AI models on your own machine instead of sending prompts to ChatGPT or Claude's servers. Think of it as a personal AI engine — you tell it which model to load (this notebook uses Qwen 3, a strong open model from Alibaba's research lab), and from then on every prompt you ask is answered locally.
What that means for you: the AI reading your statements is running inside the same temporary Colab computer as your data. Your transactions never travel to OpenAI, Anthropic, Google, or any third party — the model and the data live and die together in one sandbox.
Open the link below. In Colab, click File → Save a copy in Drive. Now it's your notebook — edits only affect your version.
Runtime → Change runtime type → T4 GPU → Save. CPU works too, but each analysis pass takes minutes instead of seconds.
Hit ▶ on the big code cell. It installs everything and warms up the model. First run takes about three minutes — go grab a coffee. After that, it's instant.
CSV, OFX/QFX, XLSX, or PDF — one for every month of the year if you want a full annual view. Direct downloads from your bank's site work best. The more months you load, the better the "past best you" comparison and the easier it is to spot annual subscriptions you forgot about.
Drill Sergeant (no mercy, converts dollars into hours-of-your-life), Supportive Friend (warm but honest), or CFO Mode (clinical, framework-driven). Optional: enter your hourly rate so the Sergeant can do the math.
You get a personalized brief covering spending velocity, your past best month, silent subscription leaks, and category breakdowns. Then chat with the buddy, run what-if scenarios on cutbacks, and download a report if you want a record.
The Colab version is the zero-install path — nothing to download, nothing to maintain, and Google's GPU is free. But if you'd rather have the same kind of tool running entirely on your laptop (no Google involved at all), here's the rough recipe:
ollama pull qwen3:8b (or llama3.1:8b, or any model your hardware can handle). Ollama downloads the model file once; after that it lives on your disk.
ollama run qwen3:8b, paste rows from your CSV into the chat, and ask the model to categorize them, flag subscriptions, or compare months. For a polished UI, install a desktop client like Open WebUI, AnythingLLM, or LM Studio and point it at your local Ollama.
The desktop route is more private (data never even touches Google's servers) and works offline, but it needs ~10 GB of disk space, a reasonably modern machine, and a few minutes of setup. The Colab notebook is the same idea with the install step removed.
One click to open it, one more to save a copy, one to hit play. About three minutes of warm-up while the model loads, and then your statements get the analysis they've been waiting for.
Open the Notebook in Colab →