Local-first finance tracker keeps your data offline and prices every AI query upfront

Aurelio is a local-first personal finance application that keeps all your data in a SQLite file on your own machine and routes AI queries through OpenRouter using your own API key. Nothing leaves your computer unless you explicitly ask the AI something.

The architecture is deliberately split. The wealth-tracking side — banks, brokers, portfolio positions, fund look-through by country and sector, real assets, debts, cash flow — works entirely offline. The AI layer is optional and additive. When you do engage it, the chat interface can search the web and, crucially, proposes every data change as a card you must explicitly confirm or reject. That confirmation step is a meaningful design choice: the AI cannot silently mutate your records.

The "analysis" feature is the most architecturally interesting piece. Two synthetic personas — described as an analyst and a confidant — argue over your portfolio, and a synthesis surfaces what deserves attention first. Whether that produces genuinely useful tension or just theatrical disagreement depends entirely on the underlying model and how the prompts are constructed; the README points to a separate document explaining the reasoning approach but doesn't reproduce it.

The default model is Claude Opus 5.5, and the README gives unusually honest per-operation cost figures. A first question costs around eleven cents; a follow-up drops to under two cents; a full analysis runs between thirty-three and fifty-one cents. Those numbers are dated to October 2026, which is a future date relative to most readers — either the README was written speculatively or the project is tracking a model not yet publicly released. Either way, the cost transparency is more candid than most AI-adjacent tools bother with.

Setup requires Node.js 22 or later, Python via uv, and a single shell script. A demo mode runs on invented household data so you can evaluate the interface before touching real records. The database gets a backup copy before any migration that changes its format — a small but important detail for a tool holding financial history.

The local-first constraint is the real value proposition here. Financial aggregation tools that phone home are common; one that keeps everything in a file you own and control, with AI as an opt-in layer you pay for directly, is a different trade-off. Whether the AI reasoning is sophisticated enough to justify the per-analysis cost is something only the linked reasoning document and actual use can answer.