Pulse Local runs its budgeting AI entirely on your Mac, no cloud or account required

A new budgeting app called Pulse Local appeared on Hacker News on October 14, and its pitch targets a specific gap. According to the Show HN post, most apps that call themselves local keep your ledger on disk but still ship the question, along with the rows needed to answer it, off to a hosted model. Pulse Local instead runs its AI models on the Mac itself, so the data and the queries about it never leave the machine.

How the on-device AI works

The developers are candid about the trade-offs. They say the bundled models are small and weaker than frontier systems on open-ended questions, so conventional code does most of the work. Supported spending answers are computed directly from transactions, then checked in a second reading before a figure is displayed. When the two readings disagree, the app reports that it is unsure rather than guessing, and AI answers are labelled as such.

Speed is another concession: with verification switched on, a money answer takes 12 to 14 seconds. To back up accuracy claims, the team says the money maths was tested against six months of a real ledger, month by month, and matched exactly to the cent.

Answers are also auditable. Labels above spending figures show which month and categories Pulse included, and selecting a label lets you change them. Conversations can continue with follow-ups such as asking for a breakdown by merchant or naming another month to compare. The app explains its figures and common budgeting terms, but deliberately stops short of telling you what to do about them.

A full budgeting app first

The team positions the product as a complete budgeting tool rather than a privacy showcase. It covers plans and targets, rules, rollover, splits, cash flow and forecasting, and all of it runs locally. Data arrives via CSV import from any bank in any currency, because a live feed would require a server holding your bank credentials. The one exception is bank sync, which carries a running cost on the developers' side and is sold separately as a Sync Pass for 29 dollars a year.

Pay-what-you-want pricing

The pricing model is unusual. Every feature is available free with no account and no nagging, while optional payments of 20 or 40 dollars act as support for future development, and buyers can type in any amount they choose. Payments are one-time rather than subscriptions, and the post says nothing in the app checks which option a customer picked — every button leads to the same download.

The current release is a beta for Apple Silicon Macs, signed and notarised by Apple so it installs like any standard app and keeps itself updated. The limitations are stated up front: the app answers in English, the ledger exists on exactly one machine, and older Intel Macs are unsupported. A Windows version is next, with no date given. The team describes itself as a small group of ML practitioners and product people, noting that once the app is installed it costs them nothing to run.

Why it matters

Financial data is among the most sensitive material on a personal computer, and the pattern Pulse Local calls out — a "local" app that quietly sends questions to the cloud — is widespread. Its design offers a workable template for privacy-first AI features: deterministic code handles the arithmetic, small on-device models handle the language, and the app accepts slower answers, weaker open-question performance and an explicit "I'm not sure" in exchange for a hard guarantee that nothing leaves the machine. The no-account, pay-what-you-want approach also lowers the barrier for anyone who wants to test whether on-device models are finally good enough for everyday personal finance.