The AI CFO That Lives in Your Office

From a Search Box to a Finance Desk

Faraday Local started as private search: point it at a folder of invoices, receipts, bank statements and spreadsheets, and ask questions in plain English. A model running on the Mac plans each question, the records database does the arithmetic, and every answer links back to the rows and pages it came from.

Now it does more than answer. It watches the books. Every file that lands in a watched folder is read, its payments are extracted and categorised, and the whole ledger is checked again. The result is one page that shows how the year is going, where it is heading and what deserves attention today.

For a growing business this is the part of a finance team that is hardest to hire early: someone who reconciles, re-checks and re-forecasts every time something changes. Faraday Local does it continuously, on hardware the business owns, with no per-question bill to make anyone ration it.

A Finance Desk That Never Clocks Out

Faraday Local works on its own between the moments when people need it. It reads, calculates, flags and drafts, and brings the decisions to the right person with the evidence attached. Every number comes from the records database, and every finding links to the file and row behind it.

That split is what lets an owner leave it running. It does the watching and the arithmetic around the clock, and people spend their time on the calls that need judgement.

Continuous Checks: Catch It the Day It Happens

A year-end review finds a double payment months after the money has gone. Faraday Local looks again every time a file arrives. When a card statement lands with the same Staples charge on August 3 and again on August 7, the summary page shows it within seconds, with both rows and the statement they came from.

Every check learns from the team. Mark a finding as fine and it is not raised again. Correct a payment’s category and the correction is applied to future payments that read the same way. Exclude a duplicate row and every total, check and forecast is recalculated without it. The more the team uses it, the sharper it gets, without anyone retraining a model.

Forecasts That Get Sharper Every Month

Most small-business forecasts are a spreadsheet someone updated once. Faraday Local re-forecasts whenever the data changes and shows exactly how good the forecast is.

The method is clear enough to explain to a board. Recurring charges, such as rent, memberships and subscriptions, are found from three payments and projected on their own schedule at their usual amount. Everything else is projected at its average over the last 12 complete months, adjusted for the time of year once there is enough history. The likely range covers eight in ten outcomes, based on how much that spending usually moves from month to month.

Then it grades itself. Each month it forecasts the last three complete months using only the data available before each one, compares the result with what was actually spent and reports the average error. As months of history build up, duplicates are excluded and categories are corrected, the forecast has more and cleaner data to work from, and the accuracy figure shows the progress on the same page.

For a finance lead deciding how much to trust a number, a forecast that publishes its own track record is worth more than a confident one that doesn’t.

Instant Answers for the Whole Team

The numbers are ready the moment anyone looks. Insights, Checks and the forecast are recalculated in the background whenever the data changes and stored, so the page opens with its numbers already there, even straight after a restart. In Faraday’s own tests on a library of 50,000 records and 3,000 files, the summary opened in 5–8 milliseconds from the stored result, where recalculating took about 2.6 seconds. After an edit, the page shows the previous numbers marked as updating and swaps in the new ones about 2.6 seconds later.

And anyone can ask. The same assistant that adds up spending answers questions about the documents around it: policies, contracts, membership agreements. Answers cite the passage they came from, so an employee checking an expense limit gets the rule and its source in one reply.

Spaces: The Right Data for Each Person

Bringing the whole team in works best when each person sees exactly what they need. Account directors read contracts but not payroll. Everyone reads the expense policy. The bookkeeper sees the bank statements. Faraday Local handles this with spaces.

Each person signs in from their own computer and sees totals, checks and answers only from their spaces, plus their own chats. Ask about spending without access to Financial data and the reply says so and names who can share it.

The boundary sits in the database, before anything reaches the model: search, records and checks only return what is in the asking person’s spaces, so the model never sees the rest. That makes it simple to open Faraday Local to a growing team with confidence.

Built for Growing Businesses

A finance assistant earns its place by running constantly, and that is where owning the hardware pays off. Re-checking the whole ledger whenever a file arrives, and re-forecasting whenever a record changes, costs nothing extra on a machine you already own. On a metered cloud service every one of those passes is a line on a bill. We set out that argument in more detail in Dollars per Task.

It also keeps the most sensitive files a company has, its ledgers, payroll, contracts and policies, inside the building, while the whole team benefits from them. As the business grows, the folders, the history and the team grow with it, and every new month of records makes the forecast better informed.

Faraday Machines installs open-weight models on Mac Studio hardware in Toronto. Faraday Local is the finance and records assistant that runs on it. If you want to see it on your own books, talk to us.

References