Know when cash gets tight. Know which invoice to chase first.

See it work, on sample data

Output from the Operator OS command line, trimmed to the key lines, run against Tallow and Tin, a fictional candle-and-balm maker that ships with the product as one of eight example workspaces. This is sample business data, not a customer's. After install (below) you can run the same three commands against the same workspace.

$ os brief
Tallow and Tin  |  2026-10-09
--------------------------------------------------------------
  cash now                  $1,180.00
  cash in 90 days           $12,657.00   (best case $18,692.00)
  lowest point              $1,012.00 on 2026-10-15
  owed to you, late         $3,120.00 across 1 invoice

$ os sim
Cash simulation   2000 runs over 90 days
--------------------------------------------------------------
  lowest               $106.80        $1,180.00      $1,180.00
  You run out of money in 7.1% of runs.
  When it happens, it usually happens around day 37.

$ os whatfirst
What to chase first
--------------------------------------------------------------
  Right now you run out of money in 7.9% of runs.
  Each line shows what happens to that number if this one gets paid.

  invoice    customer                  amount  risk of running out   bad case gain
  INV-4102   Della Frame            $3,120.00      7.9% to   0.0%         $913.15

  Ranked by the change in odds, not by the size of the invoice.

Three views of the same data: a snapshot (brief), a distribution of outcomes (sim), and a ranked answer to "which invoice matters most" (whatfirst). sim's 7.1% and whatfirst's 7.9% come from the same Monte Carlo method run with different trial counts (2,000 vs. 800 by default).

The live demo runs the same engine in your browser (a JavaScript port checked line for line against the CLI) against this workspace or any of the other seven. Nothing to install, and nothing is sent anywhere.

Who this is for

Solo operators and small-business owners who want a 90-day cash forecast, an aging report and a ranked list of who to chase, without a monthly subscription or giving a vendor their bank login. The eight example workspaces cover field service, consulting, design, maker brands, coaching, recurring services, events and subcontracting. Each was picked for the cash leak it teaches.

Technical comfort required: you'll run commands in a terminal (Terminal on Mac, PowerShell on Windows): copy, paste, read the output. No coding. There's a read-only local report view (one HTML file you double-click, no server, no login) for glancing at numbers, but entering data, running a forecast and chasing an invoice all happen at the command line. If that sounds uncomfortable, run the workbook's first module (Pre-flight, about twenty minutes) for free before you buy. It checks that your machine can run the tool and tells you if it can't.

Installation and first use

bash scripts/install.sh          # Mac and Linux
                                  # Windows: powershell -ExecutionPolicy Bypass -File scripts\install.ps1
./os doctor                      # check this machine
./os migrate                     # set up the books and the rest of the schema
./os use 01-field-service        # load a real example business
./os brief                       # see it running
./os sim                         # the same question, with the odds
./os books post                  # derive the journal from your registries
./os books check                 # three proofs that the numbers tie

Then make it yours: ./os use 01-field-service --empty, ./os setup, ./os brief. That last command printing your own business name is the finish line for day one. Requires Python 3.9 or newer; an optional git install adds version history. Both are free and already installed on most Macs; the install script checks and tells you what's missing.

How the forecast works, and what it can't tell you

os sim and os whatfirst both run a Monte Carlo simulation: thousands of randomized draws of your next 90 days (2,000 by default for sim, 800 for whatfirst). Each draw samples whether and when every open invoice gets paid (from that customer's payment-lag history where you have one, a general assumption where you don't) and whether each open deal closes, weighted by the confidence percentage you set on it. It adds your recurring income and costs and a weekly overhead figure. The spread across all draws produces "you run out of money in 7.1% of runs".

os whatfirst tests "what if this invoice gets paid" by assuming it's collected on your normal invoice terms. A shift like 7.9% to 0.0% reflects that customer's lag leaving the simulation; it does not assume same-day cash.

The simulation shows what your data says could happen. It is not a guarantee. It only knows what you've entered, so a customer you've never invoiced, a deal confidence you haven't revisited, or overhead you forgot to log will throw it off. It never invents a number (see What it will not do below), so a forecast is only as good as the rows behind it.

Where AI comes in, and what it costs

Operator OS calls no AI model and makes no network requests. There is no API client in its source, and os brief, os sim, os whatfirst, and every other command above run on the Python standard library alone, against the files on your disk.

The org chart and "AI employee" layer (os org hire) is for planning and audit. It writes a declarative spec for what a seat may do, and os org audit reads your event log afterwards to report what share of that seat's work was yours. Hiring a seat doesn't make anything run on its own.

To have an AI agent act on your data (draft a chase email, triage your week), you connect one yourself. The product is built and tested against Claude Code, which is a separate account and cost not included in the $395 one-time license. Every tool call that agent makes is checked against data/person/policy.yml first. An unregistered agent is denied by default. Sending anything that scans as a Social Security number, card number or similar is blocked outright, whatever the policy or you approved for that call; neither you nor the agent can override that block.

What it will not do:

• Never sends anything. Every draft (invoice, chase, follow-up, content) waits for you to send it.
• Never deletes a row. Dropped, lost, cancelled, written-off and expired are statuses.
• Never contacts anyone marked do_not_contact, without exception.
• Never leaves your machine. No account, server, telemetry or sync.
• Never gives tax, legal, employment or insurance advice. It hands you organized facts and tells you to take them to a professional.
• Never invents a number. Every figure traces to a row you can open.

Buy it

Email support, with a reply within two working days. A permanent, per-business license with no subscription. The workbook's first module is free to run before you buy and tells you in about twenty minutes whether your machine and comfort level are a fit.

How it compares

Further reading