── Demo

Compliance is the floor. The paid work sits on top of it.

One client engagement, followed end to end inside an accounting practice: work out what the return needs, get the documents in, check the file, watch what comes due, find the advisory work the checked file will support, and put it in front of a CPA to sign. Then the same record seen from the client's side.

Radar, board, and watch are not three products — they are three views of one obligation engine, which computes what each entity owes and when, and puts nothing on the board without a stated reason. Chase and check are two separate loops that feed it: one brings the documents in, the other reads the assembled package.

Every clip is a screen recording of the running product — real database, real file uploads, real model calls. Nothing is scripted or re-enacted. Where sample data is synthetic, it is labelled as such on screen.

── One client, end to end · The whole thing in one take

From chasing documents to a plan the CPA has signed

Every step on this page, run on a single engagement without switching clients: Riverside Trading LLC, a California S-corp selling on Amazon with one Hong Kong bank account. The compliance work surfaces the officer-compensation question; the advisory work is what puts a number on it. Sample client is synthetic — every name and number is fictitious and labelled on screen — and the ledger is a seeded QuickBooks fixture, but the engine, the statutes, and the arithmetic are the real product.

  • Four files arrive with meaningless names; three are sorted against the checklist and one is rejected for the wrong tax year — the reason stays on the record
  • The pre-filing check raises six findings, then a second independent pass refutes three of them — struck through, with the arithmetic that cleared them still on the page
  • FinCEN 114 and Form 8938 both land on evidence from the same statement — and neither was ever on the chase checklist
  • Sign-off on the package stays locked while high-risk findings are unsettled
  • The §199A figure opens its own trail: 20% of QBI against the wage/UBIA limit, the lesser one used, and every input's source named — preparer-entered or sample data
  • The 13-week curve starts from the balance on the connected ledger, and a tax date with no amount on file is flagged below the chart, never subtracted from it
  • Both deliverables come out as real spreadsheets — a multi-tab planning workbook and a formula-driven cash-flow file

── Step 1 · Know what you need, and get it in

Getting the documents in

Two browsers recording the same chase at the same time, placed side by side: on the left the client, on the right the practice. The client's side needs no account and no login — just a link. Nothing here is staged, and no message is actually sent: the drafted email sits in the approval queue on screen.

  • Last year's return goes in as a file — no copy-paste — and it works out what's actually missing this year
  • The chase email is drafted and waits in the approval queue — it is not sent in this recording
  • The client dumps everything at once; file names carry no meaning — IMG_4471, scan0032
  • The product sorts them against the checklist and shows what went where
  • A separate check then reads each file: last year's statement is rejected, with the reason, on both screens

── Step 1 (next season) · The chase remembers

The chase that remembers last season

Same client, one season later. When a chase closes it writes down what this client tends to get wrong — and the next season's chase opens already knowing it. Recorded against the real pipeline: live gap detection, real uploads, independent verification.

  • Opens on last season's note: took two requests, sent the wrong year's statement
  • The new chase surfaces that memory to the CPA before anything goes out
  • The recorded pattern repeats — a wrong-year document, rejected with the reason
  • Checklist hits zero, the loop closes, and a new note is written for next season

── Step 2 · Check the file, then watch what's due

Everything arrived. That is not the same as correct.

The chase says done. This is what happens next: the collected package gets read for what does not add up, and a standing watch works out what each client owes and when.

  • Reads the assembled package and finds Form 8938 missing — the Hong Kong account peaked over the threshold
  • 8938 was never on the checklist, so the chase could never have caught it
  • Every finding cites its source, then a second independent pass tries to refute it
  • One finding is ruled out on that second pass, and stays visible — checked and clear is a result too
  • Sign-off on the package stays locked while high-risk findings are unsettled
  • Deadlines are computed in code, never generated by a model — weekend and holiday roll-forward included

── Step 3 · Find the advisory work

From a client file to a signed recommendation

This is that paid work, run end to end on one real-estate client file: the engine reads what was confirmed off the return, runs the statutory tests on screen, refuses to price anything it cannot stand behind, and hands the CPA something they can put their name on. Sample client is synthetic — every name and number is fictitious and labelled as such on screen throughout — but the engine, the statutes, and the arithmetic are the real product.

  • Nothing reaches the engine until a human confirms it — extracted return figures sit in a queue with no auto-accept path
  • Missing an input means no dollar figure at all: half-answered strategies are never scored
  • The §469(c)(7) tests are answered live, and every number opens its own trail — what qualified it, what it assumes, what would flip it
  • The benefit counts this year's rental loss only — prior suspended losses are split out under §469(f), shown, and never added to the figure
  • Where a governing regime is still unverified, the row says so and the figure is withheld — no number you could sign by mistake
  • Open questions block sign-off; the CPA signs with a reason, and that figure is frozen against later re-runs
  • Approve produces a draft client memo — marked as draft, with the CPA as the author

── The client's side · Nothing moves until you release it

The client stops being a row in someone else's portal

Every incumbent's client portal is an upload box. Here the client connects their own account — and the same obligation rows can render twice, once for the professional and once in plain words. But not automatically: nothing the engine just derived reaches the client until the accountant releases it, and that release is recorded against their name.

  • The accountant describes a client in plain English; every obligation cites a sentence they wrote
  • One link; the client accepts with their own account — not a guest folder
  • At first the client sees only what's being waited on — the derived rows are held back
  • Nothing the engine derived speaks to the client until the accountant releases it, on the record
  • Then the same rows appear on the client's side, in plain words, credited to the accountant
  • Overdue items say what we know — tracked as open — never “you missed it”
  • A file the client already has is shared instead of re-uploaded — once they say which item it answers

── The owner's own view · No accountant yet

A business with no accountant reads its own books

No prior return, no accountant, no clean history — the books are the only input. Asking an owner to describe their business is circular: they'd need to know what FBAR is to know what to write. So it reads the books instead, then asks only what books can never contain. QuickBooks data in this recording is a seeded fixture; the derivation, calendar, and screens are the real product.

  • State, payroll, and contractors over the 1099 reporting threshold — each fact traceable to a ledger line
  • Then the blind spots, asked out loud: bookkeeping never records a foreign account
  • Silence is never read as a no — FBAR and Form 8938 hang on that one question
  • Answer it, and the obligations that depend on it appear with the reason attached

── Where to start · The practice's own filings

The cheapest way to try it involves no client at all

An accounting practice is a small business too. Point the radar at the practice's own filings first — 1120-S, quarterly 941s, the state annual report. No client data, so no §7216 consent, no security review, no integration.

  • The same obligation engine, pointed at the practice's own entity
  • Every obligation it picks has to cite a fact — no fact, no deadline on the board
  • The practice lands on the same board as its clients, tagged as yours

What it will not do

  • By default, every outbound email and call waits for your approval. Auto-send exists only as an opt-in mode you enable yourself, for routine chases.
  • It does not prepare, sign, or file returns. Filing decisions and signatures stay with the licensed professional.
  • It does not assert a deadline it cannot stand behind. Where the rules are in flux, it says so instead of guessing.
  • It does not hide a failed check behind a clean result. If a check could not run, it says it could not run.

Start free. Pay when it pays off.

Bring one file and see what the draft looks like. No card required to start.

Start free