Finance

Closes the month a little bit every single day.

Bookkeeping goes wrong in the gap between the transaction and the person who understands it. Doing it daily closes the gap: payments are matched while the context is fresh, invoices are chased on the day they age, and the month-end pack assembles itself instead of consuming a week. The one rule written into its Soul is that it never invents a number — anything it cannot prove is queued for a human, with the evidence attached.

What you stop doing

The last week of every month, and the quiet uncertainty of not knowing whether the books are current.

MI
Mira
AI Bookkeeper
On duty
Routines
  • Daily reconciliationEvery day, 07:00
  • Dunning runEvery weekday, 08:40
  • Cash positionEvery day, 13:30
  • Close pack updateEvery weekday, 17:00
Skills
  • reconcile-payments
  • chase-overdue-invoice
  • categorise-spend
  • check-renewal-against-contract
  • assemble-close-pack
Next RunEvery day, 07:00
The working day

What Mira does between waking the servers and going quiet.

Every hour below is one Routine you can open, read, and change — including the one where it stopped and asked a human.

One TuesdayMira · AI Bookkeeper8 Runs · 1 Decision
  1. 07:00

    Reconciles yesterday

    Matches 42 payments to invoices on amount, reference, and date. Three do not match cleanly — a partial payment, a currency difference, and a customer who paid two invoices at once — so those are queued, not forced.

    Finance
  2. 08:40

    Chases, politely and on time

    Sends the day's dunning: gentle at 7 days, firmer at 30, and the pre-escalation note at 60 — each from the account's own history, and each pausing the moment someone replies.

    Finance · Invoices
  3. 10:15

    Categorises against your chart, not a generic one

    Codes yesterday's spend to the accounts you actually use, and flags the two subscriptions nobody has logged into for sixty days with what they cost a year.

    Finance · Ledger
  4. 11:00

    Refuses to guess

    Stopped for you

    A cross-border refund could book to either entity, and the answer changes the VAT treatment. It writes a Decision with both readings and the amount at stake, and moves on.

    Decisions
  5. 13:30

    Updates the numbers people actually ask for

    Runway, burn, and expected cash-in, recomputed from the ledger rather than from a spreadsheet somebody last touched in March.

    Explore
  6. 15:00

    Reads the contracts, so the invoices are right

    Checks this week's renewals against the uploaded contracts — the escalator clause, the seat count, the currency — and raises the invoices that are due.

    Customers
  7. 17:00

    Files the close as it goes

    Adds today's reconciliations, accruals, and unresolved items to the month-end pack. On the last day of the month there is nothing left to assemble.

    Finance · Reports
  8. 19:30

    Reports the exceptions, not the total

    Posts what did not reconcile and why, what is now overdue past 60 days, and what it needs an answer on before the close.

    Workspace
By the end of the day
42
Payments reconciled
3
Exceptions queued, not guessed
1
Decision that needed you
What it brought to you
  • Which entity books this cross-border refund? It changes the VAT treatment.
  • Vertex is 68 days overdue and asking for terms. Extend, or hold?

Answering one performs no side effect. The employee writes the question and the options itself, and an ordinary Member can answer it.

What a bookkeeper can do

Capable in the ways that matter, bounded in the ways that count.

Reconciliation with an exception queue

Clean matches are posted; partials, currency differences, and one payment covering several invoices are queued with the candidates and the amounts. Guessing here is how books go wrong quietly.

Dunning that behaves like a person

A schedule that escalates in tone, drafts from that account's own history, and stops dead the moment a human replies. Nobody gets a firmer notice after they have already promised to pay.

A real double-entry ledger

Estimates, invoices, recurring billing, bills, vendors, reports, reconciliation, and period close ship natively — so the AI Employee is working your books, not a copy of them in a spreadsheet.

Payments stay behind a human

Money leaving the company over your threshold is an approval a Member ticks, and the server replays the exact action. The employee prepares it, evidences it, and waits.

Setting it up

Four documents, and then you stop being the trigger.

Everything that makes this a bookkeeper rather than any other role is plain, editable text. Change how it thinks by editing a document, the way you would rewrite a job description.

Soul

How it judges

  • Voice, priorities, and the lines Mira will not cross alone.
Skills

What it repeats

  • reconcile-payments
  • chase-overdue-invoice
  • categorise-spend
  • check-renewal-against-contract
  • assemble-close-pack
Routines

When it works

  • Daily reconciliation — Every day, 07:00
  • Dunning run — Every weekday, 08:40
  • Cash position — Every day, 13:30
  • Close pack update — Every weekday, 17:00
Grants

What it can reach

  • Stripe and bank Connections
  • Finance — invoices, bills, and the general ledger
  • Customers, for contracts and account history
  • The accounting policy notebook in Notes
Questions

The ones people actually ask.

Can it pay bills?
It prepares them. A payment over the threshold you set is an approval a human ticks, and the redacted payload is only replayed server-side after that — the employee has no standing power to move money.
What if it categorises something wrong?
Every posting is auditable and reversible, and the correction goes into its Soul or the relevant Skill so the class of mistake stops. The exception queue exists so that the ambiguous cases never get categorised in the first place.
Does it replace my accountant?
No — it keeps the books current so your accountant is reviewing a clean, closed month instead of reconstructing one. The judgement calls still arrive as Decisions for a human.

Build an autonomous company today.

Install Genosyn, choose an AI Model, write the first role, and put it on a schedule. Tomorrow morning, one job runs without you. The company grows from there.