AI Close Agent

A reviewer that reads your instructions, checks the books and files the follow-ups.

The Close Agent runs skills against each client’s own written instructions and their QuickBooks Online data. What isn’t done becomes a task. It only reads QuickBooks, and it runs on your firm’s own AI key.

How a run works

Instructions + QuickBooks → findings → tasks.

A run is one skill, for one client, for one period. It’s triggered on a schedule (for example, the day after your close deadline) or by someone on your team clicking Run.

1. Client instructions + QuickBooks

The agent reads the skill, the client’s notes (“what closed means” for them) and the period’s tasks, then queries their QuickBooks data and reports.

client notestasksQBO queriesreports

2. Findings

It writes a short report: closed or not closed, plus each finding as a blocker, warning or info item, with the evidence and a recommended fix.

blockerwarninginfo

3. Follow-up tasks

Findings that need action become tasks on the client’s project, assigned to the right person, so nothing lives only in a report.

assigneddue dateclient project
A Month-end close review run for September: not closed, one blocker about an Amex reconciliation, two warnings and one info finding, with follow-up tasks created
A real run report from the sample firm. You can expand “What the agent checked” to see every query and report it used.
Skills library

Start with proven checks. Make them yours.

A skill is a plain-language set of instructions. Use the ones in the library, tweak them, or write your own. Skills are versioned, so you can see what changed and when.

Month-end close review

Checks every item in the client’s “what closed means” list against QuickBooks: reconciliations, uncategorized balances and tasks marked done.

Bank reconciliation check

Verifies every bank and credit card account is reconciled through month-end.

AP / AR aging review

Flags stale bills and receivables, possible duplicate bills and unapplied payments.

Payroll tie-out

Matches payroll journal entries to the pay dates in the client’s notes.

Sales tax liability check

Compares sales tax payable at quarter end to the amount filed.

Your own skills

Write a new skill in plain language for any check your firm repeats every month.

Bring your own AI key

You pay your AI provider directly. We add nothing on top.

An admin adds your firm’s API key in AutoZonBooks. Keys are encrypted at rest and only the last four characters are ever shown again.

  • Claude runs skills today. AutoZonBooks’s skills are written and tested with Claude.
  • ChatGPT, Grok and Gemini keys can be stored for when those providers are supported for skills.
  • Every run shows its tokens and estimated cost, so there are no surprises.

What a run roughly costs

Rough cost per Close Agent run
Kind of runRough cost
Small check (one account, short notes)~$0.10–0.50
Month-end close review, typical client~$0.50–1.50
Busy client with lots of activitya few dollars

Rough, illustrative ranges, not a quote. Cost depends on the model you choose, your provider’s current prices, and how much data and how many notes a client has. Each run in AutoZonBooks shows its own token count and estimated cost, and your provider’s bill is the source of truth.

Read-only QuickBooks

It never changes QuickBooks.

Each client’s QuickBooks Online connection is used read-only. The agent can run queries and pull reports, and that’s all. It can’t create, edit, void or delete anything in the books.

  • Findings and tasks stay in AutoZonBooks. A person on your team decides what to fix.
  • Every step is logged. See exactly which queries and reports a run used.
  • Tokens are encrypted. QuickBooks tokens are stored with AES-256-GCM encryption.
Read about security

What the agent can do

Read client notes and tasks · run QuickBooks queries · pull reports like the balance sheet and P&L · write findings · create follow-up tasks in AutoZonBooks.

What it can’t do

Create, edit or delete transactions, bills, journal entries or anything else in QuickBooks.

FAQ

About the Close Agent

Does the agent close the books for us?

No. It checks whether the close is done the way the client’s instructions describe, and tells you what isn’t. Your team still does the work and makes the decisions.

Where do the client instructions come from?

Your team writes them on each client’s “Instructions & notes” tab: which accounts to reconcile, payroll dates, anything specific to that client. The better the notes, the better the checks.

How often does it run?

As often as you set it: monthly after your close deadline, quarterly for sales tax, or on demand from the client’s Close agent tab.

Is client data used to train AI models?

Data goes to the AI provider whose key you use, under your firm’s agreement with that provider. Check your provider’s API data policy; at the time of writing, Anthropic, for example, doesn’t train on API data by default.

Watch the Close Agent review a close.

We’ll run it live on a sample client and walk through the findings.