Every client chased, every month, without hiring for it.
The work in a practice is not hard, it is repetitive and it multiplies by client. An agent runs the same month-end across every ledger you keep, and stops at anything that needs your judgement.
4-day trial · Cancel anytime
An AI agent for accounting firms runs the repetitive part of practice work across every client at once: chasing missing receipts and bank statements, preparing reconciliations in Xero or QuickBooks, flagging transactions it cannot categorise, and tracking each client's filing deadlines. It drafts the chasers and the summaries; a human approves anything that leaves the firm.
At a glance
| Best for | Accounting practices & bookkeepers |
|---|---|
| Key integrations | Xero, QuickBooks, Dext, Stripe |
| Connects to | 3,000+ apps via managed OAuth |
| Setup | Connect your tools — no code required |
| Autonomy | Suggest-only to fully autonomous, with approval gates |
| Pricing | 4-day trial (card required, no charge today), then paid plans |
By Loïc Jané · Updated August 10, 2026
How the agent works
From trigger to action — Fleece AI does the work in between.
Gmail
Client documents
Google Drive
Statements
Stripe
Payments
Fleece AI
Fleece AI at work
Xero
Reconciliation staged
Slack
Deadline digest
Gmail
Client documents
Google Drive
Statements
Stripe
Payments
Fleece AI
Fleece AI at work
Xero
Reconciliation staged
Slack
Deadline digest
The work scales with the client list, and only with the client list
A practice does not get harder as it grows — it gets longer. The same fortnight of chasing, categorising and reminding repeats once per client, and the only lever most firms have is another pair of hands.
How the agent runs a practice month
It knows what each client still owes you
The agent reads the ledger and the document inbox for each client, works out what is missing against what a complete month looks like, and builds the list per client rather than one long list for the firm.
It chases, in your words, on your schedule
Reminders are drafted per client with the specific documents named, scheduled at the cadence you set, and escalated to a partner only when a client stops responding. Nothing sends without approval unless you decide otherwise.
It prepares the reconciliation
Transactions are matched against invoices and receipts in Xero or QuickBooks, the confident matches are staged, and anything ambiguous is set aside with the reason — so the review starts at the exceptions instead of the whole ledger.
It keeps the deadline board honest
Each client's filing dates live in the agent's own workspace and are checked on a schedule. You get one digest of what is due, what is at risk and what is blocked on a client, in the channel your team already reads.
Why a practice runs this differently
One agent, every client
The same month-end runs across every ledger you keep. Adding a client adds a connection, not a fortnight of work.
Approval gates on anything that leaves the firm
You are acting on someone else's books. Every client-facing message and every posting can be held at a gate until a person clicks.
A trace for every action
Each run is recorded step by step, so the answer to "why was this categorised that way" is retrievable months later.
Runs on your calendar, not your attention
Chasers on the 3rd, reconciliation prep on the 10th, deadline digest every Monday — set once, in your own timezone.
Per-client separation
Each client's work runs against its own connected accounts, so one client's data never becomes the context for another's.
Judgement stays with you
The agent stages, drafts and flags. It does not decide the treatment of an ambiguous transaction — it hands it to you with what it found.
Connects to the practice stack
Xero and QuickBooks for the ledgers, Dext for document capture, Stripe for payment data, Gmail and Drive for what clients actually send you, Slack for the digest. All through managed OAuth — no API keys.
What changes in the practice
Questions from practices
Only if you choose that. Approval gates are per action: you decide which steps need a human yes, and posting to a client ledger or emailing a client are exactly the actions most practices gate. Until approved, the work sits staged.
Each client's work runs against its own connected accounts and its own agent context. Data is not pooled across clients, and the run trace shows exactly which accounts each step touched.
No — it works them. Dext still captures documents and Xero still holds the ledger; the agent is what reads across them, notices what is missing, chases it and prepares the reconciliation for your review.
It is set aside with the reason rather than guessed. The point of the agent is to shrink the pile you review, not to make decisions you would want to make yourself.
No. You describe the month-end you already run in plain language and connect the accounts over OAuth. There are no flowcharts to draw and no API keys to manage.
That is the sensible way in. Connect a single client's stack, run one month-end alongside your normal process, and compare. Every plan opens with a 4-day trial.
Go deeper on the blog
Run the same month-end across every client
Connect one client's stack, watch a month-end run end to end, and keep the approval gates where your judgement belongs.
Powered by Fleece AI · autonomous agents for 3,000+ apps