Skip to main content
Login
Apus Platform
All business agents

Ami Documents

Reads incoming invoices and documents, runs the three-way PO–receipt–invoice match and proposes the journal entry.

Keying in supplier paperwork is the most labour-intensive job in a finance team, and the one that produces the most errors. The agent reads the document, finds the matching purchase order and goods receipt itself, and stages the journal entry so the accountant only has to review and approve.

Job in hand — here is what the agent does

You handed it over once, above. From there the agent watches the data, runs when a business event happens, does the part you delegated and brings the rest back with its evidence. This is what it has done — and what is still waiting for you.

Ami DocumentsFinance & AccountingOn dutyNo office hours · no days off

Done by the agent — nobody had to ask

  1. Read 46 incoming documents just received
  2. Three-way match: PO – receipt – invoice
  3. Pre-posted entries from past treatment
  4. Set aside documents missing required fields
  5. Attached the source document to each entry

Waiting on you

  1. Three invoices priced above their POfor you to review
  2. Entry for a spend with no purchase orderfor you to approve
Agent handled 5 · you decide 2Open within your permissions
Ami DocumentsON REQUEST

Does the batch of invoices that just came in match the purchase orders?

Most of them match; three invoices are priced above their PO and one spend has no matching order.

Post the batch that matched, and leave the discrepancies to me.

Posted the entries for the three-way matched batch, each with its source document attached.

The spend with no purchase order is outside what you delegated — the entry is drafted, waiting for your approval.

How do I handle the three price discrepancies?

I have raised a request to add the missing purchase orders, with the source documents attached, and routed it to procurement. Nothing gets posted before they confirm.

Audit log · Ami Documents · delegated by General Accountant

The figures in this example are illustrative. In production the agent creates no numbers — it reads and reorders what your system already holds.

  1. Reads inside your permissions
  2. Acts only where you delegated
  3. Every action leaves a trace

Where you meet the agent

  • Right inside the module screenSuggestions appear beside the document you already have open, not in a separate app.
  • Ask Ami in plain wordsType or speak, in the same language you already use across the system.
  • Runs by itself when there is workFires on a business event and brings the work to you, instead of waiting for you to remember.

Where the agent gets its data

The agent has no data store of its own, syncs nothing outward and keeps no copy. It reads directly on the permissioned data layer the module already runs on — within the exact rights of whoever called it.

The usual way
AI sitting outside the system
A copy has to be built
  • Data leaves the company
  • Permissions get rebuilt a second time — and can drift
  • Figures lag behind the sync cycle
The Apus wayNothing sits in between — no copy to protect, to synchronise, or to re-permission.
Ami DocumentsStanding on the data layer itself
Permissioned data layer · Finance & Accounting
  • Incoming invoices and supplier documents
  • Purchase orders and goods-receipt notes
  • Chart of accounts and posting history
Because it stands inside, the agent sees exactly what the calling account may see — not one row more.

And where does the AI model run?

Tasks that touch sensitive dataRun on Internal AI · Local LLM, inside your own infrastructure
Tasks that touch no sensitive dataUse External AI · External LLM, where reasoning power is highest

The diagram above is about where the DATA sits; this is about where the MODEL runs — two different things. Routing follows data sensitivity automatically; you do not pick per task.

See the hybrid AI model

What this agent does not do

  • It grants itself nothing.The agent pre-fills forms and can write to the system — but only within what you delegated; beyond that it stops at a suggestion.
  • There is no data shortcut.The agent sees exactly what the calling account may see — no private connection of its own.
  • It never acts anonymously.The log records which agent acted, under whose delegation and who approved — accountability stays with people.

How to switch this agent on

  1. Your company needs to be running the Finance & Accounting module — the agent reads that module's own data.
  2. No separate permission setup: the agent's rights are the rights of whoever calls it.
  3. Switched on and tuned during rollout, following your existing approval flow.

The exact scope and the order modules are switched on get settled in the demo, against your current setup.

Questions people usually ask

Can the agent change data on its own initiative?+

Not on its own initiative — but it does more than read. The agent pre-fills forms and writes to the system within exactly the scope you delegated to it; anything beyond that stops at a suggestion and waits for someone with authority. Either route, every action lands in the audit log.

What if the agent suggests the wrong thing?+

Every suggestion carries its evidence: which document, which line, and why. The reviewer edits or rejects it before anything is written — and both of those actions leave an audit trail too.

How is this different from a chatbot?+

A chatbot only answers when asked. An agent also takes a job you hand over once, then watches the data on its own, runs when a business event fires, and reports back — that is the half a chatbot never does.

How is this different from an automation flow?+

An automation flow runs the steps it was configured with. An agent reads business data, ranks it by situation and picks what needs doing — within exactly the scope you delegated, and anything needing approval still goes through your existing approver.

Does data leave the company?+

Tasks that touch sensitive data run on Internal AI placed inside your own infrastructure; only non-sensitive tasks use an external model. Routing follows the sensitivity of the data automatically.

See the agents run on your own data.

Book a demo scoped to the modules and industry you actually need.

noindex