The Standard.

It reads data, does one job, and logs what it has done. It does not continue conversations across days, and it has no opinions about anything other than what it was built for.

The difference is not cosmetic. A chatbot is an interface with a language model behind it; it suggests, guesses, fills in. An agent is a work tool that reads the structure of your systems directly — your accounting software, your online store, whatever it works in — does one thing correctly, and stays out of everything else. We have built every agent that way — narrow, on purpose.

The agent shows what it would post before anything lands in — for example — your accounting software with Oskar. You approve — the agent is never the last link.

The preview is not theatre. When something is unresolved, it stays unresolved. We would rather have one line on hold than one error in the register. The approval is an explicit action — it does not carry over from the previous session, and it does not cover postings it has not seen.

When Oskar encounters a line it cannot categorise with confidence, it stops. It does not invent an answer.

Every flag has a reason — a new supplier, an unusual VAT rate, a sum that deviates from the monthly average. We explain why the agent stopped, not just that it stopped. And once you've helped us with an answer, the agent learns from it — so the same question doesn't come up again.

Bookkeeping happens in your own accounting software. We do not copy data out to our systems; all work is done against your own API access.

Access tokens are stored locally, are never written to a log, and can be revoked by you at any time — in your accounting system. A data processing agreement is ready when we set up; we go through it line by line before you sign.

We book supplier invoices, reconcile the bank, post receipts, and run month-end and year-end close. We do not do payroll or tax advice. We have not built our own accounting platform.

The list is short on purpose. What we do, we do to a standard we can answer for. What we do not do, we say so — and if you need it, we recommend someone who actually does it.

An agent with no human accountable for it is just a chatbot with access to your systems. Every Agaas agent has a professional — a human with formal expertise in the field — who owns the method the agent works by, reviews the work, and sets the standard it is measured against.

That is the difference between a tool that suggests and an agent that does. The accountable professional answers for the work — including when the answer is that something must be redone. No agent stands alone.