What people ask before signing

No. The application is installed on your machine and the AI model runs on it too. Your credentials are entered in your own application and never leave it. I hold no copy of your data, run no server that keeps it, and have no permanent access.

No, never. It stops at the draft. Review, finalisation and sending remain human actions. This is the single most important limit of the system and it is not negotiable, even on request.

Not a blocker. If the software exposes an API, we use it: faster and more robust. If it does not, the agent drives its web interface the way a person would — opening pages, searching, typing and reading back. That is exactly what happens with Erplain.

Allow up to one minute per document when it drives a web interface, line by line, with checks along the way. The gain does not come from raw speed but from the fact that nobody has to do it: the agent runs on its own while you do something else.

Below roughly a hundred documents a month, the automation pays for itself too slowly to be worth doing. The calculator on the pricing page says so plainly: it shows \u201cvolume too low\u201d below the threshold. Better to know in thirty seconds than after three weeks.

The agent has to be adjusted. That is precisely what the monthly maintenance covers. When an agent relies on a web interface rather than an API, this is not a hypothetical risk: it is a certainty on an unknown date, and it is planned for.

The agent blocks that line and states the reason. It never fills in a guess and never matches a near-enough name. An agent that invents values to avoid stopping is a dangerous agent: in accounting, a silent error costs more than a visible halt.

You are, because you are the one who approves it — and that is exactly why the agent stops at the draft. My job is to make any error visible before approval: quantities read back, duplicates searched for, incomplete documents blocked.

Send me three real, representative documents. That is enough to see whether the format is consistent enough. I reply within 48 hours with an honest verdict: automatable, partly automatable, or leave it alone.

The first month is refunded if the agent does not do what is written on the pricing page. And before you even order, the demonstration on your own documents shows you the result.

No. The agent runs on your machine and you can re-run it without me. I deliver a tool, not a subscription to my availability. Maintenance covers changes, not the right to use it.

From two or three. With a single agent its own screen is enough and the HQ would be a useless layer. Beyond that, nobody knows any more which one ran, which one flagged something, or where the month stands.