Software has started doing things on your behalf: sending, buying, writing into systems. Here is what that changes, and what we do about it, without a single word you would have to look up.
Straight from the two people who build it. No sales team, because we haven't hired one.
EU hosted · your own keys · nothing you do trains anything
For three years, AI answered questions. You asked, it replied, and a person decided what to do next. That person was the safety mechanism, and nobody had to think about it.
In the last year, that step quietly disappeared. AI assistants now open your systems, fill things in, send messages, and spend money, hundreds of times a day, without stopping to ask. The person who used to check has been taken out of the loop, and in most companies nothing was put in their place.
Think of it as hiring someone brilliant, extremely fast, and very literal, who has keys to everything, and will never once come to your desk to check whether they should.
You would not run a company that way with a human. That is the whole of the problem, and none of it is about technology.
The first one happened to us.
Someone asks their assistant to draft a message to a contact at another company, explicitly not to send it. The assistant opens the website, writes the message, and starts sending. A human happens to be watching and stops it by hand.
A team asks an assistant to clean up a messy spreadsheet before sharing it. The spreadsheet contains real customer records. The assistant does exactly as asked, efficiently, thoroughly, and somewhere it should never have been.
The monthly AI bill doubles. Nobody can say which team, which project, or which task caused it, because the usage sits in one account and the work sits in fifteen places.
We apply your house rules before anything happens. Five rules, in plain terms.
Every action an assistant takes, what it did, on whose behalf, why, and what it cost, is recorded as it happens, not reconstructed afterwards.
You decide which actions are ordinary and which need someone to say yes first. The request goes to a person, in the place they already work.
Sending customer data outside the company, touching a system that is off-limits. Written once as a rule, applied everywhere, the same way every time.
Per team, per project, per task. Most people arrive here for the record and stay for this, because it is the number their finance director asks about.
European hosting, your own keys, and it can run inside your own walls. We do not keep your content, and nothing you do trains anything.
We would rather say them first.
When someone types into a desktop chat application, that conversation is encrypted between them and the vendor, and getting in the middle of it would mean doing something to their computer that we are not willing to do. What we can govern is the moment an assistant reaches out to act on a system, which is where the consequences live anyway.
If an assistant can act through a route we are not watching, then silence in the record means nothing. We spend most of our engineering effort closing those routes, and we will tell you which ones are still open rather than let you find out during an audit.
What did our AI do last quarter, on whose instruction, and what did it cost?
Not a dashboard nobody opens. A record your own people can read and hand to whoever asks: an auditor, a regulator, a customer, a board member who read something worrying over the weekend.
Both of us come. There is no sales person to hand you on to. If it turns out we have nothing useful for you, we will say so in the first ten minutes and give you the rest back.
TapPass.ai · control and evidence for AI that acts