Your agents read the same rules before every job. Anything risky stops and asks you first.

A real decision, waiting on a real answer.
“I have explained our return policy eleven times. To the same assistant.”
None of that is a model problem. Your agents are brilliant and they have nowhere to keep anything.
How it works
Connect it once. Every agent you already pay for plugs into the same thing, and so does every person you trust.
How you price. Who may spend. What happened last quarter and what you decided about it. Plain files, in plain words, in a folder you own.
Every agent reads it before it starts. That is the whole trick: they never boot up empty again.
Spending money. Emailing a customer. Changing a price. The work gets done right up to that line, then it waits for you, as one question with the reasoning attached.
It arrives as a link. You can answer it from your phone in the time it takes to read four lines.

Say “yes, but keep festive stock out of wholesale” and that sentence is kept, in your words, exactly as you typed it.
The next run already knows. So does the agent you switch to next year. You are not training a chatbot, you are writing down how your business works, one real decision at a time.
Everything above lives in ordinary files in your own storage, not locked inside our database. You can read them without us. You can take them and leave without asking.
That is deliberate. The reason your agents forget today is that their memory belongs to whoever built the tool. Yours should not.

A sample company, with every file open to read. Browse it.
Mainmind is in early access and we are taking a small number of businesses at a time, so we can sit with each one properly.