Your vault already carries its own instructions.
Every AI that connects to it is told the house rules automatically at the start of every chat. You do not have to do anything for that to happen.
Pasting the rules into your AI app's settings as well makes them stick harder. It matters most on ChatGPT and Grok, where the automatic instructions are weaker and get forgotten partway through a long conversation.
Five minutes, once per app.
Copy this whole block. Replace "VAULT" with your vault's name, exactly as it appears in your AI app's connector list.
My business memory lives in my VAULT connector.
1. Start every business question there: search the vault and answer from
its notes, not from memory.
2. When we agree on something new (a price, a policy, a rule, a way of
doing things), file it into the vault before we finish.
3. Email is draft only. Never send, never delete.
4. When I say wrap it up, finish the session: file any new facts, write a
short journal entry, and tell me what was saved.Claude. Settings, then Profile, then "Instructions for Claude".
ChatGPT. Settings, then Personalization, then Custom instructions.
Grok. Settings, then Customize.
Same block in all three. Do not reword it for each app.
Rule 1 stops the guessing. Without it, ask an AI what your cancellation policy is and it will write you a perfectly reasonable cancellation policy that is not yours. With it, the AI searches your notes first and answers from what is actually true. If your vault does not know, it says so instead of making something up. That is the single most valuable line in the block.
Rule 2 stops the leaking. You work something out in a chat, the chat ends, the answer is gone. This line makes filing part of the conversation instead of a thing you meant to do later.
Rule 3 is the safety rail. The vault itself cannot send or delete email, so this rule is not what makes you safe. It is what stops the AI trying, arguing, or wasting your time offering. Drafts wait for a human. Always.
Rule 4 teaches it two words. "Wrap it up" becomes a real instruction instead of a figure of speech. See What is worth remembering.
Claude reads the vault's instructions properly and holds onto them. The connector hands over the house rules at the start of the session and Claude treats them as standing orders.
ChatGPT and Grok are looser. They will connect to the vault and use the tools, but they drift. Twenty messages into a chat, ChatGPT will happily answer a question about your refund policy from general knowledge rather than searching your notes, because nothing recent reminded it to look.
Custom instructions sit at the top of every conversation in those apps. Putting the rules there is what stops the drift.
If you only use Claude, this post is optional but still worth ten minutes. If you use anything else, treat it as required setup.
Test it. Open a fresh chat in that app and ask a question you know the answer to.
"What is our refund policy?"
If it answers with your actual policy and mentions where it got it from, you are set. If it answers with something generic and plausible, the connector is not attached, or it is not searching. Say "search the vault" and watch what happens.
Do this once per app. It takes a minute and it tells you whether the setup worked.