How we protect your work
Written in plain English. No jargon, no fine print surprises. Every promise on this page is enforced by how the product is built, and you can verify each one in your project's record.
What happens, step by step
What we promise, and what we do not
We believe trust comes from honesty about limits, not from big words.
| We promise | How you can check |
|---|---|
| No AI model ever receives your complete work, your real names, or the order of your document. | Open any project to see the map of what was shared and the exact text sent for each piece. |
| Each AI model, and each company across all its models, receives less than 20% of your work. | The number on every project page is measured, not estimated. |
| No AI can tell that two pieces mention the same person, company, or secret. | Each piece uses its own one-time codes, so the same name looks different in every piece. |
| Your most sensitive lines never leave at the Zero Knowledge and Air Gap levels. | Held-back pieces are marked in red on the list of pieces. |
| Your combined result is not stored unless you choose to save it. | You are asked directly after every run. Your choice is recorded. |
| Every send is recorded in a tamper-evident log you can verify. | The audit trail on every project page, with a live verification check. |
| You can permanently delete everything, any time. | Deletes require you to type the action, then they are gone. |
What we do not promise: we cannot control what outside AI companies do with the small pieces they receive. That is exactly why we never send them enough to matter. If you want zero exposure, use the Air Gap level: nothing is sent at all.
Only what you ask for, nothing else
IPVault does exactly one job: improving and protecting your documents. The system only accepts the ten listed goals, builds every AI request from our own fixed templates, ignores any instructions hidden inside documents, and limits how fast accounts can submit work. It cannot be used as a general AI service, by design.