Terms of service

1. Who this is

Ctrl AI is operated by Ctrl AI, Inc., incorporated in [TO CONFIRM: the state or country of incorporation] and registered at [TO CONFIRM: the registered address, as it appears on the incorporation record]. Using the site means agreeing to what is on this page. If you do not, do not use it.

Legal notices go to [TO CONFIRM: the contact address for legal notices].

2. What the service does

You ask your own assistant, in one prompt, for an honest audit of what it thinks it knows about you. You paste its answer here. We read that text into a set of decisions you can accept or reject by tapping, and — if you publish — we serve the ones you ratified back to the AIs you connect, one slice at a time, with a receipt for every fetch.

That is the whole of it. It is a settings layer, and every judgment in it is yours: nothing is decided for you, no decision control arrives preselected, and nothing an AI proposes takes effect until you ratify it.

3. What it explicitly does not do

It does not give advice. Not medical, legal, financial, professional or personal. It configures another company’s assistant — ChatGPT, Claude, Gemini or whatever else you connect — and what that assistant then says is its output, not ours. We do not review it, we do not stand behind it, and we are not responsible for decisions taken on it.

Better context should make those answers better. It cannot make them right, and nothing here should be treated as a judgment about you by us.

4. The audit is a model’s self-report, and it may be wrong

This is the product’s premise rather than a disclaimer at the bottom of it. Your assistant’s model of you is out of date and stated with confidence — and its account of that model is itself a model’s output. It can be wrong about your job, your projects, what you are working on now, and what it thinks it got wrong.

What we can promise is narrower and we hold to it. Every card carries the sentence it came from, and that quote is checked against your pasted text in code: a record whose quote is not in your reply is dropped and counted. That stops us inventing something your assistant never said. It does not stop your assistant being wrong about you, which is why you rule on every item rather than us.

5. No warranty

The service is provided as it is, with no warranty of any kind — no promise that it is available, that a reading succeeds, that a connected AI fetches your context when it should, or that anything is accurate.

Some of that is designed. When a spend ceiling is hit, a reading degrades to the parts that cost nothing rather than failing. When the counter itself is unreachable it fails open. When a budget expires you get a smaller true answer instead of a blank page. Those are deliberate choices about which way to fail, not guarantees about outcomes.

6. Acceptable use

• Your own context, not somebody else’s. Paste replies from assistants you use, about you. Do not paste an audit belonging to another person to build a profile of them, and do not connect an AI to a context that is not yours to publish.
• Stay inside the limits. 20 calls per caller per hour and 5,000 a day across everyone, per task. They are generous on purpose — re-running your own audit after a bad paste is what the product is for. Scripted or automated calls beyond them are not.
• Do not attack the thing. No attempts to reach another person’s context, receipts or credentials, and no probing the connection endpoints for anything other than your own account.

Because a shared account is common, be careful with what the audit surfaces about other people in your household. The product gives you a way to mark it; publishing it as though it were yours is on you.

7. The prompt is public and free

The audit prompt is printed on the home page, copyable in one tap, and in the public repository. Use it anywhere, with any assistant, commercially or not, with no attribution and no permission from us. It works perfectly well without ever coming back here.

8. Accounts and access

There is no password. An account is an email address plus a link we send to it, which works once and expires in 30 minutes. Anyone who can read that inbox can reach the context published to it — so keep the inbox secure, and treat losing access to it as losing access here.

Publishing without an account leaves the context reachable from that browser alone until the link is claimed. Unclaimed, it is deleted after 30 days. You do not need an account to run an audit.

9. Connections, and taking them back

Each AI you connect holds its own credential, so revoking one revokes that connection and no other. Revocation is immediate: the next fetch on that credential fails.

It cannot reach backwards. Context already given to an assistant lives in that provider’s conversation and logs under that provider’s terms, and nothing here can retrieve or delete it.

10. Changes and termination

You can stop at any time: revoke every connection, and ask us to delete what is stored. There is no self-serve delete button yet, so deletion is a request — the privacy page says how.

We can suspend or end access to anyone using the service against section six, or where keeping it running would harm somebody else’s data. Features can also disappear: this product deletes a feature the moment a measurement says it does not work, and has done so twice already.

The service is free today. [TO CONFIRM: whether and how paid plans are introduced, and the notice given before a free feature changes]

11. Liability

To the fullest extent the law where you are allows, Ctrl AI is not liable for anything that follows from what an AI says after being given your context, for lost or inaccurate context, or for any indirect or consequential loss.

[TO CONFIRM: the liability cap and exclusions the company’s counsel settles on, and the consumer rights that cannot be excluded in each market]

12. Governing law

[TO CONFIRM: governing law and the courts that hear a dispute — this must be settled by the company and is deliberately not guessed here]

13. Changes to these terms

These terms change when the product changes, and the date at the top moves with them. Continuing to use the service after a change means accepting it. There is no archive of earlier versions yet.