The sales conversation for KEYSTONE

A word-for-word starting point. Make it sound like you after the first few calls.

The shape of every call (same for all 25)

Open with their world, not ours (1 minute). One sentence about a problem they likely have. Never start with "CAIN-42 is a Byzantine trust fabric"; say that later, when they ask how it works.
Discover (5-10 minutes). Ask the questions below and listen. Write down their exact words; you will use them in the follow-up email.
Show (5 minutes). Run the live demo with them watching. Nothing is pre-recorded: every button calls the real product. Let them pick an attack.
Offer (2 minutes). One offer, one price, one date. Do not list all 25 products.
Next step (1 minute). Before you hang up, agree a date: payment, a scoping call with the founder, or a "no".

Openers (pick one)

"When a customer asks you to prove a security fix, what do you send them today?"
"Most AI fixing tools hand you a change and say trust me. We hand you a sealed proof your customer can re-run."
"If one of your pen-test findings is still open, I can show you in five minutes what a provable fix looks like."

Discovery questions

Ask, then stop talking. Each question tells you whether they have the problem and how much it matters.

Tell me about the last security finding you had to fix. Who asked for it, and how did you show it was done?
Do you use AI tools to write code or fixes? Does your customer know?
What happens to the deal while a finding is open?
Who on your side spends time convincing customers a fix is real?
If you could send a single file that proves the fix, who would you send it to first?
What language is the affected code in?

The five-minute live demo

Share your screen. Everything below calls the real product, so it can occasionally be slow or rate limited (a "429" means too many tries from one address; wait a minute). Practise it three times before your first call.

Live demo: paste the sample Python with the forged-token flaw and run the detector. Point out that no AI made this call.
Signed record: show a real signed record: the old test failed, the new one passed, and the seal.
Verify it yourself: show how the customer re-checks the record offline. Say: "you don't have to trust us".
Attack lab: press run-all to show that tampered records are refused.

Objections and honest answers

They say	You answer
"We can just ask the AI to write a test."	You can, but the same model wrote the fix and the test. KEYSTONE's proof uses an independent test the model never saw, run in a sandbox.
"Can I watch it fix something live?"	Live AI fixing is switched off on the public site today; the demo shows the detector live and real signed records from earlier runs. On an engagement we run it on your code.
"Does it work on our language?"	The detector is strongest on Python. For other languages talk to the founder before promising anything.
"How many fixes has it done?"	Honestly, a small number so far: the public evidence holds four fixes that meet today's rule. It proves the method; your engagement adds your own.
"Isn't this what our pen-test firm does?"	They find problems. We close one and hand you proof. Many clients use both.

If you do not know an answer: "Good question. I will get you a precise answer from our engineers by tomorrow." Then email the founder. Never guess.

Next: the offer, the price and how to close →

FIRME AI CORPORATION · CAIN-42 is a Byzantine trust fabric. Numbers marked "our own check" were run by us, not by an independent party. Nothing on these pages is a certification.