How it works
One contract. Every change proven against it.
01 · Start
Adopt the app you have, or build a new one.
For an app you already have, we read its code and sort every part into three piles. For a new app, we build it first, then keep it the same way.
Keep
What works and is safe stays exactly as it is. Most of your app, usually.
Harden
What works but could leak, break or cost you: fixed, each fix proven by a check.
Replace
What can't be made safe: rebuilt, with your approval, and only that part.
02 · The contract
What your app must always do, in your words.
Each line is a sentence you can read, and each becomes an automatic check. You change it until it's right, then approve it. If something can't be written as a check, it doesn't go on the list, and we tell you.
What this app must always do
- 01Customers can sign in with email or Google.
- 02Checkout always charges the amount in the cart.
- 03One shop never sees another shop's orders.
- 04Order pages load in under two seconds.
- 05Nothing is deleted without the owner's say-so.
03 · Custody
Every change goes through six checks.
Fixes, features, upgrades and security patches all take the same road. When a change fails, your app stays on its last good version while we try again.
01
Your list
Every line you approved runs as an automatic check. A change that fails one doesn't ship.
02
Hidden checks
Extra tests the AI that wrote the change never sees. Their results go to the judge, so code that only passes the visible tests is caught.
03
A second AI
A reviewer from a different company reads every change. The writer never approves its own work.
04
A security gate
Leaked keys, risky patterns and libraries with known holes stop the change.
05
A judge
One last look at all the evidence together before anything reaches you.
06
You
You merge the change. Until then, nothing in your app is different.
04 · The warranty
If something on your list breaks for your users, it's on us.
For beta apps, the warranty will work like this:
- Covered: a behaviour on your approved list that breaks for your users after a change we made.
- What we do: fix it at our cost and prove the fix against your whole list. If we can't fix it, we refund what you paid for that work.
- And then: your list gets a new line, so the same failure is checked for ever after.
- Service credits: for apps we keep, a covered failure that reaches your users takes credit off next month's fee, up to the whole fee.
- Not covered: changes made outside Warrantee, behaviours that aren't on your list, and outages of services we don't control.
These are draft terms, not yet in force. Read the draft.
Who does the work
AI agents: one plans, others write the code and the tests, a reviewer from a different company checks it, and operations agents watch for trouble. No person reviews your code unless that changes, and if it does, this page will say so.
What we turn down
If we don't think we can finish a job, we say so before you pay, not after. And we won't promise what we can't check: if a behaviour can't be written as a test, it stays off your list.