Skip to content

Private beta · January 2027

Software that someone answers for.

Warrantee's AI engineers take over your app, write down what it must always do, and prove every change against that list before it ships. If something we cover breaks, we fix it at our cost.

Sign-ups open with the private beta in January 2027. Live apps first.

Watch a real job, start to finish

An example app. Results that aren't examples come from our own recorded runs.

We adopt apps whichever tool built them

  • Lovable
  • Bolt
  • Replit
  • Cursor
  • v0
  • Base44
  • Hand-written

AI made building easy. Nobody answers for what it builds.

Every AI tool sells you more code. When that code breaks, leaks or quietly charges the wrong amount, the person left holding it is you.

91%

of 200 live apps built with AI coding tools had at least one security flaw. Two in three flaws were critical or high.

Deng, Fan and Meng, arXiv, 2026

48 days

a flaw in a popular AI app builder stayed open, exposing one app's 18,697 user records.

The Next Web, Apr 2026

Jan 2026

standard business liability policies began excluding losses caused by generative AI.

Verisk ISO forms, via Big I

Three things we put in writing.

Every step in detail

01

We write down what your app must always do.

Our planning agent reads your app and drafts its contract in plain English. You change it until it's right, then approve it. Each line becomes an automatic check.

  • Drafted from your code or your idea
  • Sentences you can read, not code
  • Nothing goes on it that can't be checked
Contract · shopnest.appApproved

What this app must always do

  1. 01Customers can sign in with email or Google.
  2. 02Checkout always charges the amount in the cart.
  3. 03One shop never sees another shop's orders.
  4. 04Order pages load in under two seconds.
  5. 05Nothing is deleted without the owner's say-so.
Approved by the owner

02

Every change has to prove itself.

Fixes, features, upgrades and security patches all meet the same gate. A change that breaks a single line of your contract doesn't ship, and your app stays on its last good version.

  • Checks the agents never get to see
  • Reviewed by an AI from a different company
  • You merge every change yourself

Change

Upgrade the login library to v5

  • Your contract11 of 12 · clause 01 fails
  • Your app's own testsAll pass
  • Hidden checksAll pass
  • Review by a second AINo critical findings
  • Security scanClean
Held backYour app stays on v4

03

If something we cover breaks, it's on us.

A failure your contract covers is ours to fix, at our cost, proven before it ships. Then the contract gets a new line, so the same failure can never come back unnoticed.

  • Re-fixed at our cost
  • Every failure becomes a new clause
  • Every claim recorded in your app's ledger

Warranty claim

Closed
  1. 09:12

    Reported

    Totals are off by one cent on discounted orders.

  2. 09:20

    Covered

    Clause 02: checkout always charges the amount in the cart.

  3. 14:05

    Re-fixed, at our cost

    Proven against every clause before it shipped.

  4. 14:05

    Clause 13 added

    Discounts never change a total by more than the discount.

Between changes, we keep watch.

Launch is where most help stops. It's where ours starts.

Uptime and speed

Your live app is probed around the clock. A slowdown or an outage opens an incident before your customers write in.

Security advisories

Every library your app uses is checked against new advisories. A patch is prepared and proven against your contract.

Incidents, investigated

Operations agents trace an alert to its cause and propose a fix, within strict limits and with every action logged.

Anyone's changes, checked

Hire a freelancer or use an AI builder again: their pull requests are checked against your contract too.

Our own test run · 18 Oct 2026

Proof, with the misses left in.

14 real tasks from open-source projects like Hugo, Gson, jsoup, Pydantic and more, graded afterwards by tests the projects' own maintainers wrote. Our agents never saw them.

3
delivered, all correct
0
wrong changes delivered
5
correct changes held back
6
not solved or unfinished
Every task, every number
gson · fixrunning
  1. 0:00RequestMake Gson refuse text that isn't valid in strict mode.
  2. 0:02PlannerWrites the tests that define done, then plans one change.
  3. 1:08DeveloperEdits JsonReader.java and adds the tests. One failing run, fixed.
  4. 1:57Test engineerWrites more tests that try to break it. They pass.
  5. 3:02ReviewerA model from a different company reads the change. No critical findings.
  6. 4:28GateEvery check passes. Delivered, ready to merge.
4 min 28 s
start to delivered
103
tool calls
8
AI model calls
$0.10
of AI time

One of the three delivered tasks, replayed from the steps our system recorded.

Who else answers for your app?

You and an AI builderA freelancerA rescue agencyWarrantee
Someone answers when it breaksYou doIf they're freeFor a short windowIn writing
A written contract of what your app must doAs an auditYou approve it
Every change proven before users see itSecurity scansDependsDuring the projectEvery change
Unproven changes held backAutomatically
A warranty on what the contract covers30 to 90 daysFor as long as we keep it

How pricing works. One fixed price to adopt your app and write its contract, then one monthly fee to keep it. Both are worked out from your app and agreed before any work starts.

The warranty, in plain words.

A promise about our own work, written down before we start. No small print about AI.

Draft terms for beta apps, not yet in force. Read the draft

Covered
A line of your approved contract that stops working after a change we made.
What we do
Fix it at our cost, prove the fix against the whole contract, and add a clause so it can't recur.
If we can't fix it
We refund what you paid for that work.
Not covered
Changes made outside Warrantee, outages of services we don't run, and anything not in your contract.

Questions, answered.

Ayush Saxena

Founder

Every claim on this site has a date beside it, and the misses are published too. Ask me anything about how it works.

Is this insurance?

No. It's a promise about our own work. If something on your approved contract breaks after a change we made, we fix it at our cost. If we can't fix it, we refund what you paid for that work. The terms are a draft until the private beta.

Who does the work?

AI agents: a planner, developers, a test engineer, a reviewer from a different AI company, a judge, and operations agents for live apps. Nothing ships unless it passes the contract you approved, and you merge every change yourself.

What is the contract?

Plain-English sentences about what your app must always do, like "checkout always charges the amount in the cart". Each one becomes an automatic check. You read and approve it before we start, and you can change it any time.

My app was built with an AI builder. Is it for me?

Yes. That is who we built Warrantee for, whichever tool built your app, or if a developer wrote it by hand. If its code can live on GitHub, we can adopt it.

How much does it cost?

One fixed price to adopt your app (write its contract and close the holes), then a monthly fee for keeping it. Both are worked out from your app and fixed before any work starts. Approving a job holds its price, and it's charged only when the work is proven and delivered. Beta members get a free contract preview first.

Will you touch my live app?

Never without your approval. We work in a copy of your code, and you merge every change yourself.

Can you build a new app too?

Yes. We build it, then keep it working under the same kind of contract.

When can I start?

The private beta opens in January 2027. We'll let people in by fit and by how many apps we can look after safely, starting with apps that are already live.

Put your app in writing.

Every result we have so far is on the proof page, each with its date and what it cost.

Sign-ups open with the private beta in January 2027. Live apps first.