For the one-person company
Ship on Friday. Your testers used it on Thursday, and you already fixed what stopped them.
Pre-launch checkBefore your customers
Paste a link. AI testers, each with their own habits, device and inbox, use your product the way your customers will. You get back what broke, how to fix it, and a second test once you've fixed it. Websites and web apps today. iPhone, Android and Windows apps next.
What you can test
Checkouts, sign-ups, onboarding, dashboards, team pages. Testers use the real thing on the screen your customers hold, and circle the exact spot where it let them down, in their own words. Every card below is a real test on one of three practice apps we built with known problems, so we can measure what they catch.
Where testers go · today and next
Your customers
Paste your site. Kranth reads it, works out who would use it, and casts testers to match: where they live, what they use, how much patience they have, and why they came. Below is the cast for Kettle, a practice tea shop, and what happened when they tried to buy something.
When it breaks
Five ways a broken product comes back to bite you. The stories are made up. The bugs in them are real: our testers found every one in our practice apps before any customer did.
Talk to a tester · soon
When a tester gives up, you can ask them why. They answer from what they did and saw in the session. Below, our team questions Marcus, who left the checkout of our practice shop. Every answer is his own words from the recording.
Watch them work
Ten testers, one shop, played back. Each coloured cursor follows the path its tester really took through Kettle. The log on the right shows every click and every thought. Each tester's box below shows where they are and how much patience they have left. Pages turn red where people got stuck.
How we test the testers
A shop, a habit app and a dashboard, each with 20 problems we planted and wrote down first. Five testers used each app six times. A separate AI judge checked every report against that list. Here is what they caught, and what they missed.
They missed 22. Each miss is on the board above, and each one is something we are working on. For scale: one careful person on our team, testing by hand, found 10 of the same 60.
How it works
Nobody writes steps for them. Each tester has a background, a device and habits of their own. They use your product alone and decide what to do next, the way a person would. These are the ten from the map above, at Kettle's checkout.
Add a test login or a test card if they need one, and an errand if you have one in mind. Or let them wander.
Each one gets a background drawn from 101 personality types, a phone or a laptop, an email inbox, and habits: how patient they are and what they press first.
Each tester gets their own browser and their own session, with no script. They report what stops them, slows them or just looks wrong, with the step and the screenshot.
We merge the reports and rank them by how many testers hit each one. Every item has the steps, the fix and a way to check it. Fix it, then send the testers back in.
These are AI testers. They find what breaks and what confuses. Then go talk to the real people.
Nobody tells them where to click. That is how they find the paths your tests never walk.
In our practice shop, all ten hit the card bug and said so in their own words.
For your agent · soon
Add Kranth to Claude Code, Cursor or any MCP client, or call the REST API. Your agent sends testers to the preview link, they use it, and the fix list comes back as a comment on the pull request. Push the fix and they check again. This part comes after the web app. Below is how it will look, using problems our testers really found on Kettle.
Live event stream
/kranth test "buy any tea as a guest"Kranth checks again on your next push. Comment /kranth retest to run it now.
claude mcp add kranth.cursor/mcp.json/kranth test on a pull requestPOST /v1/testsWhere this goes
Today, ten testers go into one website and come back with a fix list. Next, they test every pull request, and phone and desktop apps too. The end game is a standing crowd shaped like your real customers, in every language and on every device, using your product every day. Your AI agent fixes what they hit before a real customer ever does.
Ship on Friday. Your testers used it on Thursday, and you already fixed what stopped them.
It writes the change, reads the fix list on the pull request, fixes it and asks the testers again. You review a green pull request.
Every market, language, device and kind of patience your customers have, without booking a lab or paying a panel.
They never meet the bug. That is the whole point.
Pricing
One credit sends one tester through one task, on a phone or a laptop, and pays for their part of the fix list. Every plan gets the same testers and the same checks. Plans start at $39 a month, or top up from $10. Plan credits you don't use roll over to the next month.
250 tester credits a month
800 tester credits a month
4,000 tester credits a month
Testers that work behind your login, on your own devices and in your own regions. We never train on your data, and the contract says so. Talk to us to set it up.
[email protected]Paste a link and tell them what to do. They use it on their phones and laptops, and you get back what broke, how to fix it, and a retest.