Terms of use
Last updated 10 November 2026 · facts checked 10 November 2026
Draft. This text has not been reviewed by a lawyer yet and is not final. It describes how the service is built today, in plain words. Passages in square brackets are still open and will be written by a lawyer. It may change before the service opens to the public.
Who we are
Kintar is a brand of Schumacher Management AB, organisation number 559599-6777, a company in Sweden. “We” means that company. You can reach us at [email protected] (mailbox still to be created before launch).
By signing in or using the service you agree to these terms. If you use it for a company, you confirm you may bind that company.
What the service does
You connect a code repository and describe a change. Kintar copies the repository to a working area, lets AI models write the change, and then checks it with your own build and tests (and, where it applies, a run in a browser). You are shown the result and decide whether to accept it. Optionally it can open a pull request.
This is early access. The service can be slow, wrong or unavailable, and features can change.
You pay only for a verified and accepted change
This is the one rule the product is built around:
- When a run starts, we reserve its price in units. A unit is $0.10, fixed. A reservation is not a charge.
- We charge only if the change is verified and you accept it. Otherwise the whole reservation is released.
- “Verified” means deterministic proof: your build and tests pass (or the browser run passes). An AI model never decides that something is verified; it may only double-check.
- The charge can never be higher than what was reserved at the start.
- A failed attempt, a run that is stopped, and a change you reject are free. If 10 attempts in a row on one repository fail verification within 7 days, new starts on that repository pause for a while, because failed attempts still cost us compute.
- After you accept a change you can still reject it for 7 days.
One exception to the rule above: computer control, if you turn it on, is billed by use and not by verified change. Computer control is off unless you turn it on. A session is billed by use, not by verified change: the engine and action cost divided by 0.7, rounded up to whole units, never more than the units reserved at the start. A session that ends in an error is not billed.
Buying units is handled by Stripe (see the privacy policy). Card payments are built but are not switched on for the public yet: today everything runs on the free tier below. Before we take real payments, these terms will be updated with the payment terms, receipts and VAT, and you will be asked to agree again.
The free tier
- Up to 10 attempts in total per account. Failed attempts count here, so retries cannot go on forever.
- Up to 3 verified and accepted small changes. A small change is about 5 files and about 10 minutes of work.
- You have 7 days to reject a verified change; a rejected change does not use one of the 3.
- One free tier per GitHub account and one per repository. A repository that one account has used is closed to others' free tier. This stops repeated sign-ups.
- No card is asked for or stored.
We can change or end the free tier. Anything you have already accepted stays yours.
Your code and what comes out
Your repository stays yours. The change a run produces is yours to use, subject to the licences of the code it builds on and to the terms of the model providers (see the privacy policy for who they are). You promise that you have the right to let us copy and work on the repository you connect, and that you will not connect code you may not share with third parties.
Your code and your task text are sent to an AI model provider. For each run, the task and the parts of your code it needs go to the provider that runs it: today Anthropic (processing in the United States or globally) or Together AI (United States). Together AI hosts open-weight models, some first published by Chinese companies; your data goes to Together in the United States, not to those companies and not to China. Which provider is used depends on the task. You can see the model used on each change. Details, retention and sub-processors are in the privacy policy.
Read a change before you accept or merge it. Verification shows that your build and tests pass, not that the change is correct, secure or fit for your purpose.
What you may not do
- Use the service to build malware, to attack others, or to break the law.
- Try to get around the limits, for example with several accounts to get more free attempts.
- Put secrets you do not want a model provider to see in your repository or task. Models receive the parts of your code that a task needs.
- Interfere with the service or other users.
We may suspend an account that breaks these rules. We do not lock accounts for non-payment while payments are off.
Data, deletion and training
What we store, for how long, where it is processed and how to have it deleted is in the privacy policy. In short: we do not train models on your code or tasks without your explicit consent, and you can delete your account yourself in Settings.
Availability, liability and law
The service is provided as it is, during early access, without a promise of uptime or of results. [Limitation of liability: to be written by a lawyer.]
[Governing law and place of dispute resolution: to be confirmed by a lawyer; we expect Swedish law.]
Changes and ending
We will tell you about material changes to these terms before they apply. You can stop using the service at any time and ask us to delete your account (see the privacy policy). We can end the service with reasonable notice.