Terms of service

These terms are between the founder of Qevra, acting as an individual while the service is in private development (“Qevra”, “we”), and the business whose workspace uses the service (“you”). Once a company is incorporated its details will be named here, and it will take over this agreement unchanged.

You accept them by creating a workspace or by calling the API. If you are doing that for a company, you are confirming you may bind it.

Qevra is an API. You index your own product knowledge; you send it a customer’s question; it returns an answer built from that knowledge, a confidence score, the fragments the answer came from, suggested actions, and a signal when it believes a human should take over.

It is not a helpdesk, an inbox or a ticketing system, and it does not contact your customers. What you do with an answer is entirely up to your own product.

Qevra is in private development. Endpoints may change, features described as planned may not ship, and there is no service level agreement. Where the documentation marks something as planned, treat it as a statement of intent rather than a promise.

This is the clause that matters most, so it is not buried at the bottom.

Answers are produced by a language model working from the knowledge you indexed. They can be wrong. The confidence score is a measure of how well the question matched your own material — it is a signal, not a guarantee, and a high score on a badly written document is still a bad answer.

Qevra escalates instead of answering when it has nothing to go on, when retrieval is weak, or when it cannot ground an answer in your sources. That behaviour is the point of the product, but it is not a filter that catches every mistake.

You decide how much human review to apply before an answer reaches a customer, and you remain responsible for what your product says. Qevra is not suitable for use as medical, legal or financial advice, and nothing it returns should be treated as such.

Keep your credentials and your API keys to yourself. A key is shown once and stored only as a digest, so we cannot recover one for you — we can only help you revoke it and issue another. Anything done with a valid key counts as done by you, so revoke immediately at Dashboard → API Keys if one leaks.

Test keys and live keys follow the same code path; only live keys are charged. Use test keys while you are building.

Do not index content you have no right to use. Do not send us special categories of personal data — health, biometrics, political or religious views and the like — or payment card numbers, in questions or in documents.

Do not use Qevra to generate anything unlawful, to attempt to extract another workspace’s data, to probe or overload the infrastructure, or to resell raw model access dressed up as your own product. Do not present a generated answer as coming from a named human who did not write it.

There are no published rate limits today. That is an absence, not an invitation: traffic that threatens the service for everyone else may be throttled or cut off.

Your documents, your questions and the answers returned on them stay yours. You grant us only the permission needed to run the service: to store that content, to split and embed it, to send the relevant parts to the model provider named in the privacy notice, and to show it back to you.

We do not train models on your content and we do not use it to improve the service for anyone else. When you correct an answer, the correction becomes a document in your own workspace and stays there.

The software, the interface and the name remain ours. Aggregate figures that identify nobody and no workspace — total requests served, say — we may use freely.

Qevra is prepaid and usage-based. An answer delivered costs $0.02. A new workspace starts with $5.00 of credit, which is a grant rather than a purchase: it is not money, cannot be withdrawn and is not refundable in cash.

Three things are deliberately free, and stay free: an escalation is never charged, because refusing to guess is the behaviour we want to encourage; an answer you reject as wrong is refunded to your credit; and indexing knowledge costs nothing, however much of it you index.

Top-ups are taken through Stripe. Card details are entered on Stripe’s page and never reach us. Credit is added only when Stripe confirms the payment, and every movement appears in your ledger.

Credit does not expire, so what you buy stays available for as long as the workspace does. We do not otherwise refund purchased credit, except where the law requires it. When credit runs out, calls that would be charged stop being served until you top up.

Prices are in US dollars and exclude any tax that may be due. We may change the price with reasonable notice; credit already bought is unaffected.

We run the service with care but do not promise it will be uninterrupted. It depends on providers we do not control — the database host, the platform it runs on, the model router — and an outage at any of them is an outage here.

We may change or withdraw parts of the service. Where a change would break a working integration we will give notice and, where we reasonably can, keep the old behaviour running under the existing API version.

You may stop at any time; ask us and we will close the workspace and delete its data. We may suspend or close an account that breaches these terms, that is being used to harm the service, or where the law requires it — with notice first, unless the situation makes notice impossible.

After closure the data is deleted, except the billing ledger, which accounting law requires us to keep.

The service is provided as it is. To the extent the law allows, we exclude implied warranties, including that Qevra will be fit for a particular purpose or that its answers will be accurate.

We are not liable for lost profits, lost business, lost data or indirect losses. Our total liability across all claims in any twelve months is capped at what you paid us in that period.

Nothing here limits liability that cannot be limited by law: fraud, wilful misconduct, gross negligence, or death and personal injury caused by negligence.

We may update these terms. Material changes are announced by email to account owners before they take effect, and the date below changes. Continuing to use the service after that means you accept the new version; if you do not, close the workspace.

Italian law governs this agreement, and the Italian courts have jurisdiction over any dispute arising from it. If you are a consumer rather than a business, the mandatory protections of your own country still apply and nothing here removes them.

Questions about any of this go to hello@qevra.dev.

Last updated 8 August 2026. See also the privacy notice and the security page.