Terms of service

These terms cover the prepaid inference API at https://api.tiyuvta.ai/v1 and the account console at inference.tiyuvta.ai. Using either means you accept them.

1. Who you are contracting with

The service is provided by tiyuvta, a sole proprietorship based in Israel. Contact: hello@tiyuvta.ai.

Payments are taken by Paddle.com Market Ltd acting as merchant of record. Paddle is the seller on your receipt, handles sales tax and VAT, and issues your invoice. Their buyer terms apply to the payment itself; these terms apply to the service you spend the credit on.

2. What is sold

Access to the models listed on the pricing page over an HTTP API compatible with the Chat Completions shape, billed per token at the rates published there. For Qwen3.8 27B that is $0.38 per million input tokens, $0.12 per million cached input tokens and $2.60 per million output tokens.

Chat completions only: no embeddings, no fine-tuning and no batch API. The live /v1/models response states what each model accepts. We add models over time; we will not silently remove one you are using — see section 8.

3. Credit

• Credit is prepaid, denominated in US dollars, and does not expire.
• Requests draw down credit at the metered rate. When the balance cannot cover a request, the API returns 402 and generates nothing.
• Our metering record is the billing record. It counts tokens, not requests, and it is visible to you in the console and in every API response.
• Credit is not a deposit, not a security, not transferable between accounts, and cannot be sold on.
• Promotional credit — including the $10 sign-up grant — has no cash value and is not refundable. Paid credit is refundable under the billing terms.

4. Automatic top-up

Automatic top-up is off unless you switch it on. When you do, you authorise us to charge your saved payment method for the amount you chose, each time your balance falls below the trigger you chose, up to the monthly ceiling you chose. You can turn it off at any time in the console, effective immediately. We stop at your ceiling and email you rather than charging past it, and we switch it off after two failed charges.

5. Your account and keys

• Sign-in is by one-time email link. Keep access to that mailbox; anyone who has it has your account.
• API keys are shown once. We store only a prefix and cannot recover a key for you.
• You are responsible for what your keys do, including spend, until you revoke them.
• One account per person or company. Creating extra accounts to claim promotional credit again is a breach of these terms and we will reclaim the credit.

6. Acceptable use

You may not use the service to:

• Break the law of your country or ours, or infringe someone's rights.
• Produce sexual content involving minors, or material whose purpose is to harass, defame, or incite violence against real people.
• Build weapons, malware, or credential-theft tooling.
• Attack the service: probe or bypass authentication, exhaust capacity deliberately, or resell raw access in a way that hides that we are the provider.
• Present model output as professional medical, legal, or financial advice from a qualified person.

Model output is generated text and can be incorrect. Consequential decisions require human review.

7. Availability, limits, and support

The service is provided without an uptime service level agreement. Measured performance and the conditions it was measured under are published on the model card. Requests that fail with our error do not consume credit.

Rate limits apply per account for abuse control, at a lower rate while you are spending promotional credit, and we raise them on request for a real workload. We do not publish a limit or a capacity figure, because capacity is scaled to demand rather than fixed. Under load, requests are shed with 429 rather than held open. Support is by email.

8. Changes

• Prices: we may change them with at least 14 days' notice by email. Credit you already bought is spent at the rate in force when you spend it, so a rise is a rise from the notice date, not retroactive.
• Retiring the model: at least 30 days' notice by email, and any unused credit becomes refundable on request even if it is past the refund window.
• These terms: material changes get an email and a new date at the top of this page.

9. Suspension and termination

You can stop at any time; unused paid credit is refundable under the billing terms. We may suspend an account for a breach of section 6, for fraud or chargeback abuse, or when a legal authority requires it. Except for fraud we will tell you why, and unused paid credit is refunded.

10. Liability

The service is provided as it is. To the extent the law allows, our total liability for any claim is limited to the amount you paid us in the three months before the claim, and we are not liable for lost profit, lost data, or indirect loss. Nothing here limits liability for fraud or for anything that cannot be limited by law.

11. Intellectual property

You keep your prompts. You keep whatever rights you have in the output. We claim no ownership of either. The model weights are Apache-2.0 from Qwen and are subject to their licence; we host them, we do not own them. Our serving stack, site and console are ours.

12. Law

These terms are governed by the law of Israel. If you are a consumer, mandatory protections of your own country still apply.