Laster katalogen …
Laster katalogen …
Dette dokumentet er kun publisert på engelsk.
Last updated 14 October 2026 · Effective 15 October 2026
1.1 These Terms of Service ("Terms") govern access to and use of the Stav.ai platform, API, dashboard, Stav Tunnel and related services (the "Service"), provided by Stav.ai AS, a Norwegian limited company (org. no. 937 371 853), Bekkelagsterrassen 9D, 1177 Oslo, Norway ("Stav", "we", "us").
1.2 The Service is offered only for business and professional use: to businesses, public bodies and other organisations, and to individuals acting in the course of their trade, business or profession ("Customer", "you"). It is not offered to consumers. Users must be 18 or over. If you accept these Terms on behalf of an organisation, you confirm that you are authorised to bind it.
1.3 The following documents form part of the agreement and apply in this order of priority if they conflict: (a) any signed order form or separate agreement; (b) the Data Processing Agreement ("DPA") at stav.ai/legal/dpa, which applies to every workspace when you accept these Terms and which we also sign on request; (c) these Terms; (d) the Acceptable Use Policy ("AUP") at stav.ai/legal/aup; (e) the supplier terms and model licences listed at stav.ai/legal/supplier-terms; (f) our documentation. The current list of sub-processors is at stav.ai/compliance#sub-processors.
2.1 The Service is in public beta. Features, models, Inference Providers, sovereignty grades, prices and interfaces may change, be limited or be withdrawn without notice while we develop the product.
2.2 During the beta the Service is provided "as is" and "as available". We do not offer a service level agreement, uptime commitment or support response time unless agreed in writing.
2.3 You should not use the beta as the only route for workloads where an outage or incorrect output could cause harm to life, health or critical operations.
2.4 We may invite you to give feedback. You grant us a free, perpetual right to use feedback to improve the Service, without obligation to you. Feedback does not include your Customer Content.
3.1 You must provide accurate information, including your organisation's legal name, country and organisation or VAT number, and keep it up to date.
3.2 You are responsible for all activity under your workspace and API keys, for keeping credentials secure, and for the users you invite. Tell us without undue delay at support@stav.ai if you suspect unauthorised use.
3.3 During the beta we may decide who to admit, and may limit volume, rate or available models per workspace.
4.1 Stav is a gateway. We route your requests to third-party inference providers ("Inference Providers") through their endpoints ("Provider Endpoints"), or to infrastructure you connect yourself through Stav Tunnel or your own provider keys ("BYOK"). We do not develop the models and do not operate the hardware they run on.
4.2 Sovereignty grades. We grade Inference Providers and Provider Endpoints on an L0–L4 scale, based on criteria such as compute country, ownership and jurisdiction. The criteria and sources for each grade are shown on each provider page. A grade is our assessment at the time stated. It may be revised, and it is not a certification.
4.3 Workspace policy. You can set requirements in your workspace, API key or router policy: a minimum sovereignty grade, permitted compute countries or regions, zero data retention, no training on inputs, and blocked Inference Providers or provider countries. We serve a request only from a Provider Endpoint that meets every requirement in that policy, based on the facts we hold for that endpoint when the request is made. If a fact that a requirement depends on is unknown, the endpoint is treated as not meeting it. If no Provider Endpoint meets your policy, the request is refused and the refusal is recorded in your request log. The facts we hold come from the Inference Providers, public sources and our own assessment. They are not legal advice or a guarantee that an Inference Provider meets any particular legal requirement. You remain responsible for deciding which policy your use requires.
4.4 Supplier terms. For traffic routed to an Inference Provider, that provider's terms, code of conduct and the licence of the model used, as listed at stav.ai/legal/supplier-terms, apply in addition to these Terms. You accept them by accepting these Terms. You do not need a separate agreement with any Inference Provider. You shall ensure that your users and end users comply with them. Under BYOK, your own contract with the Inference Provider governs its processing and billing, and we act only as a conduit.
4.5 Changes to models and Inference Providers. Inference Providers add, change and withdraw models, prices and terms all the time, and handling that is part of what the Service does. We may add or remove Inference Providers, Provider Endpoints and models at any time. Changes that follow from an Inference Provider's own changes to its prices, models or terms take effect when they take effect at that provider. An Inference Provider may change, update or replace the model or model version behind a Provider Endpoint, or serve a request with a different model, without notice to us. We are not responsible for such changes or for differences in output that result from them.
4.6 Stav Tunnel. Where Stav Tunnel is available and you route requests to machines you operate, you are responsible for that infrastructure, its security and its lawful use. The facts we show for such endpoints, such as compute country, are your own declaration.
5.1 "Customer Content" means prompts, files, inputs and outputs sent through the Service. As between you and Stav, you own your Customer Content.
5.2 We process Customer Content only to provide, secure and support the Service, as set out in the DPA. We never use Customer Content to train or fine-tune models, and we do not sell it.
5.3 Prompt content. We do not store the content of prompts and outputs by default. Your workspace may switch on prompt capture for a time-limited window. Captured content is masked for sensitive data unless your workspace allows exact capture, is encrypted, is stored in Norway, and is deleted when the retention period you set ends. Some Inference Providers keep request content for a limited time to monitor for abuse, where their terms allow it. Where we know of such retention, we show per Provider Endpoint whether it applies. "Zero data retention" means that the Inference Provider does not store the content of a request or its output after the request has been served. A provider with zero data retention may still keep the content of a specific request that it has flagged for abuse, fraud or a security incident, or that the law requires it to keep. It may also keep logs and request metadata to detect and handle abuse, for security or for billing. Inference Providers implement zero data retention in different ways under their own terms. We rely on those terms and the facts we hold for each Provider Endpoint, and we are not responsible for how an Inference Provider implements zero data retention.
5.4 Metadata. We log request metadata, such as time, model, Inference Provider, Provider Endpoint, compute country, the policy applied, token counts, cost and outcome, to provide billing, the request log and audit evidence. Request and routing logs are kept for 90 days, the audit log for 24 months, and billing records for as long as Norwegian law requires.
5.5 Sub-processors. For personal data in Customer Content, Stav acts as your processor and you act as controller, and the DPA applies. You give us general authorisation to engage the Inference Providers and platform vendors listed at stav.ai/compliance#sub-processors, and to add or remove them. The list shows each sub-processor with the date it was listed and is updated whenever it changes. When we add a new Inference Provider, we list it and notify workspace owners by email no later than when it starts serving requests. It serves your requests automatically when it meets the requirements in your workspace policy, without your approval. If you object to a sub-processor, you can block it in your policy or close your workspace.
5.6 No training on inputs. "No training on inputs" is on by default for every workspace. While it is on, we route only to Provider Endpoints whose provider states that it does not use inputs for training, according to the facts we hold for that endpoint. If you switch it off, your requests may be served by Inference Providers whose terms allow them to use inputs for training.
5.7 You are responsible for having a lawful basis for the data you send, for not sending special categories of personal data or data about criminal convictions unless we have agreed to it in writing (DPA section 7), and for any transparency obligations towards the people whose data you process.
6.1 Output is generated by third-party models, may be inaccurate, incomplete or biased, and may resemble content generated for others. You are responsible for reviewing output before relying on it.
6.2 Where you build AI systems on the Service or make output available to others, you are the provider or deployer of those systems under the EU AI Act and are responsible for the obligations that follow, including disclosing that content is AI-generated where the law requires it.
7.1 You must comply with the AUP. A breach of the AUP is also a breach of the supplier terms.
7.2 You may build your own products and services on the Service. Unless we agree otherwise in writing, you may not resell the Service, give third parties direct access to the API or to your API keys, or use the Service to build a competing inference or routing service.
7.3 Sanctions and export control. You confirm that neither you nor your end users are located in, ordinarily resident in, or acting for a country, territory or person subject to Norwegian, EU, UN or US sanctions or export restrictions, and you will not use the Service in breach of them. You state your country at sign-up and keep it accurate.
7.4 You are responsible for your users and end users as if they were you.
8.1 Usage. Usage is charged at the price shown in the Service for the Provider Endpoint that served the request, at the time the request is made.
8.2 Tiers. Your workspace has a customer tier, which we set based on your paid purchases. Your tier sets your rate limits and any free allowance for requests made with your own provider keys (BYOK). The current tiers are at stav.ai/pricing.
8.3 Credits. Usage is paid from prepaid credits in your workspace's billing currency. When you buy credits, the platform fee, the purchase limits and any VAT are shown before you pay. Under BYOK, the Inference Provider bills you directly for the tokens, and we charge the platform fee shown on stav.ai/pricing.
8.4 When your credits run out, requests to priced models are refused. We do not serve on credit. You can set spend caps and automatic top-up.
8.5 Purchased credits expire after the period stated when you buy them, which is never less than 12 months. Sign-up and promotional credits may expire sooner, as stated when they are granted. Credits are used oldest first. We notify you before credits expire.
8.6 The platform fee is non-refundable. Unused credits are non-refundable, except where the law requires a refund or where we terminate the Service without cause, in which case we refund unused purchased credits. Credits cannot be transferred. Sign-up and promotional credits have no cash value.
8.7 Changes to prices and tiers. Payments are processed by Stripe. Usage prices follow the Inference Providers' prices and can change at short notice. We may change the tiers and their terms, and move your workspace to another tier. We give at least 30 days' notice by email or in the Service of changes that make your tier more expensive or less favourable to you, and you may close your workspace before they take effect. Changes in your favour may take effect immediately. Credits you have already bought keep the expiry date stated when you bought them.
9.1 You may stop using the Service and close your workspace at any time.
9.2 We may, without notice, stop routing to one or more Inference Providers for you, for an API key or for a type of request, where continued routing would breach the AUP or the supplier terms, or would put our agreement with that Inference Provider at risk.
9.3 We may suspend your access immediately for serious or repeated breach of the AUP, the supplier terms or Section 7.3, or where the law requires it. In other cases of material breach or non-payment, we will give you written notice and 14 days to remedy before suspending.
9.4 We may terminate these Terms for convenience with 30 days' notice, or end the beta and offer new terms for general availability.
9.5 After termination, we delete or return Customer Content as set out in the DPA. We keep billing records as Norwegian accounting law requires.
9.6 We may close a workspace that has had no activity for 12 months. We notify the workspace owner by email at least 30 days before, and we do not close a workspace that holds purchased credits that have not expired.
10.1 Each party will keep the other's non-public information confidential and use it only for the purposes of this agreement.
10.2 We implement appropriate technical and organisational security measures, described in the DPA.
Stav owns the Service, the sovereignty grades, the catalogue and the documentation. These Terms give you a limited, non-exclusive, non-transferable right to use the Service during the term. You may not reverse engineer the Service or use it to build a competing grading or routing catalogue.
To the extent the law allows, and except as expressly stated in these Terms, the Service is provided without warranties of any kind, including fitness for a particular purpose, accuracy of output, or uninterrupted availability. We are not responsible for the availability, quality or conduct of Inference Providers or models, other than our obligations under the DPA in respect of sub-processors.
13.1 Neither party is liable for indirect or consequential loss, including lost profit, lost revenue, loss of data or loss of goodwill.
13.2 Each party's total liability under this agreement is limited to the greater of (a) the amounts paid by the Customer to Stav in the 12 months before the event giving rise to the claim, and (b) EUR 100.
13.3 These limitations do not apply to gross negligence or wilful misconduct, to the Customer's payment obligations, to the Customer's indemnity under Section 14, or to liability that cannot be limited by law.
You will indemnify Stav against claims by third parties, including Inference Providers, arising from your Customer Content, your use of output, or a breach of these Terms, the AUP, the supplier terms or a model licence by you, your users or your end users.
We may update these Terms. We will give at least 30 days' notice by email or in the Service of changes we make that materially affect you, and you may terminate before they take effect. Changes that follow from an Inference Provider's own changes (Section 4.5), and changes required by law or for security, may take effect sooner.
16.1 Neither party may assign the agreement without consent, except that Stav may assign it in connection with a merger or sale of the business.
16.2 Neither party is liable for delays caused by events beyond its reasonable control.
16.3 These Terms are governed by Norwegian law. Disputes shall be resolved by the Oslo District Court (Oslo tingrett) as the agreed venue.
16.4 Contact: support@stav.ai.