SERVICE LEVEL AGREEMENT FOR DEDICATED INSTANCES

1. Scope

This Service Level Agreement (the "SLA") applies to a dedicated instance of Sovereign AI Cloud ordered by a business customer under a signed order form that references it. It does not apply to prepaid credits, pay-as-you-go use, shared capacity, free or trial use, or any capability labelled PLANNED. Capitalised terms have the meaning given in the Terms of Service. Values in square brackets are proposed and are fixed in the order form once the infrastructure decision is taken.

2. Definitions

• Dedicated Instance — the GPU node, its resident models and the API endpoint reserved for the customer under the order form.
• Monthly Uptime Percentage — 100 % minus the percentage of minutes in a calendar month during which the Dedicated Instance was in Downtime.
• Downtime — a period of at least two consecutive probe intervals in which the health probe of the Dedicated Instance (GET /health and a minimal chat completion) fails from both of our measurement locations, excluding the periods of Section 5.
• Service Credit — a credit against the next monthly fee of the Dedicated Instance, calculated under Section 4.
• Scheduled Maintenance — maintenance announced under Section 6.

3. Availability commitment

We commit to a Monthly Uptime Percentage of at least [99.5 %] for each Dedicated Instance.

4. Service Credits

4.1 To receive a Service Credit, the customer submits a claim to [CONTACT EMAIL] within 30 days after the end of the month in question, naming the instance, the period and, where available, the request identifiers of failed requests.

4.2 We confirm the claim against the status-page record and our monitoring within 15 business days and apply the credit to the next invoice. Credits are not paid in cash and do not survive termination, except that a credit confirmed before termination is refunded under the Refund and Credits Policy.

4.3 Service Credits are the customer's sole and exclusive remedy for a failure to meet the availability commitment. The total of Service Credits in a month cannot exceed [50 %] of the monthly fee of the affected instance.

5. Exclusions

Downtime does not include unavailability caused by:

• Scheduled Maintenance announced under Section 6, up to the monthly allowance;
• the customer's own acts or omissions, its applications, its network, its API keys or its exceeding of the limits agreed in the order form;
• suspension under the Terms of Service or the Acceptable Use Policy;
• force majeure, including failures of the public internet, of the hosting provider's upstream connectivity or power beyond our reasonable control, and denial-of-service attacks that we mitigate with reasonable diligence;
• a model's known limitations (a slow or wrong answer is not Downtime; a failed health probe is);
• capabilities labelled PLANNED or BUILT-but-not-measured in the order form, and beta features;
• emergency security maintenance of up to [4] hours per month, announced as soon as reasonably possible.

6. Maintenance windows

6.1 Scheduled Maintenance takes place in the window [02:00–06:00 Europe/Tallinn] and is announced at least [48] hours in advance by email to the customer's administrators and on the status page.

6.2 Scheduled Maintenance is limited to [8] hours per calendar month. Maintenance beyond that allowance counts as Downtime.

6.3 Where technically possible, a model or runtime update is applied without interrupting the instance; where an interruption is unavoidable, it is scheduled in the window.

7. Measurement

7.1 Availability is measured by our status probe from [two] measurement locations in the European Union, at an interval of [60] seconds, against the instance's GET /health endpoint and a minimal chat completion against the resident default model.

7.2 The results are published on the status page, which keeps a [90]-day history per component. The status-page record is the primary evidence for a claim; the customer may submit its own logs, and in case of discrepancy the parties reconcile in good faith on the basis of request identifiers.

7.3 The status page is a static page regenerated by the probe; it sets no cookies and loads nothing from third parties.

8. Support

Dedicated instances include [2] hours of engineering support per month, by email, with a first response within [one business day] (Europe/Tallinn). Incidents that cause Downtime are handled with priority regardless of the allowance.

9. Term and changes

This SLA applies for the term of the order form. We may improve it at any time; we may reduce a commitment only with 30 days' notice and only for the next renewal period, in which case the customer may cancel before the renewal.