AI Platform
Terms of Service

This agreement covers LLM inference, GPUaaS and fine-tuning — what you own, what we commit to, and what happens when something goes wrong.

Effective
6 October 2025
Provider
Hyperfusion Computing Services FZEJebel Ali Free Zone, UAE
Governing law
DIFC lawDIFC Courts, exclusive jurisdiction
Contents
17 sectionsSLA and 2 appendices

No training on your content

We do not use your inputs, outputs or datasets to train models unless you opt in. §8

Processed in the UAE

Customer Content stays in the UAE unless you configure multi-region. §8

99.6–99.9% availability

Inference targets are monthly and per region, with service credits when missed. SLA

DIFC law and courts

Governed by DIFC law with exclusive DIFC Courts jurisdiction. §15

1

Definitions

“Services”
Provider’s AI platform services, including (a) LLM Inference (model APIs and endpoints); (b) GPUaaS (on-demand or reserved GPU instances); and (c) Fine-Tuning (training/adapter jobs, evaluation, hosting of tuned models).
“Customer Content”
Inputs, prompts, datasets, code, and configurations you provide; and “Output”: model responses generated from your inputs.
“Usage Data”
Telemetry and operational metrics about how the Services are accessed (e.g., request counts, error rates, latency).
“Fine-Tuned Model”
A base or hosted model that has been adapted using Customer Content via our Fine-Tuning Services.
“Credits”
Pre-paid monetary value redeemable for Services.
“Subscription Plan”
A time-based plan with included entitlements, limits, and overage pricing.
“AUP”
Acceptable Use Policy (Appendix A).
2

Account & Eligibility

You must provide accurate information, keep credentials confidential, and ensure your users comply with this TOS and the AUP. You are responsible for all activity under your account and any API keys issued to you.

3

Service Specific Terms

3.1

LLM Inference

Access
We provide REST endpoints and SDKs to call supported models (first-party and third-party).
Rate limits
We may enforce per-key or per-account limits to protect platform stability.
Caching
For performance, we may cache prompts or embeddings in encrypted form for short periods; see §8 for retention.
Safety
We apply automated safety filters and abuse detection, which may block or modify requests/outputs to comply with the AUP and applicable law.
3.2

GPUaaS

Provisioning
Instances are provided as on-demand or reserved; available GPU types vary by region. You remain responsible for OS/application hardening within your instance.
Usage measurement
Billed per second or minute, plus any storage/egress.
3.3

Fine-Tuning

Your datasets
You must have all rights needed to use datasets for fine-tuning and to instruct us to process any personal data in them.
Artifacts & weights
Unless a base model license forbids, you own the rights in your Fine-Tuned Model weights as between you and us; we retain platform IP (tooling, pipelines, evaluators). If the base model license does not allow weight ownership or export, you receive a right to use the tuned model on our platform.
Hosting
We host your Fine-Tuned Model for inference unless you export it and the base license allows export.
4

Acceptable Use

Do not use the Services to violate law or rights, generate or disseminate malware, infringe IP, cause harm, or process content contrary to UAE public order/morality. We may suspend/terminate for violations and will act on lawful takedown requests.

The full policy is set out in Appendix A.

5

Pricing, Credits & Subscriptions

5.1 Pricing & TaxesPricing is published in the dashboard or Order Form and may change on 10 days’ notice (except for committed/subscription terms already in effect). Fees exclude UAE VAT (currently 5%) and other applicable taxes, which you must pay.

5.2 Pre-Paid CreditsPurchase & use. Credits may be purchased in your account and automatically applied to usage. Expiration. Credits expire 9 months after purchase unless stated otherwise in an Order Form. No refunds/transfer. Credits are non-refundable and non-transferable except where required by law. Currency. Credits are denominated in [USD/AED]. Exchange rates and taxes may affect effective purchasing power.

5.3 Subscription PlansTerm & renewal. Plans are month-to-month or annual and auto-renew unless you cancel in the console before the next term begins. Included entitlements. If you exceed plan limits, overage rates apply. Non-refundability. Fees are billed in advance and are non-refundable, including for partial periods, except where required by law.

5.4 Invoices; Late PaymentInvoices are due within 15 days of issue unless paid by card/credits at time of purchase. We may charge late fees/interest permitted by law, suspend Services for non-payment, and recover collection costs.

6

Intellectual Property & Feedback

Your rightsYou retain all rights in Customer Content and, as between you and us, in Outputs, subject to third-party model licenses and applicable law.

Our IPWe own the platform, branding, models we create, APIs, documentation, and usage analytics.

FeedbackYou grant us a royalty-free license to use feedback to improve the Services.

7

Third-Party Models & Services

Some models are provided under third-party licenses (open-source or commercial). Your use of those models via our platform is subject to those licenses; where a third-party license conflicts with this TOS, the license terms for that model control your model-specific rights and restrictions. We are not responsible for third-party services outside our reasonable control.

8

Data Protection, Privacy & Security

RolesFor Hosted Services we act as processor of Customer Content, and you are the controller.

No training on Customer Content by defaultWe do not use your inputs, outputs, or datasets to train or improve our or others’ models unless you opt-in in writing or in the dashboard.

Operational logsMinimal request metadata is retained for up to 30 days for fraud/abuse, support, billing, and safety (longer if required by law or to resolve open tickets). You may request log minimization where technically feasible.

Fine-tuning storageDatasets and intermediate artifacts are stored encrypted at rest and deleted within 30 days after you delete the project or upon written request (unless longer retention is required by law or for dispute resolution).

Data localizationCustomer Content is processed in the UAE unless you configure a multi-region or consent to cross-border transfers consistent with UAE PDPL/DIFC/ADGM rules.

SecurityWe maintain administrative, technical, and physical safeguards consistent with UAE national information security requirements (e.g., UAE IAS) and a shared-responsibility model. See Appendix B.

Sub-processorsWe may use vetted sub-processors; we remain responsible for their performance and will provide a current list upon request.

9

High-Risk Use; Output Caution

You must not rely on Outputs as a substitute for professional advice or for autonomous decisionmaking in safety-critical contexts (e.g., medical diagnosis, legal decisions, aviation, nuclear). Use humans-in-the-loop and evaluate outputs for accuracy and bias before use.

10

Service Changes; Beta/Preview

We may improve, add, or remove features. Beta or preview features are provided “as is” and may be suspended at any time.

11

Term; Suspension; Termination

By youCancel renewals in the console at any time; your plan remains active until the end of the current term.

By usWe may suspend or terminate for material breach (including AUP violations or non-payment) after notice and an opportunity to cure where feasible; immediate action may be taken to protect the platform or comply with law.

EffectOn termination, your access ends. You have 10 days to export Customer Content and Outputs unless legally restricted; thereafter we delete content per §8.

12

Warranties; Disclaimers

Except for the SLA below, the Services are provided “as is” and “as available.” We disclaim implied warranties (merchantability, fitness, non-infringement) to the maximum extent permitted by law.

13

Indemnities

By youYou will defend and indemnify us against claims arising from (a) your Customer Content or your use of Outputs; (b) your breach of this TOS or AUP; or (c) your violation of law or third-party rights.

By usWe will defend and indemnify you against third-party claims alleging that our platform (excluding models you select, Customer Content, or third-party components) directly infringes IP rights, with standard exclusions (combinations, modifications, non-current versions, AUP violations).

14

Liability

LimitationEach party’s aggregate liability in any 12-month period is capped at the greater of (i) fees paid by you to us in that period for the affected Services or (ii) one month of fees.

ExclusionsNo liability for indirect, special, incidental, consequential, punitive, or lost profits, subject to non-excludable law.

15

Governing Law; Dispute Resolution

This TOS is governed by DIFC law; the DIFC Courts have exclusive jurisdiction. We may seek enforcement or injunctive relief in any competent court if needed.

16

Export & Sanctions

You represent that you are not subject to sanctions and will not use the Services in violation of UAE or applicable international export/sanctions controls.

17

Changes to this TOS

We may update this TOS with 30 days’ notice. Continued use after the effective date constitutes acceptance.

SLA

Service Level Agreement

Summary of commitments, monthly and per region, excluding maintenance and the exclusions in §4 below.

Availability targets by service
ServiceAvailability targetKey SLOs
LLM Inference API
control & data plane
99.6% Standard
99.9% Premium Support add-on
p95 latency objectives published per model family; request acceptance within 5 minutes during autoscaling
GPUaaS99.5%Provisioning SLO: 90% of instance requests fulfilled within 15 minutes when possible for in-stock SKUs; Instance MTTR: 4 hours for host failures
Fine-Tuning Jobs99.0% scheduling availabilityJob start SLO: 95% start within 60 minutes of entering “queued” (subject to quota & capacity); Artifact retention: per §8

1) How availability is measured

LLM Inference “Downtime”5xx error rate >5% or request timeouts >5% over a rolling 5-minute window, measured from Provider-side monitoring in the relevant region.

GPUaaS “Unavailability”(a) API cannot create/attach/boot instances; or (b) >10% of running instances in a region crash due to Provider fault and are not auto-recovered in 180 minutes.

Fine-Tuning “Unavailability”Platform cannot accept new jobs, or job queue halts for >30 consecutive minutes due to Provider fault.

2) Scheduled & emergency maintenance

ScheduledAt least 72 hours’ advance notice; windows are excluded from SLA calculations.

EmergencyPerformed as needed to address critical security or stability risks; excluded from SLA.

3) Response & resolution targets

Support response and restore targets by severity
SeverityResponseTarget restore
Critical (sev-1)1 hour4 hours
High (sev-2)2 hours8 hours
Medium (sev-3)4 hours24 hours
Low (sev-4)8 business hours48 business hours

4) Exclusions

Force majeure, your code or configurations, quota exhaustion due to misuse, third-party services you control, your network or ISP, DDoS or abuse you trigger, and any use outside the AUP or rate limits.

5) Service credits

Service credits apply to the affected service in the region for the month:

Service credits by monthly availability
Monthly availabilityCredit
Below target by up to 1.0%5% of the monthly fee for the affected service
Target − 2.0%10%
Target − 5.0%25%

Request credits via support within 15 days of month-end with incident details. Credits are applied to future invoices and are the exclusive remedy for SLA failures.

A

Acceptable Use Policy

You must not use the Services to:

  • Violate law or rights; process or generate content contrary to UAE public order/morality; or store/process illegal content.
  • Infringe IP, privacy, or publicity rights; scrape or de-identify datasets unlawfully; or re-identify individuals.
  • Generate malware, exploits, or facilitate harm (e.g., phishing, deepfakes to defraud, instructions for violent/illegal acts).
  • Circumvent safety systems, rate limits, or access controls; probe or stress the platform except under an approved test plan.
  • Train models on third-party content without rights or required consents; upload special-category personal data without a lawful basis.

Enforcement. We may warn, filter, throttle, suspend, or terminate; in urgent cases we may take immediate action to protect the platform and comply with law.

B

Shared Responsibility & Security Snapshot

Provider secures

The core infrastructure, control plane, and managed runtime; implements monitoring, logging, incident response, and data-at-rest/in-transit encryption consistent with UAE IAS; manages platform patches.

Customer secures

Account identities, API keys, datasets, prompts, outputs, and GPUaaS guest OS and applications; configures access control and retention; promptly reports incidents.

Questions about these terms

Write to info@Hyperfusion.io or book a call. For how we handle personal data, see the Privacy Policy.