1. Scope

1.1 These AI Specific Terms apply to the AI services that ON DIVERSITY CLOUD EUROPE, S.L. ("DIVERSITY") provides under an Order Form: dedicated inference endpoints, model as a service, training clusters, GPUs by the hour, and vector databases.

1.2 AI engagements, such as assistants, fine-tuning, agents, speech, document AI, computer vision, MLOps and AI Act readiness, are also governed by the Professional Services Specific Terms (/contracts/professional-services/) for the engagement work. Dedicated GPU servers rented on their own are provided under the Bare Metal Specific Terms (/contracts/bare-metal/), not these Specific Terms.

1.3 These Specific Terms supplement the General Terms (/general-terms/) and the Acceptable Use Policy (/acceptable-use-policy/), whose Section 7 sets out the rules for the AI services. They do not repeat what those documents settle. In particular, fees, invoicing and late payment (General Terms, Section 7), suspension (Section 8), personal data (Section 9 and the DPA), confidentiality (Section 10), liability (Section 14), notices (Section 18), and governing law and courts (Section 23) apply as written there.

1.4 Where these Specific Terms depart from the General Terms, they say so and prevail, as Section 3.2 of the General Terms provides. Capitalised terms not defined here have the meaning given in the General Terms.

2. Definitions

  • "Endpoint": a dedicated inference endpoint that serves one model for the Customer.
  • "Model as a Service": access to shared models hosted by DIVERSITY, billed per use.
  • "Training Cluster": several GPU servers networked together for one Customer.
  • "GPUs by the Hour": GPU capacity rented for short periods.
  • "Third-Party Model": a model provided under a licence or use policy of someone other than DIVERSITY.

3. Description of the service

3.1 Endpoints. An Endpoint serves either the Customer's own model or an open-weight model that DIVERSITY hosts, as agreed for each case and stated in the Order Form.

3.2 Model as a Service. Model as a Service gives access to shared hosted models and is a separate product from Endpoints. The models available change over time.

3.3 Training Clusters. The size of a Training Cluster is agreed for each project and stated in the Order Form.

3.4 GPUs by the Hour. GPU capacity is available for short periods, as stated in the Order Form.

3.5 Vector databases. Vector databases run on DIVERSITY's managed PostgreSQL, which is also governed by the Cloud Specific Terms (/contracts/cloud/).

3.6 Tenancy. Unless the Order Form states otherwise, requests to Endpoints and to Model as a Service run on a shared GPU cluster. Exclusive use of a GPU is a feature of Bare Metal GPU servers, not of those services.

3.7 Discovery call. Before an engagement, DIVERSITY offers a free 60-minute discovery call with an engineer, followed by a written diagnosis within one week. The call and the diagnosis do not oblige either party to enter into an Agreement.

4. Location

4.1 The AI services other than vector databases, and the GPUs that run them, are at present in Amsterdam (AMS1) only. Vector databases run in the site selected in the Order Form, as managed databases do under the Cloud Specific Terms. The choice of region in Section 9.4 of the General Terms does not apply to the AI services: Customer Data they process is processed and stored in Amsterdam.

4.2 Apart from the choice of region, Section 9.4 of the General Terms applies to the AI services, including the Customer's prior written consent and the notice it requires for any change of location.

5. Customer Data and models

5.1 Training. DIVERSITY's commitment is: "Your data never trains anything except your own models."

5.2 Third-Party Models. Before the first use of a Third-Party Model, the Customer must accept its licence or use policy, which DIVERSITY presents to it (Acceptable Use Policy, Section 7.3). The Customer is responsible for complying with it.

5.3 AI Act. The obligations that Regulation (EU) 2024/1689 (AI Act) places on the provider or the deployer of an AI system built with the AI services remain the Customer's (Acceptable Use Policy, Section 7.2).

5.4 Prompts and outputs. DIVERSITY does not store the prompts sent to Model as a Service or to an Endpoint, or the outputs returned, beyond the processing of each request. DIVERSITY keeps usage records (the time, the model and the number of tokens of each request) for 90 days, for billing and troubleshooting, and longer only where the law requires it.

5.5 Fine-tuned weights. Model weights fine-tuned on Customer Data belong to the Customer, whether DIVERSITY or the Customer carries out the fine-tuning. The licence of the base model continues to apply to them. The Customer may download them at any time during the term and for 30 days after the service ends. DIVERSITY then deletes them.

5.6 Outputs. Model outputs are generated automatically and may be inaccurate. The Customer is responsible for checking them before relying on them.

6. Service levels and support

6.1 Endpoints and Model as a Service are covered by the 99.9% availability commitment (General Terms, Section 5.2). Vector databases are covered as managed databases under the Cloud Specific Terms.

6.2 GPUs by the Hour and Training Clusters are not covered by the SLA. A GPU hardware failure under them is repaired, and is not an SLA event.

6.3 Support is provided under the Support Specific Terms (/contracts/support/).

7. Term and billing

7.1 The term of each AI service is stated in the Order Form. Section 16 of the General Terms applies unless the Order Form provides otherwise.

7.2 Model as a Service is billed per use. GPUs by the Hour are billed for the time used. Other AI services are billed as the Order Form sets out. Prices appear only in the Order Form.

8. End of service

Where an AI service is a data processing service under the Data Act, Section 17 of the General Terms applies, and the Order Form lists the exportable data and formats.

9. Exclusions

Unless the Order Form provides otherwise, the AI services do not include:

(a) exclusive use of a GPU for Endpoints or Model as a Service;

(b) a choice of region other than Amsterdam, except for vector databases (Section 4.1);

(c) any assurance that the Customer's AI system conforms to the AI Act or other law;

(d) the engagement work described in Section 1.2, which is ordered separately;

(e) the SLA for GPUs by the Hour and Training Clusters.