Service-Specific Terms

These Service-Specific Terms govern the technical and commercial characteristics of individual AhuraSense Services. They supplement the Terms & Services and apply only to the relevant Services.

Where a particular Service is subject to additional conditions presented during ordering, described in Documentation, or stated in an Order Form, those conditions form part of the Service-Specific Terms for that Service.

Services may have different availability, durability, performance, billing, backup, data residency and support characteristics. Customers are responsible for reviewing the specifications of the Service selected before deploying production workloads.

Resources are supplied subject to capacity, regional availability, technical compatibility, quotas and applicable compliance controls.

AhuraSense may perform maintenance, security updates, hardware replacements, migrations, capacity balancing and other operational changes reasonably required to operate the platform. Where technically possible, we will endeavour to minimise disruption.

Applicable availability commitments are governed exclusively by the Service Level Agreement or a separately negotiated commitment. Customers must not rely on undocumented implementation behaviour.

Virtual machines provide Customer-controlled computing environments. Unless expressly sold as a managed operating-system service, the Customer is responsible for:

• Guest operating system administration and software installation.
• Patching, application security and malware protection.
• Firewall configuration and credentials.
• Licensing and backup.

Charges continue for resources for as long as they remain allocated according to their billing state. Stopping an operating system from within a virtual machine does not necessarily release the underlying compute resource or stop billing.

Termination or deletion of an instance may permanently remove local or ephemeral storage associated with it.

Bare-metal and dedicated servers may require provisioning time and may be subject to minimum billing terms.

Hardware remains the property of AhuraSense or its infrastructure supplier unless expressly sold under a separate written agreement.

Hardware replacement is the Customer's remedy for verified hardware failure unless an applicable SLA or Order Form provides additional remedies.

Customers must not tamper with physical equipment, firmware, management controllers or facility systems except where expressly authorised.

GPU services provide access to accelerator resources that may include NVIDIA or other GPU families identified during ordering. Specific accelerator models and quantities remain subject to availability until confirmed. On-demand capacity may become unavailable between separate provisioning requests, even where capacity was available previously.

Customers are responsible for:

• Checkpointing training jobs.
• Storing durable model state outside temporary runtime storage.
• Designing workloads to tolerate failure appropriate to the relevant Service.

GPU errors may include device resets, memory errors, driver failures, interconnect faults, node failures and other accelerator-specific failure modes. AhuraSense may replace, migrate or restart affected nodes as reasonably required.

GPU capacity may be subject to enhanced KYC/KYB, export-control screening and end-use verification.

Reserved capacity becomes binding when the reservation is confirmed in writing, activated through the Account, or accepted through an Order Form. Unless otherwise stated, the Customer is responsible for the committed charges for the entire reservation term regardless of utilisation.

Reservations are specific to the quantity, GPU family, region and term specified in the Order. A reservation does not entitle the Customer to replacement with a newer accelerator generation.

Where identical hardware cannot reasonably be restored following a permanent failure, AhuraSense may offer:

• Equivalent or better capacity.
• An agreed commercial adjustment.
• Termination of the affected portion of the reservation.

For large dedicated clusters, provisioning milestones, capacity acceptance, interconnect specifications, replacement arrangements and any spare-node commitments should be recorded in the applicable Order Form.

Customers may deploy models owned or licensed by them or use models made available through AhuraSense. Third-party models remain subject to their respective licences and provider conditions.

AhuraSense does not grant rights in third-party model weights except to the extent AhuraSense is authorised to do so.

Customers are responsible for determining whether model licences permit commercial use, fine-tuning, redistribution, output use and the Customer's intended application.

Customers are responsible for datasets, licences, privacy permissions, data-subject rights, model rights, safety evaluation and resulting model artefacts.

Training and fine-tuning workloads may fail because of:

• Application errors or checkpoint configuration.
• Memory exhaustion or distributed-training failures.
• Dependency errors or accelerator faults.
• Other causes.

Customers should configure checkpointing suitable to the expected workload duration. Unless expressly stated otherwise, compute consumed before a Customer-controlled workload fails remains billable.

Managed databases include operation and maintenance of the database service components identified as managed in the applicable product description.

Customers remain responsible for:

• Database users and application credentials.
• Schema design, queries and access policies.
• Network exposure and retention requirements.
• Application-level data integrity.

Backups, point-in-time recovery, replicas and high-availability features apply only where enabled or included in the selected plan.

For managed Kubernetes, AhuraSense manages only the platform components identified as managed in the relevant product documentation.

Customers remain responsible for workloads, container images, Kubernetes RBAC, secrets, network policies, ingress configuration, application security, persistent volumes and worker-node configuration except where explicitly managed.

Storage capacity remains chargeable for as long as the relevant volume, bucket, snapshot or reserved storage allocation remains active.

Deleting compute does not necessarily delete attached persistent storage. Customers must verify deletion workflows carefully.

Versioning, object lock, replication, backup and lifecycle policies apply only when explicitly configured or included.

Public IP addresses, bandwidth, egress traffic, load balancers, private networking, BGP, BYOIP and other network services may be separately metered or limited.

IP addresses remain AhuraSense or supplier resources unless separately assigned to the Customer. Customers must maintain accurate routing, firewall and DNS configuration.

BYOIP customers must have legal authority to originate or authorise use of the relevant prefixes and must maintain valid route and registry information where required.

Domain registration is subject to the rules of the relevant registry, registrar, ICANN where applicable, and any country-code administrator.

A domain is not registered until confirmed by the responsible registry. Registration requests may fail even after an initial availability check.

Customers are responsible for accurate registrant data, renewal, transfer credentials, expiry dates and applicable dispute procedures. Registry fees and renewal pricing may change independently of AhuraSense pricing.

Open-source software, marketplace images, model weights, operating systems and other third-party components are governed by their respective licences. AhuraSense is not responsible for a Customer's failure to comply with third-party licensing conditions.

We may remove a third-party component where:

• Its provider withdraws it.
• A licence changes.
• Continued distribution becomes unlawful.
• A material security issue arises.