Terms of service
The operating boundaries for the Compute private preview.
Service
Compute is a private-preview service for running supported Python workloads on newly provisioned cloud GPU machines. Availability, supported hardware, and execution times are not guaranteed.
Each run is intended to own one machine from provisioning through confirmed termination. Users remain responsible for their source, dependencies, inputs, and outputs.
Credits and charges
The payment model is prepaid credit with a $10 minimum top-up. Usage is charged at the locked provider list rate plus the disclosed platform fee, metered by started machine minute.
A machine that never becomes ready is not charged to the customer. There are no automatic refunds. Exceptional credits and dispute outcomes are decided manually within 24 hours — see the ops dispute runbook.
Support: include your run id or request id when contacting the operator listed on the status page.
Acceptable use
Users must have the right to process submitted code and data and must not use the service to violate law, abuse provider infrastructure, evade quotas, interfere with other users, or compromise systems.
Compute may stop a run or suspend preview access to protect users, providers, or the service.
Preview risk
The private preview may be incomplete, unavailable, or changed. Workloads should be reproducible, and Compute should not be the only copy of source, data, or results.
Warranty, liability, governing law, termination, and formal notice provisions are intentionally not asserted by this draft and must be completed before these terms become effective.