Terms of Service

1. What NomadGPU is

NomadGPU is an operations service. We run your GPU jobs on your own Vast.ai account, on your behalf and within the limits you set. You pay Vast.ai directly for the GPUs you use. We do not resell GPU time, we do not hold your money, and we are not a GPU provider. Our fee covers only our service.

To do this you give us an API key for your Vast.ai account and, if you use your own storage, a storage key. You can revoke either at any time.

2. What we commit to

• We stop at your limit. You set spending limits per job and per day, week or month. When a limit is reached, the machine is released automatically. Limits are checked about every 30 seconds, so the final amount can exceed the limit by the cost of those last seconds and the marketplace's own billing rounding.
• The machine is released automatically at 100%. You get a warning at 80% and a message when the job stops.
• Your progress is kept. When a job stops — at a limit, on a machine failure or because you stopped it — the last saved checkpoint stays in storage, and the job can continue from it.
• We act on failures within minutes. A machine that disappears or stops is noticed within about a minute; a machine that hangs (no heartbeat from our agent, or a GPU that stops responding) within about 10 minutes; a training job that stalls with an idle GPU and no output within about 30 minutes. Then the machine is released and a replacement is searched for automatically. A job that stalls again on another machine is stopped instead, so as not to spend money on a likely code problem.
• Progress loss is limited to your last checkpoint. Our agent uploads each new checkpoint to storage within minutes of it being written (at most every sync_interval_s, 5 minutes by default). A resumed job starts from the last checkpoint that was fully uploaded. How often checkpoints are written is set by your code.
• A human answers within one business day in the support chat or by email; on the Managed plan, within the time agreed in your contract.
• You can stop a job at any time with the Stop button on your private status page, or by messaging us.
• You can revoke our access at any time by deleting the nomadgpu key in your Vast.ai account. From that moment we can no longer start or manage machines on it.

The simplest hard cap that does not depend on us: keep on your Vast.ai balance only what you are ready to spend. Vast.ai stops machines when the balance runs out.

3. What we do not guarantee

• Recovery time. When a machine fails we rent a replacement automatically. A job usually resumes 5–20 minutes later, but how fast depends on which machines are available on the marketplace at that moment and on how large your image and model are.
• Availability and prices of machines. They are set by Vast.ai and independent hosts, not by us.
• Third-party failures: outages of Vast.ai, individual hosts, storage providers or networks.
• Your code and its results. If your script fails because of a bug, a missing dependency or a wrong parameter, we show you the log and the error; fixing the code is on your side unless agreed separately. We do not guarantee the quality of trained models or generated outputs.
• Confidentiality on hosts. Machines belong to independent hosts and your job runs there unencrypted. Do not process personal, medical or otherwise confidential data this way. Checkpoints can be encrypted at rest in storage with your key.

4. Your responsibilities

• Use the service lawfully and in line with the Vast.ai Terms of Use.
• You confirm that neither you nor your owners are on sanctions lists (such as the US OFAC SDN list or EU lists), and that you do not act on behalf of such persons.
• Keep enough credit on your Vast.ai account for the jobs you ask us to run.
• Keep your contact details current so that alerts reach you.
• Third-party platform rules and indemnification. You hold sole responsibility for complying with the Terms of Service, Acceptable Use Policies and any export control or sanctions regulations of the underlying cloud providers (such as Vast.ai and RunPod). You agree to indemnify, defend and hold NomadGPU harmless against any claims, damages, penalties or account suspensions arising from your violation of these third-party terms.

5. Fees and billing

The service fee is billed after each period (week or month), with no prepayment. GPU time, disk and traffic are billed by the cloud provider to your account directly. You can cancel at any time. If the fee for a period is below $10, no invoice is issued for it: the amount is carried over and added to your next invoice.

Free trial. New Self-serve and Starter accounts get a free trial of 7 to 14 days (the exact end date is shown in your account): no service fee is charged for it. GPU time during the trial is billed by the cloud provider as usual. You can pause the trial up to three times (each pause up to 60 days, with no jobs running; the remaining days are kept) and extend it once by 7 days yourself in your account. When the trial ends and you have not confirmed that you continue (the "Continue" button in your account), running jobs are stopped — the machine is released and the progress is kept — and new jobs pause until you confirm. The service fee applies from the day after the trial.

Refunds. Payments for subscription periods or services that have already been rendered are non-refundable. If the service is terminated because of a violation of third-party platform rules, a suspected sanctions risk or a platform-level ban, no refund is issued for the remainder of the current billing period. Nothing in this section limits rights you may have under mandatory consumer protection law.

6. Data

• Your code, checkpoints and outputs belong to you.
• They are stored in your own storage bucket, or — on the Starter plan — in a separate folder of our shared bucket that only your jobs can access, with temporary keys limited to that folder.
• We keep job records (jobs, rentals, costs, events) to run the service and bill you, and an audit log of actions for one year.
• On request, or within 30 days after the end of the service, we hand over or delete your data in our storage.

7. Limitation of liability

To the extent permitted by law, our total liability is limited to the service fees you paid us in the three months before the event. We are not liable for indirect losses, lost profits or lost data beyond the last saved checkpoint, or for the acts of Vast.ai, hosts, storage or network providers.

Suspension by cloud platforms. NomadGPU acts solely as a management interface. We bear no liability for data loss beyond the last saved checkpoint, service interruption or financial damages resulting from the suspension, termination or access restriction of either our or your accounts by the underlying cloud platforms.

8. Ending the service

Either side can end the service at any time with notice by email or Telegram. We may end it immediately if continuing would create a sanctions or legal risk. When it ends, we stop all your jobs, keep your progress and you revoke our keys.

9. Changes and contact

We will announce changes to these terms at least 14 days in advance. Questions: ops@nomadgpu.com or the support chat.