Terms of Service

Effective October 20, 2026 · Version 2026-10-20

These terms are an agreement between Velobyte, Inc. (d/b/a InferOptimal) ("InferOptimal", "we", "us") and the person or organization that uses InferOptimal ("you"). They cover the InferOptimal console, dedicated endpoints, the console API, the MCP server and the command-line tools.

1. Who can use InferOptimal

Business and commercial use is allowed. If you accept these terms for an organization, you confirm you are authorized to bind it, and "you" means that organization. You must be at least 18. A separate agreement signed by both parties controls where it conflicts with these terms.

2. The service

InferOptimal runs dedicated, OpenAI-compatible endpoints for open models, on GPUs we rent from cloud providers or in a cloud account you connect. We may tune an endpoint's serving settings to run your workload faster or at lower cost. We change a running endpoint only with your consent, and a change never swaps the model you chose.

3. Your content and your data

• You own your content. Your content means the prompts, inputs, files, model weights and adapters you send or connect, and the outputs your endpoints return. You give us only the rights needed to host, process and transmit it to provide the service to you.
• We do not train on your content. We do not use your content to train, fine-tune or evaluate models, ours or anyone else's, and we do not sell it.
• Traffic you send for a comparison. If you send us requests to replay for a side-by-side comparison, we use them only for that comparison and delete our copies within 30 days of reporting the result, unless you ask us in writing to keep them.
• Usage data. We record numerical data about your endpoints, such as token counts, request timings, throughput and hardware utilization. We use it to run, bill, secure and improve the service, including to choose better serving settings. It does not contain your content.
• Who processes it. Our Privacy Policy and the data handling page list where requests go and what is kept. Some endpoints, such as GLM-5.3-Flash, can answer through the model maker's API (for GLM, Z.ai) while your GPUs start. That startup switch is shown before launch and is on by default for those endpoints; turn it off and your requests only ever reach your own GPUs.

4. Accounts and keys

Keep your account and keys secure, and limit them to people and systems you authorize. You are responsible for activity under your account and keys. If a key is lost or exposed, rotate it on the endpoint page and tell us at team@velobyte.ai if you suspect misuse.

5. Endpoints in your cloud account

When you connect a cloud account, you authorize InferOptimal to create, run and remove the resources needed for the endpoints you launch there. Your cloud provider bills you for those resources under its own terms. Stopping an endpoint asks us to release them; you remain responsible for resources in your account.

6. Model licenses

Each open model has its own license and use policy. You are responsible for following the license of every model you run, including any you bring.

7. Acceptable use

Do not use InferOptimal to break the law, infringe others' rights, create content that sexually exploits minors, attack or probe systems without permission, distribute malware, evade our limits or billing, or resell access in a way that hides the end user from us when we ask. We may suspend an endpoint or account that does, and we will tell you why unless the law or a security risk prevents it.

8. Prices, credit and billing

• Prices are shown on each endpoint in the console before you launch it. The price shown when you launch applies to that endpoint until you stop it.
• Endpoints on InferOptimal hardware are paid from your prepaid credit. Billing starts at the point shown before you launch and runs by the minute until you stop the endpoint, including idle time. Some hardware has a minimum, shown before you launch. Launching holds one hour of the price from your credit until the endpoint stops.
• If your credit runs out, your endpoints stop. Promotional and YC credit has no cash value and cannot be transferred. For a refund of purchased credit you have not used, contact us.
• Prices exclude taxes, which are added where required. Stripe processes payments; we do not store card numbers.

9. Availability and changes

GPU capacity depends on our providers, so a launch can fail or wait. A launch that fails because of us is not charged. The service is provided without an uptime commitment unless a signed agreement includes one. We may change or retire endpoints and prices for future launches, and we will give 30 days' notice before ending an endpoint type you are running.

10. Ending the agreement

You can stop using InferOptimal and ask us to close your account at any time. We may suspend or end access for a material breach of these terms, for nonpayment, or where required by law. Charges already incurred stay due.

11. Warranties and liability

To the extent the law allows, the service is provided as is, without implied warranties of merchantability, fitness for a particular purpose or noninfringement. Model outputs can be wrong; check them before you rely on them. To the extent the law allows, neither party is liable for indirect or consequential loss, and our total liability under these terms is limited to the greater of USD 100 and the fees you paid us in the 12 months before the claim. These limits do not apply to liability that cannot lawfully be limited.

12. Changes to these terms

We will email account owners and show a notice in the console at least 30 days before a material change takes effect. Continuing to use the service after that date means you accept the change.

13. Contact

Velobyte, Inc. (d/b/a InferOptimal), team@velobyte.ai.