Inference API terms

Version: October 3, 2026 · Applies to Inferoptimal Models

These terms govern the inference API supplied by Inferoptimal, Inc. ("Inferoptimal," "we," or "us"). They apply to direct customers and, subject to a separate agreement, distribution partners. They do not replace separate terms for Evo or our private agent mailbox.

1. Access and agreement

You must be at least 18 and authorized to bind the organization for which you use the API. Access requires an approved account, an API key, and acceptance of these terms and our privacy and data policy. Protect your keys, limit their use to authorized people and systems, and report suspected compromise to team@velobyte.ai. You are responsible for activity using your keys unless applicable law provides otherwise.

2. Subsidized tier and data use

The subsidized tier requires permission for Inferoptimal to collect prompts, messages, tool inputs and results you submit, and model completions, including reasoning returned by the upstream model. We may use this content to operate and secure the service, evaluate model quality, and train or improve models and systems, only where applicable law and model/provider terms permit.

You retain your rights in submitted content. You grant Inferoptimal a nonexclusive, worldwide, royalty-free license to process that content and generated responses for those purposes, including through service providers bound to appropriate safeguards. This permission does not transfer ownership of your content or authorize public publication of your raw conversations.

Prompt and completion records expire from our content store after 30 days. The same deadline applies to identifiable copies selected for evaluation or training. Trained model weights and aggregate results may remain after source records expire; deleting a source record does not necessarily undo training already performed. Our privacy policy explains separate billing and security records.

This tier is not a confidential, zero-data-retention, or no-training service. Do not submit passwords, access keys, payment-card data, regulated health information, or other sensitive information. Submit third-party or employer content only with the necessary rights and notices. If applicable law requires a separate consent or processing agreement that is not in place, do not submit the affected personal data.

3. Upstream models

Requests are processed through Z.ai using the model identified in the request, subject to our applicable provider agreement. See Z.ai’s terms and privacy policy. Applicable model licenses, upstream restrictions, and acceptable-use requirements also apply. We do not claim to own or operate the upstream models. Availability, limits, features, and rates are listed in our model catalog. A model marked unavailable cannot receive production requests. We may suspend access when funding, provider permissions, or service capacity are unavailable.

4. Charges and credits

Prices are stated in USD per million tokens. Uncached input, cached input, and output are metered separately using upstream usage counts; output may include billable reasoning tokens. The rate recorded when a request is admitted applies to that request. Estimates are authorization holds, not final charges. A request with missing or invalid usage is held for reconciliation and is not assigned an invented final charge.

Direct prepaid accounts can use only purchased or expressly granted credit. Credits are specific to this inference service and are not Evo credits. Payment confirmation comes from our payment processor, not the browser return page. Credit does not expire while the account remains open. Contact us to request a refund of unused purchased credit; statutory refund rights remain unaffected. Promotional credit has no cash value. Partner invoicing requires an agreed account limit and payment terms. No automatic card reload is enabled by these terms.

Subsidies may end or rates may change for future requests. We will display applicable rates before charging them. We may refuse new requests when account credit or the service budget is exhausted. Where required, taxes are charged separately and shown before payment. You remain responsible for valid accrued charges.

5. Acceptable use and outputs

Do not use the service unlawfully, abuse another person's rights, attempt to obtain other customers' content or credentials, bypass usage controls, or disrupt infrastructure. You must evaluate outputs before relying on them; they may be inaccurate, nonunique, or unsuitable. As between you and Inferoptimal, we do not claim ownership of your outputs, subject to applicable model licenses and third-party rights. No output is professional advice or a guarantee of correctness.

6. Suspension, deletion, and changes

You may stop using the API at any time and request key revocation, account closure, and deletion by emailing team@velobyte.ai. We may suspend access for abuse, security incidents, nonpayment, or upstream unavailability. Material changes to these terms will receive a new version and require renewed acceptance before new inference requests. Accrued payment obligations and lawful record-retention duties survive closure.

7. Service limitations and responsibility

To the extent permitted by law, the service is provided as available without implied warranties of merchantability, fitness for a particular purpose, or noninfringement. Unless a separate written agreement states otherwise, no uptime commitment applies. To the extent permitted by law, neither party is liable for indirect or consequential loss; Inferoptimal's aggregate liability arising from the service is limited to the greater of USD 100 or fees you paid for it in the preceding 12 months. These limits do not exclude liability that cannot lawfully be excluded, including applicable liability for fraud or willful misconduct, and do not limit mandatory consumer rights.

8. Contact

Questions about access, billing, content rights, or these terms: team@velobyte.ai. A separately signed agreement controls where it conflicts with these terms.