Inference privacy & data policy

Version: October 3, 2026 · Applies to Inferoptimal Models

Inferoptimal, Inc. operates the inference API at inferoptimal.com/inference/v1. This policy covers that service and its billing interface. Separate products may have separate policies.

What we collect

We process the messages, prompts, tool-related content, generation settings, and model responses needed to fulfill your requests. We also process account identifiers, contact details you provide, hashed API-key identifiers, policy-acceptance records, request IDs, model IDs, token counts, timing and error status, and billing records. Our hosting provider processes ordinary connection information, including IP addresses, to deliver and protect the service. Payment-card details are handled by Stripe; our API does not receive full card numbers.

Why we use it

We use data to fulfill requests, measure usage, manage payments and account access, prevent abuse, provide support, and comply with law. Under the subsidized tier, prompts and completions may also be used for model training, evaluation, and service improvement where permitted. Service delivery and billing rely on our agreement with you; security and administration may rely on legitimate interests; legal compliance and, where required, your consent provide other applicable grounds. Contract acceptance alone does not override privacy rights or make every use of personal data lawful.

Retention

• Prompt and completion records: 30 days from request admission, including content selected for evaluation or training. Expired records are excluded from access and training. We do not create a separate permanent raw-content training archive.
• Usage and security metadata: settled request records are retained for up to 90 days, without prompt or completion text. Unresolved billing holds are kept until reconciled and then removed after 90 days.
• Account, consent, and financial records: kept while your account is active and as needed for accounting, disputes, fraud prevention, and legal obligations; financial records may be retained for up to seven years after the relevant transaction.
• Trained models and aggregates: model weights and aggregate evaluation statistics can outlast individual source records. Source deletion does not guarantee removal of all effects of prior training. We do not treat data as anonymous merely because direct identifiers were removed.

We may preserve records longer where a binding legal obligation requires it, limiting their use to that purpose. We do not promise instant physical erasure from all infrastructure media.

Who processes data

Z.ai processes inference inputs and outputs. Its privacy policy and service terms describe its processing commitments. Cloudflare hosts the API and data stores. Stripe processes payments when prepaid billing is enabled. Authorized Inferoptimal personnel and contracted service providers may access data for the stated purposes. We do not sell raw prompts or completions or publish customer conversations. A corporate transaction or binding legal request may require disclosure subject to applicable safeguards.

The upstream provider’s policy does not remove Inferoptimal's 30-day retention or training permissions. The standard service does not promise a single-country processing location. Only verified regions are listed in the model catalog; cross-border processing remains subject to applicable safeguards.

Your choices and rights

Before using this tier, you must accept the disclosed training use. If you do not agree, do not use this tier. No private or no-training alternative is advertised as available. You may stop future submissions, revoke a key, or request access, correction, deletion, or account closure at team@velobyte.ai. Where applicable, you may object to processing, request portability or restriction, withdraw consent for future processing, and complain to your supervisory authority. We verify your authority before releasing or deleting account data. Withdrawal does not affect earlier lawful processing, and some records may be retained for legal obligations.

Security and sensitive information

We use HTTPS, hashed API keys, restricted administrative access, and time-limited content storage. No system is perfectly secure. Do not submit credentials, payment-card data, regulated health information, children's personal data, or content you do not have authority to share. This service is for adults and does not knowingly collect children's data.

Updates and contact

Material policy changes receive a new version and renewed acceptance before subsequent inference. Contact team@velobyte.ai for privacy questions. These disclosures describe the service when activated; the catalog reports whether inference is currently available.