1. Scope and Acceptance
These API Terms govern all programmatic access via api.mindrift.art: chat completions, image generation (sync and async tasks), jobs, model catalog, and account endpoints. They supplement the Terms of Service, Privacy Policy, Data Privacy Notice, and Community Guidelines, all of which apply in full to API usage. On any conflict specific to programmatic access, these API Terms prevail. Calling any endpoint constitutes acceptance. The machine-readable reference lives at api.mindrift.art/v1/openapi.json; in case of divergence between documentation prose and enforced behavior, enforced behavior governs and documentation is corrected promptly.
2. API Keys: Issuance, Custody, and Scope
- Issuance: create keys in console, API Keys. Each key carries granular scopes (text, images), an optional IP allowlist, an optional expiry, and an optional monthly budget cap. Default-deny: a scope not granted is refused with 403.
- Custody: keys are bearer secrets equivalent to passwords. Store them server-side or in secret managers only. Never ship keys in client-side code, mobile bundles, public repositories, or logs. Rotate immediately on suspected exposure and revoke unused keys.
- Attribution: all calls under a key are attributed to the issuing account, including spend, quotas, and violations, regardless of which system or subcontractor physically sent them. You are responsible for your integrations and customers.
- Transfer ban: sharing, reselling, renting, or sublicensing API access or keys is prohibited without prior written consent.
3. Quotas, Fair Use, and Safety Enforcement
- Rate limits: per-minute request quotas per category and concurrent-task caps per tier. Breaches return HTTP 429 with Retry-After and quota headers. Implement exponential backoff; aggressive retry storms are treated as abuse.
- Polling discipline: poll tasks at the documented cadence (about every 2 seconds with backoff); set a client-side timeout (90 seconds recommended) and cancel stale tasks via the cancel endpoint.
- Prohibited automation: no scraping of the catalog or other users' assets, no quota evasion (key rotation, account multiplication), no DDoS or resource-exhaustion patterns, no reverse engineering of safety systems.
- Moderation supremacy: blocked tasks are final, unbilled, and non-retriable. Resubmitting blocked content, paraphrased or not, to evade filters is a standalone violation leading to key suspension and account termination. We log moderation events with request identifiers for review and legal reporting.
- Disclosure duties: where applicable law requires labeling synthetic media or disclosing AI involvement (advertising, news, elections, likeness of real persons), compliance is entirely your responsibility as the publisher.
4. Metering and Billing
Chat meters per-token at the model's published input, output, and cached rates. Images bill per completed generation from base price times multiplier. Catalog figures are estimates: the definitive amount is computed at completion, may vary with GPU load and parameters, and is recorded per request with tokens, model, and cost in your request history. Failed technical tasks are retriable at no double charge for the failed attempt; only completed work debits the wallet. Top-ups are non-refundable per the Terms. Taxes, if due, are additional to displayed prices unless stated otherwise.
5. Compatibility, Versioning, and Support
Chat and sync image endpoints follow OpenAI-style shapes; the async task system (submit, poll, cancel, retry) is documented in the OpenAPI reference. We version breaking changes and announce removals at least 30 days in advance through docs and console notice; additive changes may ship anytime. No uptime SLA applies unless agreed in a separate written enterprise agreement. Support runs through the in-console contact flow or [email protected]: always include the request_id of failing calls (returned only for accepted, stored requests) so we can trace them.
6. Outputs, Confidentiality, and Data
Outputs assigned under the Terms belong to your account subject to compliance; the uniqueness and registrability limits of Section 5 of the Terms apply equally to API outputs. Your prompts and outputs are processed per the Privacy Policy and Data Privacy Notice, never sold, and never used to advertise to you. Request logs are retained for metering, debugging, and abuse prevention, then purged per retention rules.
7. Suspension, Termination, and Liability
We may throttle, suspend keys, or terminate API access for quota abuse, non-payment, safety violations, or legal compulsion, with or without notice proportionate to severity. Termination for cause forfeits remaining Balance per the Terms. You may stop calling and revoke keys anytime. Liability for API services falls under Sections 10 and 11 of the Terms (as-is service, $100 or 3-months-paid cap), and your indemnification duties extend to claims arising from your integrations, customers, and published outputs.