Policy Scope
This Policy governs data handling for Token API and Token Factory. The Enterprise Terms of Service and Enterprise Privacy Policy are complementary and should be read with this Policy.
Token API Data Path
Token API receives Inference Content transiently, processes it to perform the requested inference, and returns the model output to the customer. Inference Content is not persisted anywhere along this serving path.
Zero Data Retention
TokenWay applies Zero Data Retention to Token API Inference Content. Prompts, messages, attachments, other inputs, and model outputs are not logged or persisted and have a retention period of 0 days.
Inference Content is not used to train, fine-tune, distill, evaluate, benchmark, or improve models. TokenWay personnel do not review Inference Content. Customer content voluntarily submitted through an authorized support channel is handled separately for the support purpose.
Ephemeral KV Cache
The Token API serving path may use GPU memory and distributed volatile memory to cache KV tensors solely for inference acceleration. Eligible cache keys may reuse KV tensors while an entry remains live.
Cache entries are lifecycle- and capacity-bound, isolated by tenant, model, and model configuration, are not backed up or persistently stored, and do not survive relevant service or infrastructure restarts.
A retention period of 0 days means TokenWay creates no persistent record of Inference Content. Temporary KV tensors are bounded volatile processing state, not logs or persistent storage, and are removed by lifecycle or capacity eviction or a relevant service or infrastructure restart.
Non-Content Metadata
TokenWay may retain non-content service metadata, including request time, model, token counts, status, latency, error category, request ID, and billing account, for up to 180 days for metering, billing, reliability, security, and compliance. This metadata does not include the text or substance of Inference Content, including prompts, messages, attachments, other inputs, or model outputs.
Access and Operational Controls
TokenWay personnel are not authorized to access or review Inference Content, including for routine troubleshooting. Least-privilege access, environment separation, and credential controls enforce this restriction.
Token Factory Data Boundary
For Token Factory, logging, backups, access, and deletion are determined by the customer-approved deployment boundary and the applicable project agreement. This Token Factory data boundary does not weaken the Zero Data Retention requirements for Token API Inference Content.
Service Providers
TokenWay may engage service providers in the compute, data-center, network, and security categories. They may process data only for the approved service purpose and must follow TokenWay's no-training and content-retention requirements for Token API.
Security and Incidents
TokenWay uses reasonable technical and organizational measures designed to protect service data and maintains incident-response procedures. TokenWay provides incident notice as required by the applicable agreement and applicable law. No method of transmission or processing can be guaranteed to be completely secure.
Customer Responsibilities
Customers must have the rights, permissions, notices, and consents needed to submit content; protect credentials from unauthorized use; and review model outputs before use. Customers should not submit Inference Content through a support channel unless they intend it to be handled as support content.
Changes and Contact
TokenWay may revise this Policy by publishing an updated version with a revised effective date. Changes apply prospectively unless applicable law requires otherwise or an applicable agreement states otherwise.
For data or privacy questions, contact privacy@tokenway.ai. Send legal notices to legal@tokenway.ai.