Skip to main content

Zero data retention

CoreWeave Serverless Inference for models hosted and served directly by CoreWeave uses a zero data retention (ZDR) architecture by default. Customer prompts and model responses (“Inference Content”) are processed to fulfill inference requests and are not stored or logged to persistent memory during normal service operations. CoreWeave does not use Inference Content to train models. Operational metadata, such as token counts and request latency, is retained to operate and bill the service. In exceptional error conditions, diagnostic logs may contain fragments of Inference Content and are retained for up to 14 days. If a customer enables optional services, such as Agent Lens, that log or store Inference Content, such Inference Content is subject to the retention and data use practices of those services.

CoreWeave Trust Center

For more information about CoreWeave’s security, privacy, and compliance practices, visit the CoreWeave Trust Center.
Last modified on November 19, 2026