Privacy policy

Prompt and completion data

KF Inference processes prompts and completions only to provide the requested inference response. We do not use API content to train models and do not intentionally persist prompt or completion bodies in application or proxy logs.

Operational metadata

We may retain limited operational metadata such as timestamp, request identifier, requested path, HTTP status and latency. This data is used for security, reliability and usage reconciliation. Dedicated edge logs are rotated daily and retained for no more than 14 days.

Infrastructure and subprocessors

Inference is performed on privately operated NVIDIA DGX Spark systems in China. Public HTTPS termination is provided by an Alibaba Cloud server in China. Network traffic between the edge and inference system uses an encrypted private tunnel.

Security and retention

API access requires authentication. Inference ports are not exposed directly to the public internet. Content may exist transiently in memory while a request is processed and is released as the runtime reuses memory.

Requests

Privacy requests may be sent through the KF Inference provider contact supplied to the API marketplace. A dedicated domain email will be published before general availability.