Inference Data Handling — what Netra means by ZDR
Effective and last updated: October 20, 2026 · Version 2026-10-20
This policy forms part of the Terms of Service.
Netra uses "ZDR" in the sense described by OpenRouter's interpretation of in-memory caching. This means no default persistent content logging or training use, while allowing the memory caching described here. It does not mean that all data disappears immediately after a request. This reference does not imply OpenRouter certification or verification of Netra.
How native inference caching works
- Derived KV/prefix-cache state is held in RAM/GPU memory, without disk/NVMe or remote-storage offload.
- LMCache and GPU cache are encrypted.
- Cache entries can remain for up to seven days from creation. Access or reuse does not extend their expiry.
- Cache entries are shared across customers; they are not isolated per workspace.
Your caching choice
You can request that shared caching be disabled for your workspace. Once Netra confirms activation, your workspace will neither reuse existing shared entries nor add new entries to the shared cache.
Entries created before activation remain until their original seven-day expiry. Opting out does not immediately delete them. Disabling optional content recording is a separate setting.
These cache details describe Netra's native inference infrastructure. Netra requires zero-data-retention processing on supported external inference routes. The native seven-day cache limit and workspace cache opt-out described here do not describe or control an external provider's in-memory cache. ZDR does not guarantee that processing remains in Indonesia. External agents, applications and tools may also retain their own copies under their own policies.
Optional saved content follows the separate 30-day rule in our Privacy Policy.
Scope and relationship to the Terms
This policy covers Netra's V4.1 Flash Chat Completions, Responses and Messages APIs, temporary Playground and Free OCR. Free OCR coverage includes submitted images/documents, extracted text and results. Other models and tools require their own stated coverage. The native cache description above must not be assumed to describe an external provider.
This policy and the inference-data section of the Privacy Policy control conflicting general permissions concerning inference content in the Terms. Saving content, requesting support or sending feedback does not permit Netra to train or fine-tune models, conduct its own content-based evaluations, advertise, or improve models or services using that content. Content-free operational measurements are separate. Mandatory law still applies.
For processing details, cache opt-out or deletion requests, contact rifky@netraruntime.com, with farrel@netraruntime.com as backup. The Bahasa Indonesia version is controlling for Indonesian users.