Security · Zero data retention
Your data is processed, never possessed
Zero data retention isn't a setting you enable or a tier you upgrade to — it's the only way HyperInfer runs. Every prompt and completion exists solely in volatile GPU memory for the lifetime of a single request, then it's gone.
The promise
Four guarantees, written into your DPA
Never logged
Prompt and completion bodies are never written to application logs, request traces, or any persistent store. Observability operates on metadata only.
Never stored
No content is buffered, cached, or retained after a response completes. There is no message history on our side to leak, subpoena, or breach.
Never trained on
Your inputs and outputs are never used to train, fine-tune, evaluate, or improve any model — ours or any third party's. No exceptions, no opt-out needed.
Never leaves your region
Inference is pinned to your chosen data-resident region. Content never transits another region or a third-party sub-processor.
The data lifecycle
Born and gone inside one request
Follow a single token from the moment it arrives to the moment its memory is overwritten. Nothing in this path touches a disk.
- 01
Received
Authenticated over TLS 1.3 and routed only within your pinned region.
- 02
In memory
Tokens decrypted into volatile GPU memory. Never written to disk.
- 03
Computed
Inference runs on your dedicated, single-tenant accelerators.
- 04
Streamed
Output returned token-by-token straight back to your client.
- 05
Purged
GPU memory is overwritten on completion. Nothing persists.
Total persisted footprint after completion: 0 bytes.
Full transparency
What we keep vs. what we never touch
Operating a service requires some metadata — but never your content. Here's the exact line.
Operational metadata we keep
- Request timestamp and latency
- Token counts for billing
- Model ID and region
- HTTP status / error class
Content we never touch
- Prompt text or messages
- Completion / response content
- Uploaded files or attachments
Don't take our word for it
Verify it yourself
Response headers
Every response returns an X-Request-Id you can trace end to end in your own logs and tests — paired with an architecture that has no persistent write path to inspect.
Contractual DPA
Our Data Processing Agreement codifies zero retention as a binding obligation, not a best-effort policy.
Independent audit
SOC 2 Type II controls covering retention are tested annually by a third-party auditor; the report is available under NDA.