Security & privacy
Privacy isn't a setting. It's the architecture.
HyperInfer is built so that your data physically cannot be retained. Single-tenant GPUs, in-region processing, and contractual zero retention mean your prompts and completions are never logged, stored, or used to train any model.
The zero-retention promise
Four guarantees, in writing
These aren't toggles you have to find or tiers you have to pay up for. They apply to every request, on every plan, and they're written into your data processing agreement.
Never logged
Prompt and completion bodies are never written to logs, disk, or any persistent store — only volatile GPU memory during the request.
Never stored
No request or response content is retained after a call completes. Default retention is zero, with no buffering or caching of payloads.
Never trained on
Your inputs and outputs are never used to train, fine-tune, or evaluate any model — ours or any third party's.
Never leaves your region
Inference is pinned to your chosen data-resident region. Data does not transit other regions or sub-processors.
Data lifecycle
From request to purge in milliseconds
- 01
Request received
Authenticated over TLS 1.3 and routed only within your pinned region.
- 02
Loaded to GPU
Tokens held in volatile GPU memory. Never written to disk.
- 03
Inference runs
Computed on your dedicated, single-tenant accelerators.
- 04
Response streamed
Output returned token-by-token straight to your client.
- 05
Memory purged
GPU memory cleared on completion. 0 bytes persist. No logs.
Defense in depth
Controls across every layer
Tenancy & compute
- Single-tenant, dedicated GPUs
- No shared inference queues
- Isolated network namespaces
- Optional dedicated VPC peering
Data protection
- TLS 1.3 for all data in transit
- AES-256 for any at-rest artifacts
- Customer-managed keys (BYOK)
- Perfect forward secrecy
Identity & authz
- SSO / SAML & SCIM
- Role-based access control
- Scoped, rotatable API keys
- Just-in-time admin access
Audit & monitoring
- Tamper-evident access logs
- SIEM export (no prompt data)
- Real-time anomaly detection
- Per-key usage attribution
Perimeter
- Private endpoints / PrivateLink
- IP allowlisting
- No public model weights egress
- DDoS protection
Resilience
- Multi-region failover
- 99.99% uptime SLA
- Quarterly pen tests
- Incident response < 30 min
Compliance
Audited and attested
Reports, certificates and our DPA are available under NDA through the trust center. Your auditors are welcome.