Zero Data Retention
Prompts, audio, and generated output are processed in-memory and discarded the instant a request completes. No prompt content is ever written to disk.
Zero Data Retention · No payload logging
Security & privacy, by architecture.
Prompt Gateway is built so your data can't leak — because it's never stored. Prompts pass through, get processed in memory, and are gone. There's nothing to breach, subpoena, or sell.
How the architecture protects you
No payload logging
Our logs record request metadata only — timestamps, status, latency. The payload itself is never logged, not even for debugging or model training.
Encrypted in transit
TLS 1.3 everywhere. With Prompt Guard, payloads are additionally redacted and encrypted before they ever reach a model provider.
The lifecycle of a request
In memory, then gone.
1 · Receive
Over TLS 1.3
Your prompt arrives encrypted and stays in volatile memory.
2 · Protect
Redact & moderate
Prompt Guard strips PII and applies policy before anything leaves.
3 · Process
Enhance
The result is generated and streamed back to you directly.
4 · Discard
Nothing persists
Memory is released. No copy is kept — there is nothing to delete.
What we store and what we never store
What we never store
- Never stored: Your prompts or the text you type
- Never stored: Voice audio or transcripts
- Never stored: Generated prompts
- Never stored: The pages or apps you use it on
- Never stored: Anything used to train models
What we do store
- Stored: Your account email and profile
- Stored: Billing details (handled by Stripe)
- Stored: One optional, encrypted sign-in token
- Stored: Aggregate usage counts — never content
- Stored: Request metadata (time, status, latency)
- Stored: Support screenshots you choose to attach
Governance and failure modes
Admin-governed policy
- Central redaction rules and allow-lists (Prompt Guard)
- SSO / SAML & SCIM provisioning
- Per-action enable/disable — disabled reads as policy, never a bug
- Custom data residency (Enterprise)
Honest failure modes
When a safety service is unavailable, you get a clear "try again" — never a false content verdict. An outage never masquerades as a block, and a block is never silently skipped.
“Safety check unavailable — your prompt was not sent. Try again.”
Bring it to your security team
We'll walk through the architecture and sign your DPA.