Security & Privacy
We do not keep your audio.
Voice is among the most sensitive data a company can handle. The simplest way to protect a recording is to never make one — so we don't.
01
What we store, precisely
It is worth being exact about this, because “we take security seriously” is not a control. Here is the complete list of what survives a request:
02
Biometric data, handled as biometric data
A voiceprint is biometric data under GDPR, BIPA and CCPA, and a careless implementation puts its users in legal jeopardy. The Speaker Profile API therefore records an explicit consent_obtained flag on every profile, and supports a retention_days window enforced by an automatic purge — deletion happens whether or not anyone remembers to trigger it.
Deletion is real deletion: a deleted profile's embedding is removed from the database, not flagged as hidden.
This gives you the mechanisms to comply. It does not, on its own, make you compliant — consent capture and lawful basis remain yours to get right.
03
Controls
Audio is never written to disk
Streams and uploads are processed in memory and discarded. There is no recording bucket, because there are no recordings.
Voiceprints are not audio
A profile is a 192-number embedding. It cannot be played back or reconstructed into speech — it is a fingerprint, not a copy.
Per-tenant isolation
Every profile, key, job and usage record is scoped to a tenant. A request carrying the wrong tenant's id gets a 404, not a peek.
TLS 1.3 in transit
All API and WebSocket traffic is encrypted in transit. Keys are transmitted as bearer tokens, never in a query string.
Hashed API keys
Keys are stored hashed. The full value is shown exactly once, at creation — we cannot recover it for you, and neither can an attacker with database access.
Revocable at any moment
Any key can be revoked from the console and stops working immediately, on every server.
04
Operational posture
05
Certifications
VSIP is an early-stage product and is not yet SOC 2 or HIPAA certified. We would rather tell you that plainly than imply otherwise. The architecture — no audio retention, no transcripts, per-tenant isolation — is designed to make those audits tractable, and to make the data-handling questionnaire short in the meantime.
If you have a security review to run, get in touch and we will work through it with you directly.