This page says what Sayli's security posture is today. It also says what we do
not have yet, because a security page you cannot trust is worthless.
Encryption in transit
All traffic runs over TLS, including audio streaming.
Encryption at rest
All stored data is encrypted at rest with our cloud provider's default envelope encryption.
Tenant isolation
Every query that touches user data is scoped to your organization in the data model itself. Vector search carries the same per-organization filter.
Redaction at ingest
Card numbers, government identifiers, phone numbers, email addresses, and IP addresses are masked before transcript text is stored or sent to a language model.
Object access
Recordings and files live in a private bucket. Access is through short-lived signed URLs minted per request, never public links.
Access ends with membership
Removing someone from an organization closes their live sessions immediately, and stops their export and erasure at the workspaces they still belong to. A call recorded for a workspace that has since removed them stays that workspace's record.
Data portability
Every user can export everything they own as one JSON file, from Settings, without asking us.
Authentication
Sign in with Google, with Apple, or with a six-digit code emailed to the address. There is no password on a user account, so there is no credential to leak, reuse or reset, and none travels by URL. A code is single use, expires in ten minutes, and burns after five wrong attempts. Sessions use short-lived access tokens with rotating refresh tokens. The desktop app keeps its session in the operating system's credential store, and on Linux refuses to persist it at all when no real keyring is present.
Audit log
State-changing actions are recorded per organization: who did what, to which resource, when.
Telemetry hygiene
Error reports are scrubbed of transcript text, questions, answers, and tokens before they leave our systems. Analytics events carry feature usage, never call content.
Abuse limits
Per-account and per-IP rate limits bound every endpoint, tightest on unauthenticated ones.
We would rather tell you now than have you find out in a vendor review.
- No SOC 2 report. A SOC 2 Type II program is under way and no report has been
issued. We claim no certifications of any kind today, and we will not until
one is signed.
- No HIPAA business associate agreements exist with us or our vendors. Sayli
is not suitable for PHI today. A dedicated compliant inference path is on
the enterprise roadmap.
- No SAML single sign-on and no SCIM provisioning. Both are on the enterprise
roadmap.
- No self-hosted deployment.
- No bug bounty program yet. Reports are still welcome, see below.
A small set of third parties touches customer data: speech-to-text,
language-model inference, embeddings, vector search, web search, cloud
infrastructure, email delivery, billing, analytics, and error reporting. Each sees only what
its job requires. What each one's terms allow, including anything about model
training, is part of what goes out under the mutual NDA below, because it is
their document rather than ours to summarise.
We do not publish the named list. Vendor choices are part of how Sayli is
built, and we would rather not hand that to a competitor. Customers and
prospects in a security review get the full list, with regions and what each
one processes, under a mutual NDA. Ask at
security@sayli.ai.
Live status and incident history are at
status.sayli.ai.
Email security@sayli.ai. Include steps to
reproduce. We will acknowledge your report and keep you informed while we fix
it. Machine-readable details live at
/.well-known/security.txt.