Your agents’ conversations, kept apart and accounted for.
Sepal Assure is multi-tenant: many organizations share the software, but every record carries its organization and every query is constrained to it. Dedicated and customer-cloud deployments are available for agencies that need them.
Isolation
- Every table carries the organization, and the data layer refuses any query that isn’t scoped to the caller’s organization.
- Each organization has its own Google Identity Platform tenant, so sign-in, SSO configuration and user pools never mix.
- Files and transcripts are stored under the organization’s own path, and security rules check the organization on every read.
Access
- Single sign-on with Okta or any SAML 2.0 / OpenID Connect provider; multi-factor authentication is enforced by your provider, and roles can follow your directory groups. SCIM 2.0 creates and disables accounts from your directory.
- Roles match AI quality work — administrators, test designers, testers, SME reviewers, AI engineers, governance, release approvers, viewers, auditors and service accounts — and can be customized down to the permission.
- Project scope limits people to their own agents, suites, runs and findings.
- Approval routes skip anyone who submitted the request, no one approves a test they wrote, and segregation-of-duties rules flag conflicting roles.
Personal information and model use
- Production conversations are masked — names, member numbers, Social Security numbers, account and contact details — before they are stored or scored. Unmasked sampling needs a privacy-officer permission and is logged.
- Synthetic test accounts are used for test runs; real member data never needs to enter a test.
- Judge, generation and simulation models run on US endpoints only. Your data is never used to train models, by us or by our model providers.
Audit trail
- Every sign-in, permission change, test edit, approval, run, review decision, gate override, export and reveal is written to an append-only audit log with who, when, from where and what changed.
- The log is sealed daily with a hash chain so tampering is detectable, and can stream to Splunk or Microsoft Sentinel.
- Release decisions keep their full record — evidence, comments, approvals and accepted risks — for auditors and governance committees.
Accessibility
- The app is built to WCAG 2.2 AA, tested with keyboards and screen readers on every release; every chart has a table.
- A current VPAT (ACR) is available, and accessibility issues are fixed as defects, not feature requests.
Infrastructure
- Runs on Google Cloud (Firebase Hosting, Cloud Functions, Cloud SQL for PostgreSQL through Data Connect, Firestore) in US regions, with Dallas as the primary region.
- Controls aligned to the NIST SP 800-53 Moderate baseline used by TX-RAMP Level 2; the underlying Google Cloud services hold TX-RAMP authorization.
- Encryption in transit (TLS 1.2+) and at rest; secrets are held in Secret Manager, never in code or configuration files.
- Point-in-time recovery, daily backups to a second US region, a one-hour recovery point and an eight-hour recovery time, tested yearly.
We’re glad to complete your security questionnaire, the TX-RAMP assessment or the SIG Lite as part of an evaluation.