An honest, structured snapshot of how PARIE handles customer data, what controls are in place today, what is in flight, and what the roadmap looks like for the rest of 2026. Designed to be the first answer to your security questionnaire.
PARIE is a multi-tenant SaaS that delivers AI-grounded guidance and training inside enterprise web applications (Oracle, SAP, Salesforce, Workday, NetSuite, Dynamics, EPM tools, banking platforms). It runs on Google Cloud (Firebase) with Anthropic Claude as the inference layer. Founded in 2025; USPTO patent application #64/014,523 covers the underlying ensemble-reasoning method.
The product is in a phase commonly described as "design-partner stage" — operational systems are live, the architecture has been built deliberately for the multi-tenant security model that SOC 2 Type I requires, and 148+ automated tests run on every release. SOC 2 Type I certification is on track for Q4 2026. This document tells you exactly what is and isn't true today, in plain language, with no marketing.
What you can rely on today: multi-tenant rule-layer isolation, encryption at rest and in transit, append-only audit logs, brute-force lockout, forced-password-change on suspicious login, GDPR/CCPA-aligned data handling, a real DPA, a public subprocessor list, a real privacy policy, and the legal terms of a B2B SaaS (Delaware governing law, JAMS arbitration, capped liability).
What is on a customer-driven roadmap (not yet attested): SOC 2 Type I third-party audit, HIPAA BAA execution with upstream subprocessors, third-party penetration test, EU/APAC data residency. PARIE has self-collected evidence and template documents ready for each — these are engaged when a customer's procurement gate requires them. Until then, PARIE relies on its in-house tooling and the security architecture documented below.
The principle: collect the minimum we need to deliver the service, store it where the customer chose, and never train models on customer content.
| Data category | Source | Stored where | Retention |
|---|---|---|---|
| Account data (email, name, role, tenant ID) | Customer admin or self-signup | Firestore (Google Cloud, US default) | Lifetime of account + 30 days post-deletion |
| Authentication credentials (Firebase Auth tokens, hashed passwords) | Sign-in flow | Firebase Auth (Google-managed) | Until rotation / account deletion |
| Knowledge-base documents (policies, SOPs that the customer chooses to upload) | Customer admin upload | Firebase Storage; metadata in Firestore | Lifetime of account; customer can delete anytime |
| AI prompts & responses (the questions users ask + Claude's answer) | End-user copilot interaction | Anthropic API (no retention per Anthropic API terms); audit log entry in Firestore (prompt-truncated, response not stored by default) | 30 days for audit log; 0 days at Anthropic |
| Page-context labels (UI labels — button text, headings, breadcrumbs from the active tab) | Chrome extension content script | Sent to Anthropic with the prompt; not persisted | 0 days |
| Usage metrics (call count, token count, feature, latency, tier) | claudeProxy Cloud Function | Firestore usage/ + usageLog/ | 13 months (rolling) |
| Audit log (admin actions, login events, privilege changes) | Server-side handlers | Firestore adminLogs/ (append-only) | Indefinite (immutable) |
PARIE is built on Google Cloud (Firebase + Cloud Functions + Cloud Storage). Every customer is isolated at the rule layer, not just the application layer — meaning a misbehaving client request that escapes our application code still cannot read another tenant's data, because Firestore Security Rules enforce isolation at the storage layer itself.
/clients/{tenantId}/... with rules that require request.auth.token.tenantId == tenantId for every read/write.Every release runs a security test suite that tries to: read another tenant's documents, write to another tenant's audit log, escalate privilege via direct Firestore writes, and bypass tier-gating. All must fail. The CI pipeline rejects any PR that breaks isolation.
failedLogins/{emailKey} with rolling attempts array.mustChangePassword flag forces a reset on next login.staff, manager, hr_admin, client_admin, super_admin. Role-creation matrix enforces who can create whom (a manager cannot create an admin; an HR admin cannot create a super admin; etc.).adminLogs entry; logs are immutable and queryable on customer request.secrets: binding.adminLogs/ or per-tenant clients/{id}/auditLog/.update and delete on audit-log collections — only create is allowed, and the create rule requires the actor's UID match the entry's uid field. A successful audit-log entry cannot be retroactively altered, even by a super admin.parie.io and the claudeProxy endpoint. Service credits per the Master Subscription Agreement (10% credit for <99.9%, 25% for <99.0%).us-central1 with auto-failover within the region.dailyFirestoreBackup Cloud Function, exported to a GCS bucket with 90-day retention. Restoration tested quarterly.The complete, current list lives at parie.io/subprocessors. New subprocessors are announced 30 days in advance and customers may object per DPA §4.2.
| Subprocessor | Purpose | Data exposed | Certifications |
|---|---|---|---|
| Google Cloud / Firebase | Hosting, database, auth, storage, functions | All customer data (encrypted) | SOC 2, ISO 27001, ISO 27018, FedRAMP High |
| Anthropic PBC | LLM inference | Prompts + page UI labels per call (no retention) | SOC 2 Type II, HIPAA-capable under BAA |
| Stripe | Subscription billing | Billing metadata only; no card data on PARIE | PCI DSS L1, SOC 1, SOC 2 |
| Postmark | Transactional email | Email addresses + content of system emails | SOC 2 Type II |
| Namecheap | Domain + DNS | Public DNS records only | Domain registrar |
| GitHub | Source control | PARIE source code (no customer data) | SOC 2, ISO 27001, FedRAMP |
| Standard | Status | Target | Notes |
|---|---|---|---|
| GDPR + CCPA alignment | Live | — | Privacy policy, DPA, SCC clauses, data subject rights handlers all live |
| Multi-tenant rule-layer isolation | Live | — | Firestore rules + automated security tests on every release |
| Encryption at rest + in transit | Live | — | Inherited from Google Cloud; AES-256 / TLS 1.2+ |
| Append-only audit logs | Live | — | Firestore rules forbid updates/deletes on audit collections |
| SOC 2 evidence collection | Live (in-house) | — | Daily evidence packets emitted by the soc2EvidenceCollector Cloud Function — covers CC6.1, CC6.6, CC7.2, A1.2, CC2.2, CC9.1. Available to customers' auditors under NDA. |
| SOC 2 Type I third-party audit | Roadmap | Customer-driven | Engaged when the first customer's procurement gate requires it. PARIE has self-collected evidence ready to hand to an auditor on day 1. |
| HIPAA BAA | Template ready | Q3 2026 | PARIE BAA template published at /baa. Anthropic + Google Cloud BAAs requested before any PHI is processed; status confirmed in writing per BAA Exhibit B. |
| ISO 27001 | Customer-driven | 2027+ | Engaged alongside SOC 2 Type II if a customer requires both. |
| Penetration test | Customer-driven | Before first regulated deal | Will be commissioned when a customer's procurement gate requires a current report; output shared under NDA. In the interim PARIE runs ESLint security rules and Dependabot on every release. |
If your procurement gate requires SOC 2 Type I, HIPAA BAA, or ISO 27001 on day 1: we engage with your timeline. PARIE will accelerate the relevant track in exchange for a contracted commitment, and where appropriate, partner with a SOC 2-certified channel partner (e.g., Donyati's compliance umbrella) to bridge the gap during certification. Talk to us before assuming the gap is a deal-breaker.
SECURITY_INCIDENT for first response within 4 business hours; PGP key on request.SECURITY_VULNERABILITY. We acknowledge within 2 business days, fix within the SLA above, and credit the reporter unless they prefer otherwise.SECURITY_INCIDENT — first response in 4 business hoursSECURITY_VULNERABILITY — acknowledged in 2 business daysDATA_SUBJECT_REQUEST — 30-day turnaroundCOMPLIANCE — for security questionnaires, attestation packets, custom DPA negotiation