Building a Privacy-First AI Platform: Sevrel's Architecture
CRE documents contain confidential rent figures, deal terms under negotiation, and proprietary valuations. Sevrel combines qualified provider data-use terms, organization-scoped authorization, database row-level security, and instrumented audit events. These controls reduce risk; they are not a promise that software can never fail or that every access is represented in one audit stream.
The Architecture at a Glance
Sevrel's privacy architecture has three key components:
Enterprise AI Provider (Tiered Routing)
Sevrel uses an enterprise AI provider with tiered model routing (fast, standard, and deep tiers). Inference requests use TLS. Current public disclosures describe the applicable provider's paid-API training and retention posture, including up to 30-day abuse-monitoring retention and no Sevrel Zero Data Retention agreement. Verify current vendor terms and any executed customer agreement before relying on them.
Per-Organization Scoping
Customer-content paths carry organization context, and PostgreSQL Row-Level Security policies add a database enforcement layer. CI exercises cross-organization isolation; current production role and policy state still requires live operational evidence.
Application & API Layer
The web application and API run on managed infrastructure with DDoS protection, TLS encryption, and edge caching. Session cookies are HttpOnly, Secure, and SameSite=Lax. Server-side integration secrets are not intentionally exposed to browser code; browser authentication libraries still handle the user tokens needed for their documented sign-in flows.
Privacy Design Principles
- No secrets in the frontend. API keys, tokens, and credentials exist only server-side. The browser never sees them.
- Protected session cookie. The Sevrel session cookie is HttpOnly, Secure in production, and SameSite=Lax. Browser OAuth libraries maintain their own documented transient token/cache state.
- Encrypted credentials at rest. OAuth tokens for Egnyte and Microsoft Graph are encrypted with symmetric keys before database storage.
- Per-organization isolation. Organization-scoped tables use PostgreSQL Row-Level Security and application-layer tenant checks, with automated cross-organization tests.
- Append-only audit log. Selected security and product events are recorded with correlation context; event coverage varies by path.
Tiered Model Routing
No single model is best at everything. Sevrel automatically routes each query across three tiers based on complexity:
Fast
Fastest responses. Used for titles, intent classification, and simple lookups.
Standard
Best balance of speed and depth. The default for RAG queries, agent workflows, and general chat.
Deep
Deepest reasoning. Reserved for deal underwriting, complex legal analysis, and systematic research.
What This Means in Practice
For a CRE organization evaluating AI tools, Sevrel's architecture answers the questions that IT security and legal teams care about most:
- “Where does our data go?” — TLS-encrypted to our enterprise AI provider (Anthropic). Sevrel's current disclosure says standard abuse-monitoring retention can last up to 30 days and no Zero Data Retention agreement is in place; confirm current vendor terms and the executed customer agreement
- “Is our data used for training?” — Current public provider terms for paid API data are described in Sevrel's privacy disclosure; there is no asserted Sevrel-specific contractual no-training guarantee
- “Who has access?” — Organization roles govern ordinary access; audited internal privileged-support paths also exist, and PostgreSQL RLS adds a tenant-control layer
- “Can we audit usage?” — Instrumented security and product events write to an append-only database design with correlation context; event coverage varies by path