← Back to aserva

Security at aserva

aserva handles your customers' conversations and order context. This page describes the controls that protect them — in plain language, without overclaiming.

Tenant isolation and access

Strict multi-tenant isolation

  • Every record is scoped to your organization; the tenant key is always derived from the authenticated session, never from request input.
  • Vector search namespaces, caches, and rate-limit buckets are partitioned per organization.
  • Two-tenant live isolation is verified with automated proof checks, not assumptions.

Authentication and RBAC

  • Dashboard access uses Supabase Auth with JWT session validation and middleware session refresh.
  • Roles: owner, admin, agent, viewer — with granular permission checks on sensitive operations.
  • Public API keys are stored as SHA-256 hashes; the plain key is shown exactly once at creation.

Data protection

Encryption and secrets

  • Third-party integration credentials are encrypted at rest with AES-256-GCM.
  • Secrets are never logged; automated audits scan the repository for leaked credentials and known advisories.
  • Data in transit is TLS-only across app, API, widget, and webhook surfaces.

PII handling

  • AI guardrails scrub sensitive data patterns (card numbers, government IDs, phone, email) before LLM processing.
  • GDPR webhook compliance endpoints are implemented for Shopify (customer data request, redact, shop redact).
  • Customers can request data export and deletion through platform-mandated flows.

AI and action safety

Approval-gated commerce actions

  • Refunds, returns, cancellations, exchanges, and discounts are proposed — never executed autonomously.
  • Widget confirmations require signed, purpose-bound, scoped, short-lived, one-time action tokens.
  • Merchant and platform automatic-action switches both default to OFF.
  • Every action proposal, confirmation, and rejection is written to an immutable audit log.

Webhook and abuse protection

  • All provider webhooks (Shopify, Stripe, email, Facebook, Instagram, and Zid) are signature-verified or authenticated before processing.
  • Idempotency keys and replay protection on delivery-sensitive endpoints.
  • Rate limiting with sliding windows on public AI endpoints and the public API.
  • CSRF protection and strict route audits run in CI on every change.

Operational policies

  • Incident response policy — severity ladder (SEV1-4), the alerting that actually runs, escalation path, postmortems, and status communication.
  • Data retention policy — retention per data category, grounded in the actual data model, including what is and is not deleted automatically today.
  • Subprocessor list — providers the platform actually uses, categorized by role, with integration surfaces listed separately.

Subprocessors

aserva uses the following subprocessors to deliver the service. This list is kept current; material changes are announced before they take effect. The detailed subprocessor page adds purposes, data categories, and integration surfaces.

SubprocessorPurposeData region
SupabaseAuthentication and PostgreSQL database hostingUnited States
VercelApplication hosting and CDNUnited States
UpstashRedis cache, rate limiting, session memoryGlobal (AWS us-east-1)
PineconeKnowledge base vector storageUnited States
Moonshot AI (Kimi)Primary LLM inferenceProvider-managed
OpenAIFallback LLM and image analysisUnited States
ResendOutbound email deliveryUnited States
SentryError monitoring and diagnosticsProvider-managed

Connected commerce platforms (Shopify, Salla, Zid, WooCommerce) process data under their own terms when you link your store.

Compliance status

  • GDPR: platform-mandated data request and redaction webhooks implemented; PII scrubbing in the AI pipeline.
  • SOC 2: not yet certified. The controls above are implemented and tested in CI (route audits, credential audits, dependency advisories); certification is on the roadmap and this page will be updated when obtained.
  • Data residency: primary infrastructure in US regions; EU residency planning is available for enterprise discussions.

Found a security issue?

We run a coordinated disclosure process. Please report vulnerabilities privately and allow time for remediation before public disclosure.

Contact the security team