Skip to main content

Level 1 — System Context

The System Context view answers: who interacts with Noxys, and which external systems does Noxys depend on? It is intentionally free of internal detail.

Diagram

Actors

ActorDescriptionInteraction with Noxys
EmployeeEnd user of generative-AI toolsSends prompts via browser, IDE, or desktop integration
Customer AdminCISO, IT operations, security opsConfigures policies, reviews alerts, manages users
DPO / ComplianceData Protection Officer, compliance leadExports audit trail, DPIA, transparency reports
Customer SIEMSentinel, Splunk, Sekoia, Elastic, CrowdstrikeReceives streamed audit events
Customer IdPEntra ID, Okta, Google WorkspaceAuthenticates console users (SSO), provisions accounts (SCIM)

External systems

SystemWhy it matters
AI providers (OpenAI, Anthropic, Google, Mistral, …)The destinations Noxys mediates. Connections happen only after policy evaluation and Risk Signal sanitisation.
StripeSubscription billing for Pro / Enterprise tiers (see ADR-010).
Postmark / SESTransactional email — invitations, password reset, alert digests.

Trust boundaries

  1. Customer organisation — owns Employees, Admins, IdP, SIEM. Noxys is invited in via signed Terms + DPA.
  2. Noxys SaaS (EU) — the sovereign control plane runs in the EU (OVH / Scaleway, see noxys-infra). No customer data leaves the EU.
  3. External AI providers — third parties; Noxys ensures only sanitised payloads cross this boundary, per the customer's policy.

Privacy-by-design promise

Even at this zoomed-out view, the architecture preserves a critical invariant: raw prompts are never persisted by Noxys. Detection happens inline; only a SHA-256 hash plus metadata is stored. The transparent boundary diagrammed above is therefore a policy boundary, not a data lake.