Security & Data
Confidential work deserves clear boundaries.
Our public register separates safeguards we can verify today from the principles that govern us and the work still ahead.
Each statement is labelled by what it is today: verified, a governing principle, in progress, or on the roadmap.
“Verified” means confirmed by our own engineering review, not by an external auditor. We hold no certification.
Four established pillars
The boundaries that shape security, confidentiality and governance.
Open a pillar to read the register items behind it, each with its exact status.
Detailed controls
Explore the controls behind the principles.
Each control keeps the status it holds today. Nothing here is a certification.
Workspace-scoped tenancy
VerifiedWorkspace boundaries are enforced at the data layer, including workspace-consistent relationships between client records.
Client access boundaries
VerifiedAt the data layer, client-facing access is limited to what a workspace member may read; internal-only records are unavailable through client access.
Invitation-only access
VerifiedAccess is reviewed before activation; public self-registration is closed at the platform level and the product is sign-in only.
Trusted operations
VerifiedSensitive operations are confined to trusted server contexts.
Client intelligence remains private
PrincipleWe do not use one client's intelligence for another client, and it never moves into a shared or global intelligence layer.
Presentation-safe intelligence
VerifiedRaw research material remains in the backend; clients receive a presentation-safe projection.
Research policy gate
VerifiedCertain entities are excluded from research by a direct-name policy check.
Identifier-only workflow events
VerifiedWorkflow events carry only identifiers and run controls, never the brief, mandate or research content. The model call itself is executed through the orchestration step, so the prompt sent to the model and the model's response pass through the orchestration provider.
Server-side model boundary
VerifiedModel calls, prompts and orchestration logic remain server-side. Only presentation-safe outputs required by the product experience are exposed to the client.
No training on client data
PrincipleWe do not train models on client data. Third-party processing remains subject to the provider's terms.
Versioned decision records
VerifiedEvery decision run is recorded with its versions and outcome.
Beauty-industry governance
PrincipleBy design, the system is governed by beauty-industry expertise rather than generic automation.
Protected decision system
VerifiedThe protected decision system remains server-side as a versioned module. Model calls, prompts and orchestration logic remain server-side. Only presentation-safe outputs required by the product experience are exposed to the client, with internal references removed from client-facing text.
Canonical source evidence
VerifiedEvery output claim links to server-canonical sources, and model citations are checked against the sources actually retrieved.
Run audit record
VerifiedDecision runs retain operational metrics, versions and outcomes for auditability.
How Nawa Nova handles your data
How Nawa Nova handles your data
A simple view of where information lives, when AI is used and how intelligence moves through the platform.
- User browser
- Nawa Nova application
- Supabase / workspace data
- Server-side intelligence
- Anthropic / Claude
- Inngest orchestration
- Presentation-safe outputs
Lovable is used to build and host the Nawa Nova application.
The application provides the authenticated interface for creating projects, defining briefs and reviewing intelligence and decisions.
Workspace and project records are not stored in the application layer; the interface reads them from and writes them to Supabase.
Authenticated workspace and project records are persisted in Supabase Postgres.
They include workspaces, memberships, projects, briefs, decision mandates, analysis runs and persisted outputs such as winning concepts and decisions.
Workspace membership governs which records an authenticated user can access.
Row-level security policies and table grants restrict client access according to that membership.
Claude is called from server-only modules for bounded research and decision-reasoning tasks.
It receives the context assembled for each task, and its structured responses are validated before they move forward.
Claude is not the system of record; durable outputs are persisted in Supabase.
Prompts, orchestration and the NB BEAUTY CODE™ methodology remain in server-only modules.
Inngest coordinates the longer-running asynchronous intelligence workflow.
Brand Intelligence, Competitive Intelligence and Market & Signals run as separate streams before the validated evidence moves to the Decision Engine.
The workflow defines sequencing, limited retries and failure handling.
Workflow events carry identifiers and run controls, not the brief, decision mandate or research content.
The model call itself is executed through the orchestration step, so the prompt sent to the model and the model's response pass through Inngest.
Inngest is not the primary project-data store; project records and persisted outputs remain in Supabase.
The product interface requests selected workspace, project, brief, mandate, run and winning-concept fields.
Live intelligence reaches the browser through a presentation-safe projection containing display values, evidence labels, confidence and source references.
Raw research packs have no client grant and are not read by the product interface.
Server prompts and orchestration logic remain in server-only modules rather than the browser bundle.
- Nawa Nova app
- Product interface / authenticated experience
- Supabase
- Workspace and project persistence / access control
- Anthropic / Claude
- Model reasoning
- Inngest
- Workflow orchestration / retries / sequencing
- NB BEAUTY CODE™
- Proprietary decision methodology and intelligence logic
Nawa Nova combines these layers into a governed decision system; it is not a thin interface over a model.
Forward register
What we are building next
Stated openly as in progress or on the roadmap, never as available today.
Formal lifecycle procedures
In progressWe are formalising documented retention, deletion and revocation procedures.
Data-residency commitments
RoadmapWe plan to define and publish clear data-residency commitments.
Independent assurance programme
RoadmapWe are early-stage, claim no current certification, and plan to establish an independent assurance programme.
Restricted-entity protection
In progressWe are developing reconstruction-resistant protection for restricted entities.
Security contact
A question about security or data?
For questions about personal data, including access, correction and deletion requests, write to privacy@nawanova.com. For any other question about security or this register, write to hello@nawanova.com. We answer with the same labelling you see here.
Access is reviewed before activation. Request access below and we will confirm the appropriate level during onboarding.
Request access
Request access
Tell us what you need to decide and we’ll confirm the right access path.
Already invited? Sign in