01 Security & privacy
Trust is a feature with a changelog.
A security claim you can’t verify is marketing. Below: what’s enforced today, what’s true in the alpha, and what’s planned — each labeled, because the labels are the point.
02 Principles
The non-negotiables.
Provider credentials stay server-side
Claude API keys and any future provider credentials live in server configuration — never in the browser, never in client-side code, never in logs. The alpha never asks you to paste a key into chat.
Approval boundaries hold
Publishing, spending money, deleting data, sending messages — anything destructive or financially consequential pauses for your explicit approval. No irreversible action runs on autopilot, in any mode.
Tools run inside permission boundaries
Every capability — file access, code execution, data lookups — is explicitly authorized per workspace. Browser and computer execution stay gated until they can be implemented safely; we’d rather say “not yet” than imply control we don’t have.
Your work is exportable
Work State is yours. Export goals, decisions, artifacts, and history any time, in readable formats. No lock-in at the data layer — that would contradict the entire product.
Company-grade tenant isolation
Full tenant isolation, SSO, and audit-grade access controls arrive with company spaces. The alpha is personal workspaces only — the trust boundary is small because the surface is small, and we say so instead of implying more.
Independent review
Third-party security review before company spaces launch. Until then: small surface, explicit boundaries, honest labels.
03 Separation
Personal and company, separated by design.
When company spaces launch, the rule is simple and structural: company administration governs company-owned data — skills, shared context, team budgets, audit trails. It does not grant access to personal spaces, private conversations, or personal work history. Ever.
This isn’t a policy document taped over shared infrastructure. Personal and company data live in separate authorization domains from the start, so the boundary doesn’t depend on anyone’s good behavior.
What admins can and cannot see
Can: company skills, shared project state, team spend, audit logs of shared executions.
Cannot: your personal goals, private work history, personal artifacts. The admin console has no view into personal spaces — by construction, not by promise.
04 Straight answers
Asked often, answered plainly.
Where does my data live?
In the alpha: in NWZUP’s hosted Work State store, per personal workspace. Provider calls go to Anthropic’s API under our key management; your content is subject to Anthropic’s commercial terms for API usage. Export any time.
Do you train models on my work?
No. We do not train models on user work, in alpha or ever — that would make us the thing this product exists to replace. Provider data-handling follows each provider’s published API terms.
What happens if I leave?
Export everything first — it’s yours. Then we delete your workspace data on request. The ledger’s anonymized operational metrics (timings, error rates) may be retained for reliability; never your content.
Is the alpha SOC 2 certified?
No — and we won’t imply otherwise. The alpha is personal workspaces with a small surface: server-side credentials, approval gates, permissioned tools, exportable data. Formal certification is part of the company-spaces launch, and this page will say so when it happens.
Small surface. Explicit boundaries.
That’s the alpha’s security story — and we’d rather you read it here than discover it later. Request access if that’s the deal you want.