Draft, 14 September 2026. Not yet reviewed by qualified counsel; not to be relied on until it is. Same structure as HelpingNeighbor's notice, written for what this product actually does.
Effective: [date on publication] · Operator: [legal entity], trading as LumenNexus, lumennexus.ai · Contact: privacy@lumennexus.ai
LumenNexus is a data-governance control plane. Organisations ("workspaces") put data products in it so that who may read what is decided in the database, every decision is written down, and what left the boundary can be shown. Two things follow, and the rest of this notice is detail on them:
We do not sell personal data. We do not use it for advertising. There are no third-party trackers on the site.
[Legal entity], operating LumenNexus at lumennexus.ai. Hosted in the European Union (Hetzner Online GmbH, Germany/Finland). One operator; the contact address above reaches a person.
HttpOnly, Secure, SameSite=Lax, 14 days). No analytics or advertising cookies.| Purpose | Data | Basis (GDPR) |
|---|---|---|
| Running your account and the workspaces you belong to | Account, membership | Contract |
| Keeping the audit trail every workspace is sold on | Actions, actor email | Contract; the workspace's legitimate interest |
| Refusing repeated wrong passwords | Sign-in attempts | Legitimate interest (security) |
| The second sign-in step | Email, code hash | Contract; legitimate interest (security) |
| Investigating incidents | Server logs | Legitimate interest (security) |
| Billing | Stripe customer id, plan | Contract |
| Answering you | Support messages | Legitimate interest |
The assistant (Candela) answers from documents and products inside your workspace, filtered by your clearance before anything is retrieved, and cites what it used. Depending on how the workspace is configured, the model call goes either to inference the operator runs on private infrastructure or to an endpoint the workspace's owner has chosen; a workspace can set a no-egress policy that refuses any public endpoint. We do not use workspace content to train any model. We keep counts of model calls, not their content (§2.2).
Off by default, and never offered at all to a workspace with a no-egress policy. An owner or admin decides it for the workspace; you can refuse it for yourself, and your refusal wins.
When it is on, a service called Fast-Back (listed in §6) remembers how you like to be helped. It knows you by an opaque id for you-in-this-workspace — not your name, address or role — and holds the preferences you state.
If the workspace also switches on the separate remembering conversations setting, your own questions may be stored there as well, after addresses, telephone numbers, national and health identifiers, payment cards, dates of birth, and this workspace's own product, column and people names are removed. The assistant's answers are never sent — they quote your documents, so not one of them leaves, whatever the settings say. You are shown exactly what would be stored before it is stored, and the page What we remember about you lets you read, correct, export or erase all of it at any time. Erasing your account erases this too.
adr/0016.The current list, with what each receives, is kept at SUB-PROCESSORS.md and changes are notified to workspace owners at least 14 days before they take effect.
Whatever jurisdiction you are in, you can:
GET /api/me/export from your account, or ask us. The export reports counts of governed content, not the content — that belongs to the workspace.POST /api/me/erase or ask us. Erasure pseudonymises: your email and name are replaced by a stable pseudonym, your sessions end, and the audit entries you made remain under the pseudonym. An erasure is itself recorded.For data inside a workspace, exercise these rights with the workspace owner (the controller); we will help them answer within the time the DPA sets. Residents of California and other US states with privacy statutes have the rights those statutes give; we do not sell or share personal information as those statutes define it.
| Data | Kept |
|---|---|
| Account and memberships | Until you erase the account or the last workspace closes |
| Audit trail | For the life of the workspace, and afterwards under its own retention rules; never edited |
| Sessions | 14 days, or until you sign out |
| Sign-in attempts | 24 hours |
| Sign-in codes | 10 minutes |
| Server logs | 90 days |
| Database backups | 14 days |
| Model usage counts | For the life of the workspace |
Isolation between workspaces is enforced in the database engine, not in application code; every table holding workspace data carries a row-level security policy and a test that it is forced. Passwords are hashed; sessions are opaque tokens hashed at rest; the audit trail is hash-chained and verified; the site is served over TLS with HSTS. What is in place, what is partial and what is not yet built is published honestly in our SOC 2 control mapping, and the security posture page inside every workspace says which of its answers are measured, declared, or unknown. Report a vulnerability to the address in /.well-known/security.txt.
The service is for organisations and is not directed at anyone under 16. We do not knowingly hold a child's account; tell us and it will be erased.
We will post changes here with a new effective date and, for anything material, tell workspace owners at least 14 days before. Continued use after the effective date is acceptance of the change.