Your Brand Data Is Not
Our Business Model.

1 Who Gets In (And Who Doesn't)

• • Invite-only accounts. An organisation admin invites a user by work email; the user sets their own password through a time-limited link. No default or shared credentials. Email addresses are verified.
• • Single sign-on: SAML 2.0 through our identity provider (Microsoft Entra ID, Okta, Google Workspace), configured per organisation and email domain. Google sign-in is also available.
• • Multi-factor authentication via TOTP (Google Authenticator, Microsoft Authenticator, Authy, 1Password). Any user can turn it on. Org admins can make it mandatory and the platform blocks access until enrolment is complete.
• • Removing a user takes effect immediately at the database layer, not just in the interface.

2 Your Data, Your Rules

• • Your data is never used to train, fine-tune or evaluate any AI model. Not by us, and not by the providers we use (see section 5).
• • We store only what the service needs: account data, authentication data and technical logs. We do not build profiles of your customers.
• • Outputs download in open formats (DOCX, PDF, CSV, PNG, MP4) at any time. A full export of an organisation's content is provided on request.
• • Deletion on request, confirmed in writing. After termination your content stays exportable for 30 days and is then permanently deleted.

3 Your Data Can't See Their Data

• • 900+ row-level security policies inside the database. Row-level security is enabled on every table, not most of them.
• • Every row carries partner, organisation and workspace identifiers, and the database checks them against the caller on every read and write. A bug in application code cannot return another tenant's rows.
• • White-label partners are isolated from each other and from gimmefy's own tenants in exactly the same way.
• • The isolation perimeter is checked with automated security advisors on every quarterly review.

4 Not Everyone Gets the Keys

• • Organisation roles: Org Admin (users, settings, billing), Manager (day-to-day content and workspaces), Member (create and use).
• • White-label partners hold a partner-level role for provisioning and support across the client organisations they host.
• • Role changes are checked in the database, so nobody can escalate their own privileges through the API.
• • Our own engineers: named individuals, least privilege, MFA on every production console, quarterly access review. Customer content is opened only to resolve a request you raised or an incident, and that access is logged.

5 What Happens When We Talk to the AI

The question enterprise security teams ask most, answered exactly.

• • Sent per request: your prompt, the brand context relevant to that task (passages retrieved by semantic search, never the whole vault), files you attached to that request, and the thread history.
• • Never sent: another tenant's data, your credentials, billing data, your user directory. Nothing goes to a provider unless a user picks a capability that uses it.
• • Every provider is accessed through its commercial API. Anthropic, OpenAI and Google state in their terms that API data is not used to train their models.
• • Retention after the request: Anthropic deletes within 30 days, OpenAI holds up to 30 days for abuse monitoring, Google holds 55 days for the Gemini API. Then it is gone.
• • Zero data retention: we run on standard commercial terms today, so those windows apply. Enterprise and white-label clients who need ZDR can have it arranged with the providers that offer it, and their workspace restricted to those providers.
• • Around 50 models from roughly 20 providers are live at any time. The current roster is in the product under AI Stack, and can be restricted to named providers per workspace.

6 Built on Certified Infrastructure. Configured by Us.

• • Two production regions: Frankfurt, Germany (AWS eu-central-1) for EU and UK customers, and Tokyo, Japan (AWS ap-northeast-1) for Asia-Pacific. Your account lives in one and stays there.
• • Database, authentication, file storage and 200+ edge functions run on Supabase (SOC 2 Type II) inside AWS (ISO 27001, SOC 1/2/3, PCI DSS). Media rendering on Google Cloud Run, in-region.
• • The database is not reachable from the public internet. Every API call carries a short-lived, authenticated token.
• • Daily backups with point-in-time recovery, retained in-region. AES-256 at rest, TLS 1.2+ in transit.
• • Separate development, staging and production environments. No customer data outside production.

7 Continuous Protection

Our application layer defends against the common web vulnerabilities:

• • Origin allow-listing on every backend function.
• • HTML and SVG generated by models is sanitised with DOMPurify, and previews render inside sandboxed iframes.
• • Protection against server-side request forgery on every function that fetches a URL.
• • Input validation with type, size and length limits. Strict file-type allow-listing on uploads.
• • Peer-reviewed pull requests, automated tests and a fixed promotion path (dev, main, staging, production) for every change, including every access-control policy.

Third-Party Subprocessors

Every third party that can receive customer data, what it gets, and where it processes. Nothing reaches an AI or data-collection provider unless a user chooses a capability that uses it.

What's In Progress

Listed so nobody mistakes them for things we already do.