Legal

Security

Last updated 7 August 2026

Compliments OS is an alpha product. This page describes what is genuinely in place today and, just as importantly, what is not yet. We would rather you know before you evaluate us than after.

Separation between restaurant groups

Every record in the system belongs to an organization, and every query is scoped to the organization of the person making it. No page or action accepts an organization identifier from the browser — it is always resolved from the signed-in session on the server. A record belonging to another group does not return a redirect or a blank page; it does not resolve at all.

This is covered by automated tests that run on every change.

Access inside a group

Five roles — owner, admin, marketing, manager, and viewer — with a defined set of capabilities each. Permission is checked on the server for every action that changes data, not merely by hiding buttons. Sending a live campaign, merging guest records, and managing integrations are restricted to owners and admins.

Stored credentials

Integration credentials are encrypted at rest with AES-256-GCM before they touch the database, and are never decrypted for display — the interface shows only a masked hint recorded when you saved them. If the encryption key is not configured, the system refuses to store a credential rather than storing it in the clear.

Guest communication

Email sending is off by default and requires several independent conditions before a real message can leave the system, including an explicit in-app confirmation by an owner or admin. Consent is re-read from the database at the moment of sending rather than trusted from an earlier preview. Every marketing message carries a working, guest-specific unsubscribe link.

Review responses drafted by AI never publish on their own. A person approves every one.

AI handling

The model receives the minimum guest context needed for the task and is never permitted to write to your data. Segment definitions produced by AI are validated against a fixed list of allowed fields before they run — the model cannot construct a query directly. Anything the AI produces is labelled as a suggestion and waits for a person.

Logging

Tokens, credentials, email addresses, and phone numbers are stripped from log output and from audit metadata before either is written. Meaningful changes — sending, merging, publishing, connecting an integration — are recorded in an audit trail.

What is not in place yet

This is an alpha, and the honest list matters more than the reassuring one:

  • No SOC 2, ISO 27001, or other certification. We do not claim one and you should not assume one.
  • Authentication is email and password only. There is no single sign-on, no multi-factor authentication, and no self-service password reset yet.
  • Guest deletion and data-export requests are handled manually by us rather than through a self-service tool.
  • No formal penetration test has been carried out.
  • There is no published uptime commitment. This is a product being built alongside its first users.

Every one of these is on the path to a production release. None of them is a surprise we would rather you discovered during a pilot.

Reporting something

If you believe you have found a security problem, email alex@rubixagency.com with the details. We will acknowledge it and tell you what we are doing about it. We will not pursue anyone who reports a genuine issue in good faith.