The technical detail
For the people who will ask how it actually works.
Skip this section if you just want to sell things — nothing here changes what you do day to day. It is here because sooner or later an IT team, a compliance officer or a marketplace will ask, and you should be able to hand them a straight answer rather than a promise.
WorkflowSeparation of duties
Every product moves through draft, review, approved and published, and the person who approves cannot be the person who created it. Enforced by the system rather than by policy. Eleven distinct roles, from catalogue manager to SEO manager to publisher.
AuditEvery AI call on record
Each model call is logged with the model, its version, the prompt version, how long it took and what it cost. When a listing is challenged, you can reproduce exactly how it was produced instead of arguing about it.
SpendBudgets checked before the call
Per-organisation credit budgets, per-service spending caps and daily request quotas, resolving across platform, organisation and user level. The check runs before the expensive call, not after the invoice. Bulk jobs return an estimate and wait for confirmation.
TenancyIsolation enforced by the database
Multi-tenancy is PostgreSQL row-level security, not application filtering. A missing WHERE clause cannot leak another tenant's catalogue, because the database refuses the read rather than trusting the query.
ProvidersBring your own AI provider
Provider and model are selected per capability, per organisation. Credentials are encrypted at rest with AES-256-GCM. Your data goes to the provider you chose, not wherever the vendor happens to hold a contract.
PrivacyAccess control and erasure
Multi-factor authentication, configurable password policy, and GDPR Article 17 erasure that anonymises personal data while preserving the audit record and referential integrity.
Happy to walk an IT team or a security reviewer through any of this on a call — including the parts that are not finished yet.