Data model

Every table carries a site_id. That column is the tenant boundary: queries are always scoped by it, and the API refuses cross-site reads even when an ID is guessed correctly.

IDs are UUIDs generated on insert. Content additionally has a numeric content_id for humans — the #111 you see in the list table.

content

One row per record of any type.

Column	Type	Notes
id	text (UUID)	Primary key
content_id	integer	Human-facing number, auto-increment
site_id	text	Tenant
type	text	Collection name — posts, pages, insights…
slug	text	URL segment, unique within a type
title	text
body	text	HTML
status	text	publish · draft · trash
meta	JSON	Terms, excerpt, featured image, custom field values
created_at / updated_at	timestamp

media

Metadata for files stored in R2.

Column	Type	Notes
id	text (UUID)
site_id	text
key	text	Object key in R2, e.g. demo/2316-hero.webp
url	text	Public URL
filename	text	Original name
mime_type	text
alt / caption	text
width / height	integer	For images
size_bytes	integer
source_url	text	Where it was imported from, when applicable

users

Column	Type	Notes
email	text	Identity used to sign in
site_id	text
name	text	Display name
role	text	Matches a row in roles.name
active	boolean
created_at	timestamp

roles

Column	Type	Notes
id	text (UUID)
site_id	text
name	text	Machine name, e.g. staff
label	text	Shown in the UI
capabilities	JSON array	e.g. ["content.read","content.write"]

taxonomies

Column	Type	Notes
id	text (UUID)
site_id	text
name	text	Machine name, e.g. categories
label	text
description	text
hierarchical	boolean	Category-style parent/child
content_types	JSON array	Which types it applies to

terms

Column	Type	Notes
id	text (UUID)
site_id	text
taxonomy	text	FK-ish to taxonomies.name
term_key	text	Original external id, when imported
name / slug	text
description	text

collections

Column	Type	Notes
id	text (UUID)
site_id	text
name	text	Becomes content.type
label	text
fields	JSON array	{ name, label, type, options }

Relationships

taxonomies (1) ──< terms
      │                  │
      └── content_types  └── referenced from content.meta.terms[] { taxonomy, id, name, slug }

collections (1) ──< content   via content.type = collections.name
roles       (1) ──< users     via users.role = roles.name
media       (1) ──< content   via content.meta.featured_image

Term attachments live inside content.meta.terms as an array of small objects. That is why renaming a term has to propagate: the name is copied into every record that references it.