Find the answer. Keep building.

A page has an id, a display name and a route path. The path must start with "/" and be unique across the site.

Every page renders a document made of nodes — sections, text, images and component instances — that you arrange on the canvas.

Exactly one page is marked as the home page. Pages can also opt into a layout and carry SEO, navigation grouping and access rules.

A document is a small envelope — version, id, type (page, layout or component), name and a root array of nodes.

Each node has an id and a kind, plus optional props, styles, slots, bindings, events, actions and conditions.

Styles are split into base plus optional tablet and mobile viewports, with hover, focus, active and disabled states.

Mark one page as home. The platform enforces a single-home invariant — the first page flagged as home wins, and later home flags are removed by the schema normaliser.

A missing or empty active page is a schema error and blocks publishing, so make sure the project always points at a real page.

The canvas is a visual editor over the document tree. Common node kinds include section, text, image and component-instance.

A layout renders the active page through a page-content node, which acts as the drop target for the page body.

Styles can be set for base (all sizes), tablet and mobile viewports.

Interaction states are written separately — hover, focus, active and disabled — so each one can override the resting style.

A style value can be a literal or a theme token reference such as color.brand.primary.

For themed values, prefer token references over frozen numbers so the site stays consistent when the theme changes.

Pages carry id, name, path, document, an optional layout, an optional access rule, navigation grouping and a home flag.

They also carry route params (for dynamic pages), a data source, SEO metadata and a status of draft or published.

A page with route params must also set a data source. A dynamic page without a data source is a publish error (routing missing source).

Route params let one page render many records — for example a CMS entry or directory profile — driven by the bound collection.

Component references are validated for circularity before publish. A circular component reference is an error and blocks publishing.

Flatten the cycle — for example by moving shared content into a reusable component — before publishing.

A collection has fields whose types include text, number, boolean, date, image, file, url, email, object, array, any, richtext, select and relation.

Field keys must be unique per collection. A duplicate field key is a publish error.

A query references a collection, applies filters and sort, and can cap results with a limit.

Filter operators are eq, neq, gt, gte, lt, lte, contains and in. A query that references a missing collection or field blocks publish.

Variables are scoped to app, page or component. Formulas support literals, bindings, unary and binary operators, and a small function set.

Formulas and bindings that reference missing collections, fields or pages are publish errors — fix dangling references before publishing.

Site auth is controlled by a master switch, with options for public registration, email verification and password reset.

Providers include emailPassword, magicLink, google, apple, github and microsoft. Session length and multi-device sessions are configurable.

Access rules cover public, guest, authenticated, roles, owner, condition and nobody.

Roles are created and assigned in the editor, and a default role can be set for new people who register.

Each collection permission controls read, create, update and delete, with an optional owner field.

Collection access is enforced in SQL, so a condition rule on a collection permission cannot be compiled and blocks publish. Page-level condition rules are fine.

A workflow is a trigger plus an ordered list of actions. Triggers include manual, pageLoad, event, webhook and schedule.

Actions cover state, navigation, data, auth, UI, form validation, branching, running other workflows and API requests.

Bindings draw from sources like variables, records, queries, context, form, event, formulas, action results, workflow inputs and triggers.

Conditions use eq, neq, gt, gte, lt, lte, contains, startsWith, endsWith, empty, notEmpty, truthy and falsy, grouped with and/or.

Circular workflow references are errors and block publish, just like circular component references.

Break the cycle before publishing — for example by splitting the loop into distinct workflows.

AI-assisted building never writes a whole project document. It expresses every change as an operation plan with a base revision.

The platform applies the plan to that revision; a stale base revision is rejected so an older plan cannot clobber newer work.

Site users are editor-managed, so AI plans never author user records. Secrets are metadata only — there is no operation that writes a secret value.

Domains, redirects, environment flags, integrations and custom code are also deny-listed. AI plans reference them, never author them.

Plans are checked against publish rules before they are applied — single home page, valid paths, https connections, no dangling references.

If a check fails, the plan is rejected with the specific error rather than silently producing an invalid site.

A .sima file is a ZIP containing a manifest and a project document. The project document is authoritative.

Importers only read the manifest and project files; extra knowledge files are inert and ignored.

Before publishing, the project is validated for structure, routing, data, auth, automation and asset references.

Only error findings block publish; warnings and info do not. The result reports issue counts and whether publishing is allowed.

A project that passes validation becomes a published snapshot served by the published runtime.

The published runtime enforces tenant isolation and permissions, so a published site behaves the same way the editor preview does.

Custom domains are configured at the editor level — they are not authored through plans or packages.

If your domain does not connect, check that the domain was added in site settings and that the DNS record matches what the editor shows.

Site settings control whether HTTPS is enforced and how the www variant behaves.

A redirect or HTTPS change is applied through the editor settings surface, not through an AI plan.

Redirects are configured in site settings and are not authorable via operations.

When a redirect behaves unexpectedly, review the redirect list in settings for ordering or conflicting entries.

Sima has no native billing engine. Payments are composed from an integration entry, secret metadata, API request steps and workflows.

Charge, refund and subscribe calls are made against your payment provider through api.request steps in a workflow.

The payment provider key lives in the automation secrets list as metadata only, and the integration references it by secret id.

A secret id that is missing from the secrets list is a publish error, so the connection cannot silently ship without its key.

Analytics is an integration entry referencing a secret id, plus a privacy setting for whether analytics waits for consent.

There is no native analytics engine either — it is modelled the same way as billing, through integrations and automation.

Platform users (workspace members and editors) and site-level end users are separate namespaces.

Site users live inside the project and are managed in the editor; platform users manage the workspace and its sites.

A workspace can be personal or team. Team workspaces let you invite members by email and assign roles.

Ownership can be transferred, and members can be removed from the workspace settings screen.

Every resource is scoped to its workspace and site. Cross-tenant access is never allowed.

If you cannot see a site, check which workspace is active and whether your role grants access to that site.

When publishing is blocked, the validation result lists issue counts. Only errors block publish; warnings and info do not.

Common errors include duplicate field keys, missing data references, circular components or workflows, and non-https API base URLs.

Queries and bindings must reference existing collections, fields, pages, roles, workflows, requests, connections, secrets and assets.

A dangling reference is a publish error. Re-point the query or binding at a real target, or create the missing item.

AI-assisted changes carry a base revision. If the revision is stale, the plan is rejected rather than applied over newer work.

Refresh from the current project state and re-author the plan against the latest revision.

A condition rule on a collection permission cannot be compiled to SQL and blocks publish.

Replace it with roles, owner or authenticated access, or move the condition to a page-level rule where it is allowed.