CMS sync

Returns {connected: false} when the site has no connection, so this is safe to call before one is set up.

Forgets the connection and the per-page links to it. Content is kept on both sides — this stops syncing, it does not delete anything.

What this site does about its CMS unattended: whether the nightly comparison covers it, what it may pull without being asked, and the URL the CMS calls when its content changes.

Sets the standing rules. autoPull is off, unedited (pages the CMS changed that nobody has touched here since we last sent them) or all; a page that changed on both sides is never pulled automatically under any of them. Nothing here ever pushes — copy reaching a live website stays a person's decision.

Stores the credentials for a CMS and verifies them against it. Nothing is synced yet — pull or push afterward.

Compares both sides and returns the plan — what would be created, updated, and what conflicts. Read-only: nothing moves until you pull or push.

Brings the CMS's content into the site. Preview it first with GET /sites/{id}/cms/plan, which reports what a pull would change without changing it.

A pull touching more than a handful of pages answers 202 with a job instead of the result: hundreds of rate-limited calls to a CMS take minutes, which is longer than any request should be held open. Follow the run at GET /sites/{id}/cms/sync-jobs/{job_id}. Send background: true to be given a job whatever the size.

Sends the site's content to the CMS. Local edits win over anything changed in the CMS since the last sync; the plan endpoint reports those conflicts before you commit to them.

Answers 202 with a job once the push is larger than a handful of pages — see pullFromCms.

Same comparison as the plan, and it remembers the verdict on the site so the page tree can show where the two sides disagree without comparing again. Use the plan endpoint for a read that leaves nothing behind.

Mirrors the CMS's component definitions into a content-schema library, so generated content is shaped the way the CMS expects it.

Sends the workspace's content schemas to the CMS as component definitions, so pages pushed afterward have somewhere to land.

Every recent pull and push for this site, newest first: who ran it, what it moved, how long it took and which pages it could not settle.

Progress and outcome of one run. done and failed advance against total while it works; stalled reports a run whose worker was lost, which can be resumed without losing what it applied.

Stops the run at its next checkpoint. Pages already synced stay synced — this is a stop, not a rollback.

Continues a run that stalled or failed, from its last checkpoint. Pages it already settled are not touched again.

Starts a new run over exactly the pages the named run could not settle. The original run keeps its record.

Issues the URL to register in the CMS, so a change there is noticed in seconds rather than at the next scheduled comparison. Called again it replaces the URL, which is how one is rotated. Send secret to have deliveries checked against the provider's signature. Refused where the connected CMS cannot call a URL.

Withdraws the URL. Deliveries to it stop being accepted immediately; the site falls back to the scheduled comparison.

Both versions of one page, field by field, with the differing fields first — what a conflict needs in order to be decidable.

Settles a page that changed on both sides by keeping each field from the side named in choices — {"title": "local", "body": "remote"}. Fields nobody names keep the local value; _name, _slug and _text name the three that are not content fields. The merged copy is written here and then sent to the CMS under the same approval gate as any push, so on a site under review it waits for a reviewer and the response reports awaitingApproval.

Replaces the local copy of this page with the CMS's. The local version is discarded.

Pushes a single page named explicitly. force bypasses the approval gate and the nothing-changed check, since the caller asked for this page by name; a genuine conflict is still reported unless resolution says how to settle it.

The providers a site can connect to, and the fields each one needs to connect. Static catalog — it reflects no workspace data.