Guided migration from an external CMS into Externa collections using the existing AI remote JSON import path — not a second sync engine.

Pipeline details: AI imports & sync. Host allowlist: AI configuration / Environment variables. Webhook alternative: AI API.

Profiles

ImportRemoteJson accepts optional profile:

Profile	Reshape before flatten
`directus`	Drop Directus system fields (`user_created`, `date_*`, …); promote first translation scalars (`title`, `body`, …) onto the row
`wordpress`	Unwrap `{ rendered: "…" }` envelopes on first-level keys (`title`, `content`, `excerpt`)

Without profile, behavior is unchanged: extract list → flatten (nested → JSON string).

Recipe (Directus)

Allowlist the CMS host in AI_REMOTE_IMPORT_HOSTS.
Dry-run:

ImportRemoteJson
  url=https://cms.example/items/pages
  profile=directus
  dry_run=true
  upsert_key=slug
  auth_bearer=…

Confirm proposed fields / preview rows (zero writes).
Re-run with dry_run=false (or omit). Large payloads queue via ImportCollectionJob.

AI chat presets: Import from Directus URL / Import from WordPress REST.

Recipe (WordPress REST)

Same flow with profile=wordpress and a /wp-json/wp/v2/posts (or custom post type) URL. Application Passwords: pass auth_header as Basic … when Bearer is not used.

Limits & gaps

Topic	Detail
Row caps	Sync/async thresholds (~200 queue, max 500 per call) — chunk URLs or repeat
Media	Featured images / file UUIDs are not auto-downloaded into Externa `File` rows in v1
Relations	Stored as JSON strings unless you reshape upstream
Locales	Directus `translations[]` → first locale scalars only; full locale matrix is out of scope
Write-back	One-way import only

Prefer outbound webhooks + CI when the goal is rebuild/purge, not content migration.