Skip to main content
Background sync recipes use background agent tasks to discover and fetch content from connected services without holding an HTTP request open. The common pattern is:
  1. Launch a read-only background task with a focused prompt and optional skill.
  2. Poll the task until it returns a session_id.
  3. Fetch the persisted tool trace for that session.
  4. Correlate the agent’s resource manifest with successful tool responses.
  5. Transform those responses into records for your index or data pipeline.
Each server exposes different nested-resource identifiers and retrieval tools, so the final correlation step is server-specific. Keep the task lifecycle generic, then use the appropriate exporter for the source system.
Background sync traces can contain complete customer documents, database rows, comments, and other sensitive content. Keep platform credentials and generated files server-side, apply your normal retention controls, and never commit raw traces to source control.

​
Available recipes

Sync Notion files

Discover Notion pages, child pages, databases, rows, and comments, then build resource-centric JSON from the task trace.

Public cookbook source

View the reusable task runner, Notion exporter, tests, and full usage guide.

​
Adding another source

Future recipes can reuse the same background-task runner and replace only the exporter. A source-specific exporter should define:
  • the stable resource identifier used to deduplicate fetches;
  • which tool response contains the canonical content;
  • how nested resources refer to their parent;
  • how pagination and retries appear in the trace; and
  • which ambiguous calls should be reported instead of guessed.
This structure works for other document systems such as Google Drive, Confluence, Overleaf, or GitHub, while keeping source-specific parsing isolated.