Data Sharing

The product model for data sharing is “publish, then subscribe.” A publisher selects a Catalog object and target workspace; the subscriber mounts it as a read-only subscribed library. This is not export and does not give the subscriber an independently writable copy.

Not currently automatable through the SDK

In the reviewed main branches of moi-python-sdk and moi-go-sdk, neither SDK defines public Publication, Subscription, or Data Share types or methods.

More importantly, the product implementation recorded in this documentation repository still uses stateful frontend mocks for data sharing rather than a production backend. Publication, subscription, and cancellation state shown by that interface is not durable cross-workspace authorization.

Consequently:

  • Do not construct guessed endpoints for publication, subscription, renewal, or revocation.

  • Do not grant downstream access based on frontend mock state.

  • Do not feed pre-refresh UI state into audit or billing systems.

  • Do not describe a Catalog download or export task as data sharing; their authorization and lifecycle differ.

Preserve the resource model for future integration

Once a production backend and public SDK support are released, integration code should keep these resources distinct:

Resource

Owner

Relationship to persist

Publication

Publishing workspace

Source Catalog object, target workspace, permission, and validity period

Subscription

Subscribing workspace

Publication, mount location, local name, and subscription state

Mounted Object

Subscriber’s Catalog

Read-only object and its source Publication

A safe workflow should:

  1. Verify the source with its Catalog ID and full path before publication.

  2. Name target workspaces explicitly; do not produce a public link.

  3. Treat the mounted object as read-only and retain its source workspace and Publication identity.

  4. Inspect query, knowledge-base, and workflow dependencies before unsubscribe or revocation.

  5. Treat expiration and revocation as expected access loss, not as a network failure to retry indefinitely.

These are product-resource and integration constraints, not a list of callable methods today. Implement automation only after the official SDK publishes the corresponding types and methods.

Available alternative

If the immediate requirement is to deliver data to an external system, use Data export in AI Studio. If another workspace needs continuing read-only access, wait for production data-sharing and SDK support rather than substituting guessed authorization calls.