Bucket Forking

Forking a bucket works like forking a repository on GitHub: you get your own independent bucket that starts out identical to the original. It is instant and takes no extra storage until you change something, so you can give every experiment, pipeline run, or sub-agent its own writable version of real data, and make changes freely without any risk to the original. Afterwards, a diff shows exactly what changed.

What a Fork Is

You can fork any bucket you can read: one your team owns, or one that another team has shared with you. The result is an ordinary bucket that belongs to your team. You can upload, overwrite, and delete objects in it, share it, or fork it again.

A fork starts with the latest version of every object in the source, as it was when the fork was created. Older versions and delete markers are not carried over, and forks are always versioned.

Independent From the Source

A fork is a separate bucket with its own access, not a view of the original:

  • Nobody gets access automatically. Accounts, users, and agents that can access the source bucket cannot see the fork unless its owner grants them access. The fork owner decides who works with it.
  • Changes stay on their own side. What you write to the fork never reaches the source, and what happens to the source does not reach the fork.
  • The source can disappear. If the source owner deletes objects or the whole bucket, the fork keeps all its objects.
  • Losing access doesn't matter. If a share is revoked after you forked, the fork is unaffected.

Objects in a fork are protected like any other IronShard data, and the fork owner's geo-fencing and access rules apply to the new bucket.

Comparing a Fork With Its Source

A diff shows what changed since the fork was created: objects changed only in the fork, only in the source, or in both, as well as objects added or deleted on either side. It is the way to review what happened on each side. Changes are not merged back automatically.

Where It Is Available

Forking is available in the IronShard console and through MCP, so an agent can fork a bucket it has access to as part of a task. It is not available through the S3 API, because S3 has no equivalent operation.

For a large bucket, creating the fork takes a little time. The fork becomes usable once all its objects are in place, and if the connection is interrupted, repeating the request continues where it left off.

Who Uses Forks

TeamUse Case
Data Scientists / MLFine-tune models or test preprocessing on a real dataset without touching the original
AI Agents, RAGAn isolated, writable version of the data for each task or run
DevOps / PlatformTest transformations or migrations before applying them for real
Teams sharing dataWork privately and with write access on a bucket that another team shares read-only

Current Limitations

  • There is no merge back to the source yet.
  • Snapshots as a separate feature are not offered. A fork is the way to keep a point-in-time version of a bucket.
  • The storage policy of a fork cannot be changed after creation.
  • Billing for data shared between a fork and its source is not final.