ironshard.branch, snapshots + forks

Snapshot any bucket.
Branch it instantly.
Recover anything.

Copying a bucket to experiment is slow and expensive. Overwrite production and the old state is gone. Ad-hoc copies leave no record of what ran, or who promoted it. Branch fixes all three, without touching the source.

Early AccessTalk to Us

How snapshots, copy-on-write branches, and merges work: the Branching and Snapshots docs.

How Branch works Immutable snapshots accrue down a main timeline. A writable branch forks from a snapshot, takes a new write and diverges, while point-in-time recovery reads back from any snapshot. mainrestore to any snapshotfork →experimentnew write — diverges snapshots accrue on main · branches diverge and take writes · recover from any point
// how it works

Snapshot, branch, experiment, promote.

The source is never touched, and every action is logged automatically.

01

Snapshot — immutable, instant

Capture a bucket’s exact state at a point in time. Zero-copy, instant, and read-only — it never changes.

instant · immutable
02

Branch a writable fork

Fork from any snapshot or live state. Copy-on-write, so storage is zero until the branch diverges.

zero-copy until divergence
03

Experiment in isolation

Run pipelines, migrations, or training against the branch. Many branches run in parallel; the source is untouched.

parallel · isolated
04

Promote or discard. Log is automatic.

Promote through a governed workflow or discard at zero cost. Every action is written to an immutable log.

governed · audit-logged
// workflows

A safe workspace for anything.

Spin up an isolated branch, run a job, check the results, then promote or discard. Nothing touches the source until you say so.

AI & ML data prep
agent experimentationRAG corpus preparationdataset normalizationclassification & labelingtraining-set curationembedding / index regeneration
Transform & enrich
data cleaningformat conversionmetadata enrichmentobject reorganizationdeduplicationcompression / re-encoding
Govern & validate
shadow testingdocument redactionPII detection & maskinganonymization for sharingdata-quality validationrestore / DR drills
Runs on any IronShard bucket, native or mirrored. Data living in external S3? Mirror brings it in, then branch it here. For the mechanics underneath — snapshots, copy-on-write, governance on every branch — see the Branching and Snapshots docs.
// review & promote

Nothing ships until a human says so.

When the work is done, IronShard shows you exactly what changed: added, modified, removed, plus cost and policy impact. Then you decide.

Review and promote A branch arrives at a review window summarizing what changed. A human then promotes it, keeps it as an artifact, or discards it. branchreview+ 128 new~ 34 modified− 12 removed✓ policy · approvalsest. cost +$0.40Promotemerge back · or managed bucketKeepaudit artifact · or exportDiscardzero cost · production untouched
Agents experiment on branches. Production lives in managed buckets.
// use cases

What this makes possible.

Agent sandboxing

Give an autonomous agent its own branch to read, write, and delete freely. If a run goes wrong, discard it at zero cost. Production is never exposed.

autonomous · fully isolated

Shadow testing

Run a new model or agent against real data on a branch, side by side with production. Compare outputs and promote only the version that wins.

real data · zero production risk

Recovery & reproducibility

Return to any past state. Restore a bucket to the moment before a bad write or deletion, or pin an exact snapshot so a result reproduces no matter how much the source has moved on.

point-in-time · reproducible

Audit & compliance

Every change on a branch is recorded immutably: what an agent read, what it wrote, and who promoted it. A tamper-evident trail you can hand to a regulator.

HIPAA · EU AI Act · SOC 2