One mutable data plane for checkpoints, evaluation outputs, training data, and agent artifacts.
Sync large artifacts without Git history.
Buckets are for checkpoints, evaluation dumps, scratch data, and other files that change independently from a repository release.
Use one owner-scoped Bucket ID from the CLI, the browser, or an S3-compatible client.
$ mega buckets sync ./checkpoints \
mega://buckets/mega/training-artifacts
Scanning 184 files · 6.37 GB
✓ uploaded changed objects
✓ Bucket available at a stable path
CLI and S3 workflowsSync directories or connect existing object-storage tooling.
Owner-scoped accessPublic or private Buckets inherit the owner access boundary.
Workload mountsAttach Bucket data to Spaces and Jobs without publishing a repository revision.
Move data independently from compute.
Training and inference workloads can read a stable Bucket path while producers continue writing new outputs.
Keep model and dataset repositories versioned; keep high-churn intermediate files mutable.
- CLIread + write
- Spacemounted data
- Jobcheckpoint output
Give changing artifacts a stable home.
MEGA Storage Buckets