Pay once. Store forever.
Ztakora Drive is the permanent data layer for datasets, model weights, embeddings and artifacts - stored once, referenced by hash and kept retrievable behind policy. Pricing and availability figures are design targets, not offers.
Storage that
Outlives the Subscription
Stored Once. Never Re-Uploaded.
Licensed and private corpora are stored once, content-addressed and referenced by hash on the Grid - never re-uploaded per run.
Checkpoints That Stay Retrievable
Checkpoints and weight files are kept retrievable for reproducible evaluation and audit, resolved by content hash long after the run.
Immutable Job Outputs
Evaluation records and job outputs accompany any published benchmark - reviewable evidence that does not move when a plan changes.
Secure by design.
Private by default.
Encrypted and Off-Chain
Large or private material stays encrypted and off-chain while the chain holds the pointer, the license and the access policy that make the record reviewable.
Policy-Bound Retrieval
Encryption and scoped permissions decide who can read a file - without exposing it publicly or relying on a single administrator's workspace.
Licensing and Provenance Records
Every file carries its licensing terms and provenance record, anchored to permanent storage - a single upfront payment keeps it retrievable, not a subscription that can lapse.
See the difference
Drive makes
Can we reuse last month's training corpus for the next run?
Yes - as long as the storage subscription is still active. Please re-upload the dataset for this run, keep it inside the current plan's quota, and note that files are deleted after a retention window once the account lapses.
Already stored. The corpus went into Ztakora Drive once, is content-addressed and referenced by hash - no re-upload per run:
- Stored once, referenced by hash
every job pulls the same content address instead of a fresh upload - A single upfront payment
the file stays retrievable - not tied to a monthly subscription that can lapse - Licensed and private material stays encrypted
off-chain, while the chain holds the pointer, the license and the access policy
The run reuses the exact corpus the previous job trained on - verifiable from the compute receipt.
Who can read the checkpoint - and can it still be audited in a year?
Access follows the current workspace plan. Administrators can change permissions, files can be deleted, and audit history depends on whatever the cloud provider still logs.
Retrieval is bound to policy, not to a workspace seat:
- Scoped permissions decide the reader
encryption and policy control who can read a file without exposing it publicly - Checkpoints stay retrievable
weight files remain addressable for reproducible evaluation and audit - Provenance is the record
licensing and provenance anchor to permanent storage as immutable job outputs
A year later the same hash still resolves - with the license and the policy that were committed on the Grid.
Verifiable compute receipts without putting datasets on-chain
How the runtime can keep private data and heavy compute off-chain while anchoring compact, reviewable proofs of what was run.
