Anchors model identity, versions, licenses and the access policy attached to each model, so a registry can be verified independently of its author.
Version Anchors
Every compute job carries a precise model version anchor, so a receipt names exactly the weights that were run - not a moving target.
Evaluation Summaries
Signed evaluation summaries commit to the chain when useful; high-frequency benchmark detail stays off-chain and reviewable by reference.
Compute Receipts
A compact, replayable proof per job: model version, dataset slice, runtime build, cost and outcome - reviewable long after the run finishes.
Job Definitions
Training and inference jobs declare their inputs, hardware class, cost ceiling and reproducibility bounds before execution starts.
Passkey Identity
Passkey-based identity gates who can run, read and charge for each model - without turning the wallet into the product.
Scoped Access Policy
Versioned, scoped permissions decide who can reach a model, a dataset or a job - policy lives on-chain where it must be verifiable.
EVM Wallet Interop
The Grid exposes an EVM-compatible surface so registries, receipts and policies can be referenced from EVM tooling and contracts.
EIP-6963 Discovery
Wallet-provider presence is detected through the EIP-6963 announcement standard - detection only, never an account-access request.
Solana Wallet Interop
Solana wallet flows connect to the same identity and signing boundary, keeping cross-ecosystem access reviewable.
Cross-Chain References
Registries and receipts can be referenced from other chains without fragmenting settlement - the Grid keeps the verifiable record.
Verifiable Data Feeds
Oracles and external signals enter as signed, reviewable references - evaluation detail stays off-chain until a summary is worth committing.
Event Indexing
Chain events are indexed into job state, receipt history and audit trails so product surfaces show reconcilable evidence, not vibes.
Formal Verification
Critical contract properties get deterministic test coverage and adversarial review before registries, receipts or policy scopes expand.
Content Addressing
Datasets, weights and artifacts are addressed by content hash and referenced on the Grid - the pointer is verifiable, the file stays private.
Permanent Storage
A single upfront payment stores a file and keeps it retrievable - an AI pipeline is not tied to a monthly subscription that can lapse.
Policy-Bound Retrieval
Encryption and scoped permissions decide who can read a stored file, without exposing it publicly or depending on one administrator.
Encryption
Large or private material stays encrypted and off-chain while the chain holds the pointer, the license and the access policy.
Provenance Records
Licensing terms and provenance records anchor to permanent storage, so every dataset reference carries the evidence a reviewer needs.
Hardware Classes
A published hardware class lets every node machine be audited against the same bill of materials - specifications describe the intended class, not a delivery promise.
GPU-Subnet Scheduling
Heavy compute is scheduled to GPU node machines across independent providers - routing without a central broker.
Capabilities - Ztakora | One verifiable compute stack