OpenGradient Docs

Appearance

Proof Settlement ​

Proof settlement is the process by which inference proofs and attestations are verified and recorded on the OpenGradient ledger. This makes all AI inference operations end-to-end verifiable and auditable by every full node on the network.

OpenGradient uses configurable verification methods: ML execution supports ZKML, TEE, and Vanilla verification, while LLM execution uses TEE verification (see ML Execution and LLM Execution for details).

How It Works ​

After an inference node executes a request and generates a proof:

  1. Proof Submission: The inference proof or TEE attestation is submitted to the network
  2. Validator Verification: Full nodes verify the proof during the consensus round
  3. Block Inclusion: The verified proof is included in the block and recorded to the ledger
  4. Finalization: Once 2/3+ validators agree, the proof is permanently settled

Network state is persisted on blockchain nodes. For large proofs like ZKML, the proof data is stored on Walrus with only the reference recorded on-chain—keeping the blockchain efficient while maintaining full verifiability.

Verification Guarantees ​

By including inference proofs in OpenGradient blocks, transactions that use AI models become end-to-end verifiable:

  • Full Node Verification: Every full node on the network verifies inference proofs as part of block validation
  • Consensus Security: Proof verification is part of the state transition function, ensuring all validators agree on inference correctness
  • Immutable Record: Once settled, inference proofs are permanently recorded and cannot be tampered with

This provides the same level of security as regular EVM state transitions—all inferences are verified and secured by the entire validator set.

LLM Settlement Modes ​

For LLM inference, clients can choose from different settlement modes that control how much data is stored on-chain. See Settlement Modes for the full comparison of SETTLE_INDIVIDUAL, SETTLE_BATCH, and SETTLE_INDIVIDUAL_WITH_METADATA.

Cross-Chain Verification ​

Individual users and light clients can also verify specific inferences they are interested in, allowing off-chain and cross-chain users to gain full confidence in their results.