OpenGradient Architecture

OpenGradient is an EVM-compatible blockchain network that extends standard blockchain capabilities with native AI inference. The network enables developers to execute ML and LLM models directly from smart contracts, with all inference operations atomically executed as part of blockchain transactions.

OpenGradient's novel Hybrid AI Compute Architecture (HACA) leverages node specialization to create a powerful, versatile platform that provides efficient and scalable inference capabilities while remaining fully verifiable and decentralized.

NOTE

As an EVM-compatible blockchain, OpenGradient supports all standard Ethereum tooling and frameworks. In addition to standard EVM capabilities, the network includes native precompiles for AI inference that are validated by specialized validator nodes.

HACA Components

The HACA architecture is made up of the following core node types that work together to power the OpenGradient blockchain:

• Full Nodes: Blockchain validator nodes that execute EVM transactions and maintain the blockchain state. These nodes handle standard EVM operations, native AI inference precompiles, and compute the full state-transition function (STF) for the OpenGradient blockchain.

• Inference Nodes: Specialized validator nodes with optimized hardware designed to execute AI model inference and generate verification proofs. These nodes run inference operations requested by smart contracts and produce verification proofs (ZKML proofs, TEE attestations, or Vanilla inference results) depending on the execution mode and verification method selected. They support various hardware architectures including CPUs, GPUs, TPUs, or software-based TEEs.

• Storage Nodes: Decentralized storage infrastructure powered by Walrus that maintains the Model Hub's filestore. Model files uploaded to the network are stored on Walrus, making them available for inference execution. Models are referenced by their Blob ID and can be retrieved by inference nodes when needed.

• Data Nodes: Secure enclave nodes that provide trusted data access services for smart contracts. Using TLS protocol through Nitriding, these nodes establish encrypted connections with third-party data sources (APIs, databases, oracles) and expose secure data access interfaces that smart contracts can query. This enables smart contracts to access external data needed for model inference.

How It Works

When a smart contract on the OpenGradient blockchain calls an AI inference function:

1. Transaction Execution: The transaction is processed by Full Nodes, which validate the transaction and identify AI inference precompile calls
2. Model Retrieval: Models are retrieved from Walrus storage and provided to Inference Nodes
3. Inference Execution: Inference Nodes execute the model with the provided inputs using specialized hardware
4. Verification: Inference Nodes generate verification proofs (ZKML proofs, TEE attestations, or Vanilla inference results) depending on the selected verification method to prove correct execution
5. State Update: Full Nodes validate the proofs and update the blockchain state with the inference results

All of this happens atomically within a single blockchain transaction, ensuring that model inference is as reliable and composable as any other smart contract operation.

Below is a diagram describing the HACA components and workflow in the OpenGradient blockchain: