OpenGradient Architecture
OpenGradient is the network for open intelligence - a decentralized AI platform that enables secure, verifiable AI inference at scale. The underlying infrastructure is a decentralized network built on CometBFT consensus, currently in testnet. The network serves as a verifiable control plane and ledger for AI operations, ensuring all inference requests and results are cryptographically secured and auditable.
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.
OpenGradient Components
The HACA architecture is made up of the following core node types that work together to power the OpenGradient network:
Full Nodes: Coordination nodes that maintain the network's verifiable ledger. These nodes serve as the control plane—handling inference node registration, payment processing, and settlement. After inference is complete, full nodes verify proofs and attestations to ensure all operations are secure and auditable.
Inference Nodes: Stateless worker nodes that provide AI compute resources to the network - GPUs for local inference or access to external model providers like Anthropic or OpenAI. These nodes use TEE attestations or cryptographic proofs like ZKML to ensure privacy, security, and verifiability for all AI operations. Full Nodes verify these proofs, so users can trust that their inference is executed correctly and securely.
Storage: Decentralized storage infrastructure powered by Walrus. Model files and large inference proofs are stored on Walrus, referenced by Blob ID, and retrieved by inference nodes when needed.
Data Nodes: Secure enclave nodes that provide trusted data access services for applications. Running inside TEEs, these nodes establish encrypted connections with third-party data sources (APIs, databases, oracles) and expose secure data access interfaces. This enables applications to securely access external data needed for model execution.
How It Works
OpenGradient delivers web2-like latency while maintaining full verifiability through asynchronous settlement:
- Node Registration: Inference nodes register on the network and are verified - for TEE nodes, this confirms they run correct, untampered software.
- Direct Inference: Requests are routed directly to inference nodes - no blockchain in the critical path.
- Instant Response: Results are returned to the user immediately with ultra-low latency.
- Asynchronous Settlement: After inference, the network handles payment and verification. Full nodes verify signatures and attestations (TEE or ZKML proofs) and record results to the ledger.
This architecture separates the fast path (inference) from the verification path (settlement), giving you the speed of web2 with the security and auditability of a decentralized network.
Below is a diagram describing the HACA components and workflow in the OpenGradient network:
