Ritual is an open AI infrastructure platform for providing and using decentralized, secure, and efficient computation services.

The Infernet Node is a lightweight off-chain client that fulfills compute workloads by listening for on-chain or off-chain requests and delivering workflow outputs and optional proofs via on-chain transactions or the off-chain API.

The Infernet SDK is a set of smart contracts allowing users to subscribe to off-chain compute workload outputs. The SDK enables smart contracts to request compute (either on- or off-chain via EIP-712 signed subscriptions), and receive a callback from an Infernet Node when the compute workload has been processed.

The Ritual Superchain is an execution layer to support AI-native operations. Ritual’s crypto-economic AI coprocessor supports computational tasks within the network to power a new generation of AI-enabled smart dApps.

A suite of sovereign modular execution layers, each containing specialized Stateful Precompiles (SPCs) overfit to different classes of arbitrary computation, mostly revolving around AI models, ranging from classical to foundation models.

As the core units of the Infernet SDK, subscriptions are a request (eitehr one-time or recurring) made by a compute consumer to an Infernet node, instructing it to process some compute. Users initiate compute subscriptions; nodes fulfill subscriptions.

The Infernet ML library is a Python SDK providing a set of easy-to-use abstractions for creating and deploying machine learning workflows. It provides an interface for data pre-processing, inference, and post-processing.

Jobs that are processed off-chain with the result being delivered off-chain.

Compute that is requested from a smart contract which is processed off-chain and the verifiable result being delivered back to the smart contract.

Requesting a compute via HTTP request with the result being delivered to a smart contract.

Guardians enable nodes to control what computations and outputs are acceptable to execute, providing nodes with full autonomy on controlling the ingress and egress of model objects.

Routers are responsible for optimally routing AI tasks (such as completion, chat, summarization) and embeddings for web2 applications to nodes.

The Ritual Superchain consists of different classes of nodes with different functionalities and resource requirements. This includes full and validator nodes, as well as Ritual specific nodes (proof nodes, model caching nodes, and privacy nodes).

Models are first-class, primitive objects within the Ritual protocol and are consumed by SPCs. Nodes can persist models locally or fetch them from a permissionless model registry.

Stateful precompiles (SPCs) are precompiles with state access. The Ritual protocol requires hyper-optimized operations which can efficiently compute a variety of AI model specific functionality. Some SPCs:

• can be implemented as a composition of other SPCs (i.e. finetuning & inference)
• can fully exploit various types of parallelism (i.e. embeddings)
• are sequential by construction

General Message Passing is used to allow applications (on any chain) to tap into the Superchain’s execution functionality via compact, bi-directional transports.

Portals use native smart contracts to enable data evaluation on source chains, before tapping the Ritual Superchain. As such, Portals minimize data sent between chains by optimizing for static analysis of AI model inputs which localizes compute to source chains.

Frenrug is an on-chain AI agent managing a portfolio of Frentech keys and is powered by Ritual.