Ritual Knowledge Hub

Technical Architecture

Breakdown of Ritual’s Tech Stack

Ritual Knowledge Hub

Architecture

Exploring Ritual’s Tech Stack

Ritual’s technical architecture consists of different layers. Let’s have a look how its tech stack is organized and which components are part of it.

Ritual Superchain

AI coprocessor for blockchains

Infernet

Ritual’s first production implementation

Architecture Overview

Ritual’s stack is organized into the Ritual Coprocessor – consisting of GPUs, a VM, and the node set – as well as Infernet, the protocol’s first production implication.

Ritual Superchain

Ritual’s crypto-economic AI coprocessor will power a new generation of AI-enabled smart DApps. As an integral part of the Ritual architecture, the coprocessor supports computational tasks within the network.

  • Handling off-chain computation requests
  • Delivering results back to
    the on-chain environment

  • Specifically for tasks related to machine learning inference

Ritual Coprocessor

Elements of the Ritual Superchain

Routers

Ritual Supercomputer

Stateful Precompiles (SPCs)

Ritual Supercomputer

General Message Passing (GMP)

Ritual Supercomputer

Portals

Ritual Supercomputer

Guardians

Ritual Supercomputer

Modular Stateful Precompiles

Ritual Supercomputer

Shared Sequencer

Ritual Supercomputer

DA Layer

Ritual Supercomputer

Model Storage

Ritual Supercomputer

Permanent Storage

Ritual Supercomputer

Restaking

Ritual Supercomputer

Ritual Coprocessor

Superchain Elements in Detail

Routers

Routers are being presented at tasks such as completion, chat, summarization, and web3 app embeddings. An incentive structure is in place to optimally route requests to Ritual nodes. Settings for service quality, privacy, computational integrity, and cost can be defined by the user.

Stateful Precompiles (SPCs)

In order to efficiently compute AI models, Ritual requires hyper-optimized operations for which SPCs (precompiles with state access) are used. Some SPCs can be implemented as composition of other SPCs (for finetuning & inference). Some SPCs will be capable of fully exploiting various types of parallelism while others are sequential by construction.

General Message Passing (GMP)

Compact, bi-directional General Message Passing transports are used to allow applications on any chain to access the Superchain’s execution functionality.

Portals

Portals use smart contracts to allow data evaluation on source chains before using the Ritual Superchain. In doing so, Portals minimizing the amount of data sent by optimizing for static analysis of AI model inputs.

Guardians

Guardians ensure that nodes have full control over the ingress and egress of model objects, allowing nodes to control what traffic and outputs are acceptable to execute.

Modular Stateful Precompiles

Seamless integration of SPCs into other VMs from the community is made possible through their special architecture.

Shared Sequencer

By abstracting sequencing from execution layers, Ritual allows transaction atomicity across layers. This functionality will be powered by a shared sequencer and pooled security.

DA Layer

Users are able to choose their preferred data availability layer thanks to Ritual’s modular design.

Model Storage

Data objects are first-class citizens in the Ritual Superchain (by using models). With its permissionless, censorship-resistant storage layer, Ritual allows nodes to dynamically on- & off-load ephemeral models.

Restaking

The security of the Ritual Superchain will be bootstrapped by  staked security from other chains (Ethereum Mainnet via Eigenlayer). Later on, organic security from within the own ecosystem will be used to a larger degree.

Permanent Storage

Some general purpose artifacts and metadata require persistent storage so that nodes can deterministically reconstruct the computation of models.

Infernet

Infernet is the first production implementation of the Ritual protocol. Smart contracts can request Infernet to compute an output with on-chain proof.

  • Coordinator propagates requests
  • Forwards request to node capable of doing the computation
What is Infernet?

Ritual’s Infernet is a lightweight library for bridging off-chain compute on-chain. With the help of Infernet, smart contract developers can request computation to be executed off-chain by Infernet Nodes and delivered to their on-chain smart contracts via the Infernet SDK.

What is Infernet’s Primary Application?

Infernet’s primary application is in servicing ML Inference workloads. With Infernet, users can do the following:

Users can build and host ML models, deploy them to Infernet nodes, and create subscriptions to consume inference outputs and optional succinct execution proofs (for classical ML models that fit within a circuit) from smart contracts on-chain.

  • Build and host ML models
  • Deploy ML models to Infernet nodes
  • Create subscriptions to consume inference outputs (and optional execution proofs) from smart contracts on-chain.

What are Infernet’s Use Cases?

  • Host arbitrary compute workloads via Ritual Core Workflows
  • Be consumed via easy interfaces on-chain (Infernet SDK)
  • Be used for on-chain and off-chain compute processing via Infernet Node
  • Be permisionlessly accessed by any builder, smart contract developer, or node operator

Go Back

Ritual Academy

Up Next