Ritual Chain Brings AI Inference and Persistent Agents Into an EVM Layer 1

Ritual Chain is an EVM-compatible Layer 1 designed to let software call AI models, perform private computation and schedule its own future activity from within on-chain applications. The launch matters because it puts those capabilities in the chain’s execution layer, where contracts can use them alongside asset transfers, rather than relying solely on services outside the network.

Ritual, a lab for autonomous intelligence founded by former Polychain Capital partners Niraj Pant and Akilesh Potti, says the network is intended to support agents that persist, control funds and interact with external services over extended periods. Its public testnet has been live since June and has processed more than 40 million transactions from both agents and people. Ritual reports block times below 300 milliseconds and says the chain includes 15 native precompiles for functions such as language-model inference, web access and autonomous scheduling.

Delegated computation changes the workload

AI inference can be expensive to repeat, particularly if every validator must independently run the same model computation to reach agreement. Ritual’s execution and verification design assigns computationally intensive work to specialized nodes, then has the wider network verify the result. The company describes this as “execute-once, verify-many-times.” That structure aims to support larger workloads without requiring every validator to reproduce them.

The design shifts attention to how delegated work is checked and what evidence validators can use to accept a result. Ritual’s launch integrations include Phala Network, which supplies GPU-backed trusted execution environments for confidential inference. The source says this hardware attests to what it ran. Such environments are intended to shield prompts, keys and strategy data from compute operators, though the confidentiality claim depends on the execution environment and its attestation process.

Ritual also says Zellic audited the protocol. An audit is one input into understanding a system’s security; it does not by itself establish that every component or future application is free of defects. For developers, the practical questions include how delegated results are verified, how failures are handled, and what protections apply when an agent is given access to funds or private inputs.

Agents designed to outlast a session

Many AI agents operate only while a person’s session and infrastructure remain available. That model can suit interactive tools, but it creates constraints for software expected to run recurring services or manage funds over time: the operator may shut down the process, revoke an API key or disappear.

Ritual’s proposed model gives an agent an on-chain identity and state that persist beyond the machine or person that deployed it. A heartbeat mechanism can restart an agent if it stops, within the permissions and budget assigned to it, without a separate keeper network or off-chain scheduling job. The agent can also use a non-custodial wallet to hold and spend funds, price services and settle transactions without a human signing each action.

That autonomy makes permission design central. A restart mechanism can help a service continue running, but an agent’s permitted actions and spending limit determine what it can do after a failure or unexpected result. Private execution can reduce exposure of sensitive inputs to node operators, while leaving developers responsible for deciding what information to provide and what authority to grant.

Niraj Pant, Ritual’s co-founder, said: “As we move toward AGI, intelligence is beginning to separate from human agency and enter the economy on its own terms. It can hold assets, make commitments, transact, and persist independently.” He added that the network is intended to provide infrastructure for these actors to operate. The remarks describe Ritual’s ambition; the launch details establish the available mechanisms, while adoption and agent performance remain to be seen.

Launch integrations and developer access

Ritual Chain launches with integrations from Phala Network, Nous Research, LayerZero and Stargate. LayerZero provides cross-chain messaging, while Stargate enables USDC transfers between Ritual Chain and other networks from launch. Nous Research is bringing its Hermes application to the chain.

Ritual’s approach differs from adding model access through an external oracle: contracts can call a language model, retrieve web data, request private computation or schedule later execution through native precompiles. Developers can deploy standard EVM contracts to use these functions, according to the company, without adopting new tooling.

That lowers the friction for EVM developers to experiment, but it also puts new demands on application design. A contract that relies on model output or web data must account for the source and limits of that information, as well as what happens when an inference call fails or returns an unexpected response. Native access changes where these capabilities sit in the stack; it does not make their outputs inherently dependable.

Funding and next steps

Ritual has raised more than $30 million from Archetype and Polychain Capital. Its advisors include NEAR co-founder Illia Polosukhin, Gauntlet founder Tarun Chitra and BitMEX co-founder Arthur Hayes. Archetype led the company’s 2023 round. Archetype founder and general partner Ash Egan said the launch brings programmable value and programmable intelligence onto the same network, and noted that the company’s current direction was not available when Archetype led that round.

Developers can find documentation at docs.ritualfoundation.org and inspect network activity through the Ritual Chain explorer. The next meaningful indicators will be how applications use the native functions, how the delegated verification model performs under real workloads, and whether agents can operate within clear limits over long periods.