Inference Chains

Inference chains serve as the communication layer between user requests and model responses. Each inference chain is either verifiable or trusted. For verifiable chains, OPML or ZKML can be utilized to guarantee the compute was performed on the response using economics or cryptography accordingly. When a user provides a request, the message is passed to an inbox contract to the inference chain. After the request is fulfilled the inference chain posts the response to the outbox contract which triggers a callback on the applications contract.

An example of this dynamic can be seen in the following diagram.