Verifiable Computation Proofs

Background: The functionality of agent memory certification relies on Trusted Execution Environments (TEEs) and model API providers faithfully executing computational effort for API responses.

Question / Future Work: Investigating mechanisms to prove verifiable computational effort when the model API provider might have an incentive to collude with market participants, potentially necessitating the use of more expensive or complex solutions like Zero-Knowledge Proofs (ZKPs) or executing LLM inference inside GPU TEEs.