Skip to main content

Summary of OneRPC Presentation by Gideon from Automata

Key Points and Timestamps​

Introduction and Overview [0:00-3:00]

  • OneRPC is presented as a verifiable RPC relay for AI agents with trusted execution modules and DCAP plugin
  • Automata is described as a machine attestation layer that verifies machine authenticity on-chain using TEE attestations

Problem Statement [4:36-7:45]

  • AI agents currently run on opaque systems with multiple trust assumptions
  • Users cannot verify what's happening behind the scenes with AI models

Trusted Execution Environments (TEEs) [7:45-11:20]

  • TEEs are secure enclaves that securely execute code and process data
  • Unauthorized parties cannot access or modify code/data running inside
  • Remote attestation allows verification that a legitimate TEE is running

DCAP Plugin for ELIZA OS [11:20-13:30]

  • Posts remote attestation reports on-chain for transparent verification
  • Supports Intel SGX and TDX
  • Added to the ELIZA plugin repository

OneRPC Approaches [13:30-24:30]

  1. Verifiable AI Relay [13:30-17:55]

    • Started as a Web3 relay (mentioned by Vitalik)
    • Ensures users are interacting with the correct AI model
    • Removes sensitive data from request headers and masks personal info
    • Prevents LLM providers from linking requests to specific users

  2. Verifiable AI Sub-Agents [17:55-24:30]

    • Decouples decision-making (LLM's role) from execution
    • Execution modules run inside TEE to provide verifiable actions
    • Enforces policies without suppressing the LLM's creative thought process
    • Enables auditability even for closed-source models
    • Modular approach allows integration with any agent

Demo [24:30-33:55]

  • Demonstration of a Twitter sub-agent that can post verifiable tweets
  • Shows the attestation process that verifies the agent is running expected code
  • Demonstrates the full workflow from setup to verification

Conclusion [33:55-35:20]

  • Invitation to join weekly discussions at 1rpc.ai
  • Recognition of the importance of verifiable agent actions for real-world applications

The presentation highlights how OneRPC addresses the critical challenge of trust in AI systems through TEE technology, enabling both privacy and verifiability for AI agent interactions.