Supernet
  • Supernet|Guides
  • Multi-Agent Collaboration
    • Overview
    • Best Route
  • Decentralized GPU System
  • Whitepaper
    • Whitepaper of Supernet
  • distributed network
    • Supernet | Introduction
    • AI Operating System
      • Transformer Decoder Architecture
      • LLM OS
    • AI OS Design
      • Decentralize Validator
      • Decentralize Storage
    • TEE-Enhanced Design
    • AVS of Agent
    • Fault Tolerance
  • Product
    • Supernet | Intelligent Node
    • Supernet | AI OS SDK
    • Supernet | AI Explorer
    • Supernet | Agent Studio
    • Supernet | Public API
Powered by GitBook
On this page
  • Core Design
  • Workflow
  • Advantages
  • Applicable Scenarios
  1. distributed network

AVS of Agent

Before reading this, you should understand the functionality of AVS in EigenLayer.

PreviousTEE-Enhanced DesignNextFault Tolerance

Last updated 5 months ago

To verify the authenticity of agent services executed within a distributed network and ensure the compliance and transparency of node behavior.

Core Design

Task Validation Mechanism

  • Task Distribution: AGENT tasks (e.g., model execution, decision-making processes) are distributed by on-chain smart contracts or task aggregators to multiple nodes.

  • Node Execution: Each node executes the assigned task independently and generates results.

  • Result Verification: The results are validated on-chain through cryptographic proofs, such as hash verification or zero-knowledge proofs (ZKPs), ensuring the nodes genuinely executed the tasks.

Validation Layer: Result Consistency and Behavior Monitoring

  • Consensus Mechanism: Results are validated using majority consensus among nodes. Signatures (e.g., BLS) are aggregated to ensure efficient verification.

  • Proof of Execution: Nodes provide execution proofs (e.g., logs or ZK-SNARKs) as evidence of task completion. These proofs are submitted on-chain for transparency.

  • Random Sampling: Random nodes are periodically selected to re-execute tasks, ensuring result authenticity and catching fraudulent behavior.

Reward and Penalty System

  • Reward Distribution: Nodes receive rewards based on task completion and verification outcomes, using an ERC-20 token model.

  • Slashing Mechanism: Nodes that fail to execute tasks, submit fraudulent results, or violate compliance standards will have their stakes slashed.

Signature Aggregation for Efficiency

  • BLS Signatures: Task results and execution proofs use BLS signatures for aggregation, reducing on-chain storage costs and gas consumption while maintaining verification efficiency.

Workflow

Task Submission (On-Chain)

  • AVS Consumers submit AGENT tasks to the AVS smart contract for execution.

Task Execution (Off-Chain)

  • Nodes receive and execute the tasks, generating outputs and proofs of execution.

On-Chain Validation

  • Aggregated BLS signatures or ZK proofs are submitted to the AVS contract.

  • Execution results and node behaviors are verified on-chain.

Rewards and Penalties

  • Rewards are distributed to compliant and successful nodes.

  • Malicious nodes face penalties, including slashing of their stakes.

Advantages

Authenticity

Combines cryptographic verification and consensus to ensure tasks are genuinely executed.

Decentralization

Prevents centralized control by distributing tasks to multiple nodes.

Economic Incentives

Encourages reliable node behavior through rewards while penalizing malicious actions.

Efficient Verification

Signature aggregation reduces gas costs and on-chain computation.

Applicable Scenarios

Task Execution Validation

Verify distributed nodes completing AGENT tasks, such as model decision-making or service computations.

Data Integrity

Ensure nodes process and validate accurate, untampered data.

Service Quality Monitoring

Monitor node uptime, performance, and compliance with AVS standards.

AVS workflow in restaking.