This project addresses the critical trust deficit in the burgeoning AI agent economy by providing a decentralized protocol for auditing and verifying AI agents. It targets AI developers seeking to prove their agents' capabilities and buyers needing to assess risk before hiring, offering verifiable proof of agent performance, security, and track record through a trust-first marketplace.

How It Works

AgentLens employs a multi-faceted approach combining on-chain audit scores derived from a Multi-Dimensional Dynamic Reputation Model (MDDRM), hardware-enforced Intel SGX Trusted Execution Environment (TEE) attestation for audit integrity, and Zero-Knowledge Proofs (ZK) for verifying audit calculations and agent fingerprints without exposing proprietary code. Audits are conducted in a Docker sandbox, with results cryptographically anchored on-chain, ensuring that trust is verifiable rather than merely claimed.

Quick Start & Requirements

• Primary install/run: Requires Node.js 20+, Docker & Docker Compose, and Rust for ZK circuit compilation. Local development involves starting a Polygon Edge local node, deploying smart contracts using Hardhat, and running the React frontend.
  • Install dependencies: cd contracts && npm install, cd ../sandbox && npm install, cd ../frontend && npm install.
  • Start local blockchain: cd infra/polygon-edge-local && docker compose up -d.
  • Deploy contracts: cd contracts && npx hardhat run scripts/deployV3.js --network edge_local.
  • Run frontend: cd frontend && npm run dev.
• Prerequisites: Node.js 20+, Docker & Docker Compose, Rust, Polygon Edge local node.
• Links: Documentation, Integration Guide, Live Demo: http://203.91.76.159/.

Highlighted Details

• Dimensional Risk Profiling: Agents are evaluated across 6 dimensions (Security, Task Execution, Cognitive, Environment, Engineering, Compliance) to generate a comprehensive risk profile.
• Intel SGX TEE Attestation: Audits run within hardware-isolated enclaves, with cryptographic proofs (MRENCLAVE) anchored on-chain to guarantee execution integrity.
• Zero-Knowledge Proof Verification: Utilizes circom and snarkjs (Groth16/BN128) to prove audit score calculations and agent identity fingerprints without revealing source code.
• Dynamic Reputation (MDDRM): On-chain reputation scores dynamically adjust based on audit results, user reviews, and time decay.
• Benchmark Validation: Tier-1 LLM agents like OpenAI GPT-4o and Anthropic Claude Sonnet 4.5 achieved perfect 100/100 audit scores, demonstrating vendor-agnostic auditing. Failed cases, such as a re-test of Zhipu-GLM4-Agent, highlight the protocol's ability to differentiate genuine pass/fail outcomes. Hardware-anchored execution is confirmed via SGX-DCAP attestations for all audits.

Maintenance & Community

The project is independently developed by a student seeking collaborators, researchers, and contributors passionate about Web3, AI Agents, ZK Proofs, and TEE. Contact is available via 3172791717@qq.com. The project adheres to a Contributor Code of Conduct.

Licensing & Compatibility

• License: GNU Affero General Public License v3.0 (AGPL-3.0) for community, research, and non-commercial use.
• Commercial Use: A commercial license is available upon contact for proprietary SaaS platforms or private enterprise deployments, offering an alternative to AGPL obligations.
• CLA: All contributors must sign a Contributor License Agreement (CLA).

Limitations & Caveats

The "Official Platform (Coming Soon)" indicates that the fully managed marketplace and hosted audit services are not yet deployed, requiring users to set up their own infrastructure for local development. The project's development by a single student may present a higher bus factor and potentially impact development velocity. The AGPL-3.0 license imposes strong copyleft requirements that necessitate careful consideration for commercial integration.