Arvyon

Arvyon is a ZK-verified autonomous AI agent framework for trustless policy enforcement on blockchain. It enables users to deploy AI agents that act on-chain autonomously, while every decision is paired with a zero-knowledge proof of policy compliance, stored as an immutable on-chain audit record.

---

The Problem

AI agents operating on blockchain can execute transactions, vote in DAOs, and manage assets autonomously. However, traditional frameworks require users to blindly trust off-chain logic, leaving no mathematical proof that the agent actually followed the rules it was given.

---

The Solution

Arvyon introduces a completely trustless, end-to-end framework built on four primitives:

• Transaction Intent Schema (TIS) — A structured declaration of what the agent intends to do and its reasoning, generated before every action.
• ZK Policy Proof — A zero-knowledge proof that the intended action complies with the user-defined policy (e.g., maximum trade limits), without revealing internal complex AI logic on-chain.
• Policy Decision Record (PDR) — An on-chain immutable log of every agent decision, queryable by anyone, forever.
• Executor Contract — A smart contract that mathematically verifies the ZK Proof on-chain. If valid, it executes the real transaction (e.g., swapping tokens on Uniswap); if invalid, the transaction is perfectly blocked.

---

Tech Stack

Layer	Technology
AI Agent Core	Python, LangGraph, Ollama (Llama 3.1)
ZK Proof Engine	Circom, snarkjs (Groth16)
Smart Contracts	Solidity, Hardhat
Frontend dApp	Next.js, React, Tailwind CSS, ethers.js

---

Project Architecture

Arvyon operates as a continuous pipeline bridging off-chain AI reasoning with on-chain verification.

1. Setup & Policy Registration in Dapp

A user interacts with the Dapp to create a policy for their AI agent. This policy is hashed and stored on the Sepolia testnet via the PolicyRegistry smart contract.

2. The Agent Loop (LangGraph Pipeline)

The core autonomous loop :

• Observe: The agent fetches the registered policy bounds and current market data (via CoinGecko).
• Decide: The agent asks an LLM (e.g., Llama 3.1) to reason about the market data and propose an action.
• Intent & ZK Proof: The agent structures the decision into a TIS and generates a mathematical Zero-Knowledge (ZK) proof using Circom/snarkjs that proves the action complies with the policy bounds.
• Submit: The agent submits the decision and ZK proof to the blockchain.

3. On-Chain Verification (Smart Contracts)

When submitted to Sepolia:

• The Executor contract runs an on-chain verification of the Groth16 ZK proof.
• If mathematically valid, the action is approved and the transaction is forwarded to the target protocol (DeFi execution).
• The decision is permanently logged into the PDRLogger.

4. Audit & Transparency (Dashboard)

The Next.js dApp listens to the blockchain and displays a live, human-readable Dashboard. Users can watch a live stream of the agent's thought process (via the Live Agent Terminal) and see an immutable feed of ZK Verified executions.

---

Built With

• circom
• hardhat
• langgraph
• nextjs
• solidity