ChainWhisperer — Onchain AI Copilot

Hyperion HyperHack Ideathon
1

Project Name

ChainWhisperer — Onchain AI Copilot

Problem Statement

Navigating smart contracts is risky and opaque for most Web3 users. Many sign transactions without fully understanding what the contract will do, potentially exposing themselves to scams, rug pulls, or unintended permissions. This lack of transparency especially affects non-technical users and newcomers to Web3.

Solution Overview

ChainWhisperer is a browser extension that acts as a real-time, onchain AI copilot. It detects smart contract interactions across dApps, Metis Testnet explorer, and wallets like MetaMask, providing natural language explanations, risk flags, and safe usage guidance before the user signs. By embedding a conversational AI assistant directly in the browsing experience, ChainWhisperer demystifies smart contract behavior and empowers users to interact with Web3 safely. Our approach combines onchain AI with real-time UI integration, creating a unique trust layer for contract interactions.

Project Description

ChainWhisperer brings contract safety and clarity into the hands of every Web3 user through a lightweight, intuitive browser extension. The extension automatically recognizes when users view a contract or prepare to sign a transaction, then analyzes the contract using Hyperion’s onchain AI capabilities and explains what it does in plain English.

Key features include:

  • Contract Auto-Detection on Metis Explorer, dApps, and wallets
  • Conversational Chat UI for asking questions about contract functions
  • Transaction Risk Warnings for dangerous patterns like unlimited approvals or owner-only withdrawals
  • Function Deep Dive showing expected inputs/outputs in simple terms

Built on the Hyperion testnet, ChainWhisperer uses parallel execution for fast analysis and can integrate with Alith, Hyperion’s AI co-agent, to deliver personalized safety insights. This project excites us because it democratizes smart contract comprehension and adds a human-friendly layer of defense to Web3—a critical step toward mainstream adoption.

Community Engagement Features

To drive adoption and stress-test functionality during the campaign, users will be rewarded for completing key in-extension actions:

Testable Features / Tasks

  • Open a dApp and receive a contract summary: +10 pts
  • Ask the AI a question about a contract function: +15 pts
  • Trigger a transaction and receive a risk warning: +20 pts
  • Submit feedback or flag a false positive/negative: +25 pts

Gamification Approach

  • Users earn XP and level up their “Whisperer Rank”
  • A leaderboard tracks top contributors
  • Special badges for early testers, bug reporters, and safety advocates

Onboarding Incentive

This system makes learning Web3 safer and fun, encouraging exploration while rewarding caution. It helps us collect valuable feedback while building a user-driven safety layer.

Getting Involved

Anyone can get involved by:

  • Contributing ideas, UX feedback, or contract samples on our this post
14 Likes
2

Hello @prakhar728 , Can the AI answer complex questions like “Can this contract drain my wallet?”

8 Likes
3

Yes, ideally it would.

Once the contract’s code is fetched, by feeding it to the LLM and giving the LLM a pattern of security risk contract I will enable the extension to answer these questions.

5 Likes
4

awesome vision, i’ve been think of something after reading it so i will ask to seek clarification..

  1. How does ChainWhisperer analyze smart contracts without access to verified source code?
    Are you decompiling bytecode or using AI pattern recognition to infer behavior?
  2. What mechanisms are in place to ensure the AI’s contract explanations are accurate and not misleading?
    Can users see a confidence level or audit trail for each explanation?
2 Likes
5

When ChainWhisperer flags a transaction as risky, how can I understand exactly what the risk is? And if I think the warning is incorrect, can I report it and earn points

1 Like
6

Appreciate the work you’re putting into this!

7 Likes
7

Interesting concept. How does ChainWhisperer plan to handle complex contracts that use delegate calls or proxies? Will it be able to interpret the logic accurately in real-time?

5 Likes
8 
1. How does ChainWhisperer analyze smart contracts without access to verified source code?
Are you decompiling bytecode or using AI pattern recognition to infer behavior?

Ans: I will be decompiling the bytecode first - this will the general function structure, definition and coniditonal statements. These decompiled statements, results will be fed into the LLM to review the definition and functionality and judge the security.

2. What mechanisms are in place to ensure the AI’s contract explanations are accurate and not misleading?
Can users see a confidence level or audit trail for each explanation?

Ans:

  1. The prompt for the LLM would include strict rules to define the Confidence score
  • A confidence level (Low / Medium / High)
  • A reason for the confidence level (e.g., “based on matched pattern” vs. “unknown logic”)

  1. For unverified: A warning appears — “Generated explanation is based on inferred bytecode structure. Proceed with caution.”

  2. User Flagging & Feedback Loop:

  • Users can flag inaccurate explanations
  • These are collected and used to fine-tune the AI prompts, reduce overconfidence, and improve heuristics
2 Likes
9

Yes. I will make sure to add this functionality.
Especially for unverified contracts the UI will have Warning text:
“Generated explanation is based on inferred bytecode structure. Proceed with caution.”
In case the user feels the explanation or inference is incorrect they can flag the response and get points for it.

3 Likes
10

I am yet to look into the data returned for Proxy contracts, specifically what solidity code is fetched. Based on that I’ll refine the process.

Thanks for raising this question, I didn’t think of this edge case yet.

2 Likes
11

Thanks, noted. The added warning makes sense.

3 Likes