GuardDog

GuardDog - Autonomous AI Wallet Security Agent

Inspiration

Every week, DeFi users lose millions to honeypot tokens, rug pulls, and malicious contract approvals. The tools that exist like Revoke.cash, GoPlus, De.Fi all do the same thing: they send an alert. You still have to wake up, open your wallet, and act manually. If you're asleep or offline, you lose everything before you see the notification.

We built GuardDog because the gap isn't information, it's autonomous execution.

What It Does

GuardDog is the first autonomous wallet security agent that doesn't just alert but it acts. The agent monitors wallets 24/7 and automatically moves threatened tokens to a secure GuardianVault smart contract the moment a threat is detected. No human approval required.

The protection loop:

User enables protection once by signing one transaction
Agent scans every 5 minutes across BSC, Base, ETH, and opBNB
Threat detected (score ≥ 75) - agent executes batchProtectTokens() onchain
Tokens safe in GuardianVault; only the user can withdraw
Telegram alert sent with BSCScan transaction proof

How We Built It

Smart Contracts (Solidity + Hardhat)

GuardianVault.sol - secure token custody, guardian can move tokens IN but never OUT. Only the wallet owner can withdraw.
ThreatRegistry.sol - onchain community threat intelligence database, 0–100 aggregate scoring

Autonomous Agent (TypeScript + OpenClaw)

Runs 24/7 on Render - 30+ days zero downtime
Three parallel threat detection sources per scan:
- Community ThreatRegistry (onchain reports)
- Bytecode analysis (30+ dangerous function selectors)
- OKLink AML professional address screening
Gemini AI generates threat explanations and wallet risk assessments
MongoDB persists all scan history, threat detections, and protection events

Frontend (React + Vite + Tailwind)

Live Wallet Risk Score (0-100 with AI insight)
Bytecode Scanner; detects SELFDESTRUCT, hidden mint, blacklist in raw contract code
Pre-transaction AI screener; every tx screened before you sign
Community bounty system with onchain badges
Surge Watch; monitor new token launches for threats

AI Layer (Gemini via Google Cloud)

Wallet risk scoring with personalised 2-sentence AI insight
Threat explainer - plain English explanation of what each threat does
Transaction screener - SAFE / CAUTION / DANGER verdict before signing
Contract bytecode explainer - Claude reads raw code and explains it

Challenges We Faced

Trust model - designing a system where the guardian agent has zero ability to steal funds even if compromised. Solved at the smart contract level: protectTokens() can only move tokens IN, never OUT.
RPC reliability - BSC testnet RPC timeouts caused scan cycle freezes. Solved with per-network gas limits and direct read providers bypassing wallet origin restrictions.
Multi-chain complexity - deploying and verifying contracts across 4 testnets with consistent ABIs and network configurations.
Real-time threat detection - combining three data sources (community, bytecode, OKLink AML) into a single coherent risk score without false positives.

What We Learned

The hardest part of autonomous security is not detection but it is building a system users can trust enough to give it partial control. Every design decision in GuardianVault was made to minimise trust requirements: no admin keys, no upgrade paths, onchain enforcement of guardian limitations.