@shieldai-xyz/sdk

v0.1.0-alpha

Published

10 days ago

Runtime security SDK for autonomous AI agents — spend controls, anomaly detection, MCP gateway, EU AI Act compliance

0High
0Medium
0Low

shieldai2026

AI agents security web3 defi arbitrum mantle spend-control anomaly-detection eu-ai-act compliance mcp blockchain

@shieldai/sdk

Before your AI agent spends your money, someone should be watching.

Runtime security SDK for autonomous AI agents — spend controls, anomaly detection, MCP gateway, EU AI Act compliance.

Install

npm install @shieldai/sdk

Quickstart

import { ShieldAI } from '@shieldai/sdk'

const shield = new ShieldAI({
  agentWallet: '0xYourAgentWallet',
  dailyLimit: 50,        // USD
  whitelist: ['0x...'],  // approved counterparties
})

shield.watch() // that's it — agent is now monitored

Validate Transactions

const result = await shield.validate({
  from: agentWallet,
  to: '0xRecipient',
  amountUSD: 10,
  timestamp: Date.now()
})

if (result.allowed) {
  executeTransaction()
} else {
  console.log('BLOCKED:', result.reason)
  // reason: 'Daily limit exceeded' | 'Counterparty not whitelisted' | 'Anomaly detected' | ...
}

MCP Gateway

// Intercept every MCP tool call before execution
const mcpResult = await shield.interceptMCP({
  tool: 'uniswap_swap',
  params: { tokenIn: '0x...', amountIn: '1000000' }
})

if (mcpResult.allowed) {
  executeMCPCall()
}
// Blocks: high-velocity attacks, non-whitelisted contracts, critical tools

Agent Status

const status = await shield.getStatus()
// {
//   status: 'safe' | 'warn' | 'frozen',
//   anomalyScore: 0.03,
//   txToday: 12,
//   spentToday: 18.50,
//   remainingLimit: 31.50,
//   complianceScore: 100
// }

Freeze / Unfreeze

// Manual kill switch
await shield.freeze('Suspicious activity detected')

// Unfreeze when safe
shield.unfreeze()

// Auto-freeze on anomaly — happens automatically

Alerts

shield.onAlert((alert) => {
  console.log('ALERT:', alert.type, alert.severity)
  // Send to Slack, PagerDuty, Telegram, etc.
})

EU AI Act Compliance

// Generate compliance report (Art. 9, 13, 14, 72)
const report = shield.getComplianceReport()
console.log(report.score) // '83%'
console.log(report.articles) // array of compliance checks

// Export as JSON for regulatory inspection
const json = shield.exportCompliance()
fs.writeFileSync('compliance_report.json', json)

Advanced Config

const shield = new ShieldAI({
  agentWallet: '0x...',
  dailyLimit: 100,          // USD daily limit
  perTxLimit: 25,           // USD per transaction
  whitelist: ['0x...'],     // approved counterparties only
  anomalyThreshold: 0.8,    // 0-1, default 0.8
  alertWebhook: 'https://...', // POST alerts here
  rpcUrl: 'https://...',    // custom RPC
  euAiActLogging: true,     // EU AI Act compliance logging
  mcpGateway: true,         // MCP tool call interception
})

Security Features

| Feature | Description | |---------|-------------| | SpendGuard | Daily + per-tx limits enforced on every transaction | | Whitelist | Only approved counterparties can receive funds | | Anomaly Detection | Isolation Forest baseline — auto-freeze on anomaly | | MCP Gateway | Intercepts every tool call before execution | | Kill Switch | Manual freeze + auto-freeze on critical anomaly | | EU AI Act Logs | Auto-generates Art. 9, 13, 14, 72 compliance evidence |

On-Chain Contract

SpendGuard.sol is deployed and verified on Arbitrum One:

Contract: 0xE03C389DF391549E44c2aa807576c9eE2956C2d8
Network:  Arbitrum One
Verified: 2026-04-04 · Exact Match ✓
Block:    #448899709

View on Arbiscan ↗