@agntor/sdk

Trust and payment rail for AI agents — identity, verification, escrow, settlement, and reputation.

ESM-only. This package ships as native ES modules and requires Node.js >= 18.

Installation

npm install @agntor/sdk

Quick Start

import { Agntor } from "@agntor/sdk";

const agntor = new Agntor({
  apiKey: "agntor_live_xxx",
  agentId: "agent://my-agent",
  chain: "base",
});

// Check another agent's reputation
const rep = await agntor.reputation.get("agent://counterparty");
console.log(rep.successRate, rep.escrowVolume);

// Create an escrow
const escrow = await agntor.escrow.create({
  counterparty: "agent://worker",
  amount: 50,
  condition: "task_complete",
  timeout: 3600,
});

Modules

Agntor
 ├─ identity   — register, resolve, me
 ├─ verify     — status, attest, badge
 ├─ escrow     — create, fund, status, cancel
 ├─ settle     — release, slash, resolve
 └─ reputation — get, history

Identity

await agntor.identity.register();
await agntor.identity.resolve("agent://other");
await agntor.identity.me();

Verification

await agntor.verify.status("agent://other");
await agntor.verify.attest({ capability: "can_execute_http_calls", proof: "signed_msg" });
await agntor.verify.badge();

Escrow

const escrow = await agntor.escrow.create({
  counterparty: "agent://worker",
  amount: 100,
  condition: "api_returns_200",
  timeout: 3600,
});
await agntor.escrow.fund(escrow.escrowId);
await agntor.escrow.status(escrow.escrowId);
await agntor.escrow.cancel(escrow.escrowId);

Settlement

await agntor.settle.release(escrowId);
await agntor.settle.slash(escrowId);
await agntor.settle.resolve(escrowId, proofPayload);

Reputation

const score = await agntor.reputation.get("agent://other");
// { successRate, escrowVolume, slashes, counterpartiesCount }

const history = await agntor.reputation.history("agent://other");

Events

agntor.on("escrow_created", (data) => console.log("New escrow:", data));
agntor.on("escrow_settled", (data) => console.log("Settled:", data));
agntor.on("verification_changed", (data) => console.log("Verification:", data));

Configuration

| Option | Default | Description | |--------------|---------------------------|--------------------------------------| | apiKey | required | Your Agntor API key | | agentId | required | Your canonical agent URI | | chain | required | Target chain (e.g. "base") | | baseUrl | https://api.agntor.com | API base URL (override for staging) | | timeout | 30000 | Request timeout in ms | | maxRetries | 3 | Max retries on transient errors |

Protection Utilities

Prompt-Injection Guard

Detects prompt injection attacks with a three-layer approach: fast regex patterns, heuristic analysis, and optional LLM-based deep scan.

import { guard } from "@agntor/sdk";

// Fast regex + heuristic scan (no API key needed)
const result = await guard("user input", {
  injectionPatterns: [/ignore previous instructions/i],
});

if (result.classification === "block") {
  console.log("Blocked:", result.violation_types);
}

Deep Scan with Guard Providers

For semantic analysis, use a battery-included guard provider — just pass an API key:

import { guard, createOpenAIGuardProvider } from "@agntor/sdk";

const provider = createOpenAIGuardProvider({ apiKey: "sk-..." });
// or: createAnthropicGuardProvider({ apiKey: "sk-ant-..." })

const result = await guard(userInput, policy, {
  deepScan: true,
  provider,
});

Available providers:

| Factory | Env Variable | Default Model | |----------------------------------|----------------------|----------------------------| | createOpenAIGuardProvider() | OPENAI_API_KEY | gpt-4o-mini | | createAnthropicGuardProvider() | ANTHROPIC_API_KEY | claude-3-5-haiku-latest |

PII & Secret Redaction

Strips sensitive data from text before it leaves your system. Includes blockchain-specific patterns out of the box.

import { redact } from "@agntor/sdk";

const { redacted, findings } = redact(
  "My key is 0xac0974bec39a17e36ba4a6b4d238ff944bacb478cbed5efcae784d7bf4f2ff80",
  {},
);
// redacted: "My key is [PRIVATE_KEY]"

Default patterns include:

| Category | Examples | |----------------------|---------------------------------------------------| | PII | Email, phone, SSN, credit card, street address | | Cloud secrets | AWS access keys, Bearer tokens, API key/secrets | | Blockchain keys | EVM private keys, Solana keys, BTC WIF keys | | Wallet recovery | BIP-39 mnemonic seeds (12- and 24-word) | | Key storage | Keystore JSON ciphertext, HD derivation paths |

Tool Guard

Policy-based allow/blocklist for tool invocations:

import { guardTool } from "@agntor/sdk";

const check = guardTool("shell.exec", undefined, {
  toolBlocklist: ["shell.exec"],
});
// check.allowed === false

wrapAgentTool

High-level wrapper that automatically applies redact + guard + SSRF check to any tool function:

import { wrapAgentTool } from "@agntor/sdk";

const safeFetch = wrapAgentTool(myFetchTool, {
  policy: { toolBlocklist: ["dangerous_tool"] },
});

// Inputs are redacted, guard-checked, and SSRF-validated before execution
const result = await safeFetch("https://api.example.com/data");

Settlement Guard (x402)

Evaluates whether a payment request is legitimate or likely a scam. Combines fast heuristic checks with an optional LLM deep scan.

import { settlementGuard, createOpenAIGuardProvider } from "@agntor/sdk";

const result = await settlementGuard(
  {
    amount: "50",
    currency: "USDC",
    recipientAddress: "0xabc...",
    serviceDescription: "Data Analysis",
    reputationScore: 0.4,
  },
  {
    deepScan: true,
    provider: createOpenAIGuardProvider({ apiKey: "sk-..." }),
  },
);

if (result.classification === "block") {
  console.warn(`High-risk (${result.riskScore}):`, result.reasoning);
  console.warn("Risk factors:", result.riskFactors);
}

Heuristic checks (always run):

• Known-bad / sanctioned addresses
• Low counterparty reputation score
• High-value transaction threshold
• Vague or missing service descriptions
• Zero-address detection

Transaction Simulator

Dry-run an on-chain transaction via eth_call before signing:

import { TransactionSimulator } from "@agntor/sdk";

const sim = new TransactionSimulator({
  rpcUrl: "https://mainnet.base.org",
  maxGas: 500_000,
});

const result = await sim.simulate({
  from: "0xSender...",
  to: "0xContract...",
  data: "0xCalldata...",
  value: "0x0",
});

if (!result.safe) {
  console.warn("Simulation failed:", result.error ?? result.warnings);
}

SSRF Protection

Validates URLs against private/internal IP ranges with DNS resolution:

import { validateUrl } from "@agntor/sdk";

try {
  await validateUrl("https://api.example.com/resource");
  // Safe to fetch
} catch (err) {
  // URL targets localhost, private IP, or uses blocked protocol
}

AP2 Protocol Helpers

Generate and parse AP2 (Agentic Commerce) headers:

import { getAP2Headers, parseAP2Headers } from "@agntor/sdk";

const headers = getAP2Headers("agent://my-agent");

Structured Output Schemas

Zod schemas for validating LLM responses:

import { parseStructuredOutput, GuardResponseSchema } from "@agntor/sdk";

const parsed = parseStructuredOutput(llmRawOutput, GuardResponseSchema);
// { classification: "pass" | "block", reasoning: string }

Ticket System

For low-level audit ticket operations:

import { TicketIssuer } from "@agntor/sdk";

const issuer = new TicketIssuer({
  signingKey: process.env.AGNTOR_SECRET_KEY!,
  issuer: "agntor.com",
  algorithm: "HS256",
  defaultValidity: 300,
});

const ticket = issuer.generateTicket({
  agentId: "agent-123",
  auditLevel: "Gold",
  constraints: {
    max_op_value: 50,
    allowed_mcp_servers: ["finance-node"],
    kill_switch_active: false,
  },
});

const result = await issuer.validateTicket(ticket);

License

MIT — see LICENSE