@oceanalt/core

Client for OceanAlt's Trust Decision Core — the step before an AI agent's money moves.

• decide({ to, amountUsdc?, network? }) → allow | review | decline with reason codes and evidence (free, no key)
• endpoint(url) → is this payment endpoint on a phishing / scam list? (address lists cannot see impersonated endpoints)
• intentCheck({...}) → decode EVM calldata or a whole Solana transaction and compare with the declared payment (approve / SetAuthority dressed as a payment → mismatch; undecodable → unknown, never safe)
• verifyRequirements(paymentRequired) → offline Ed25519 verification of signed x402 payment requirements (tampered payTo in transit → digest_mismatch)
• guardOutbound(url, body) → secret-exfiltration scan (private keys, API keys, mnemonics) before a request leaves the agent
• ANOMALY → stable anomaly vocabulary (agent.anomaly.*) for your own monitoring

import { OceanAltClient } from "@oceanalt/core";
const client = new OceanAltClient({ onRecord: (r) => console.log(r) });
const d = await client.decide({ to: "0x93c67dabee533a9689a1e3a67201352b6b29dcd4", network: "eip155:8453" });
if (d.decision !== "allow") throw new Error(`hold: ${d.reason_code}`);

No keys, no custody, no telemetry unless you forward onRecord yourself. network accepts chain names or CAIP-2 ids. OceanAlt is not a licensed institution and not an AML/KYC provider; allow means "no risk signal found now", never "safe". Docs: https://oceanalt.com/en/api-docs · spec: https://oceanalt.com/en/rap