@agentoracle/verify

Trust verification SDK for AI agents. Verify claims before your agent acts on them.

Install

npm install @agentoracle/verify

Quick Start

import { verify } from "@agentoracle/verify";

// One line — verify any claim
const result = await verify("Bitcoin was created by Satoshi Nakamoto in 2009");
console.log(result.evaluation.overall_confidence); // 0.97
console.log(result.evaluation.recommendation);     // "act"

Full Client

import { AgentOracle } from "@agentoracle/verify";

const oracle = new AgentOracle();

// Evaluate claims
const eval = await oracle.evaluate("SKALE charges $0.05 per transaction");
// → { evaluation_id, evaluation: { confidence, claims[], recommendation } }

// Free preview
const preview = await oracle.preview("What is x402?");

// Database stats
const stats = await oracle.fingerprints();
// → { database_stats: { total_keys: 374, feedback_count: 0 } }

// Submit feedback
await oracle.feedback("eval_123", "accurate");

Verification Gate Middleware

Embed trust verification into any Express API. Incoming data gets auto-verified before your handler processes it.

import express from "express";
import { createVerificationGate } from "@agentoracle/verify/middleware";

const app = express();
app.use(express.json());

// Block submissions below 0.6 confidence
app.post("/api/submit",
  createVerificationGate({ minConfidence: 0.6 }),
  (req, res) => {
    // Only reaches here if content passes verification
    console.log(req.verification.confidence); // >= 0.6
    res.json({ accepted: true, verification: req.verification });
  }
);

// Annotate but don't block
app.post("/api/ingest",
  createVerificationGate({ block: false }),
  (req, res) => {
    // Handler decides what to do with low confidence
    if (req.verification.confidence < 0.3) {
      return res.status(422).json({ error: "Unverifiable" });
    }
    res.json({ processed: true });
  }
);

Middleware Options

| Option | Default | Description | |--------|---------|-------------| | minConfidence | 0.5 | Minimum confidence to pass (0-1) | | field | "body" | What to verify: "body", "query", or a body key | | block | true | Block requests below threshold, or just annotate | | baseUrl | agentoracle.co | API endpoint | | extractContent | — | Custom function (req) => string |

Standalone Verification

For non-Express environments:

import { verifyContent } from "@agentoracle/verify/middleware";

const { pass, confidence, recommendation } = await verifyContent(
  "The earth is flat",
  { minConfidence: 0.7 }
);
// pass: false, confidence: 0.03, recommendation: "reject"

API Methods

| Method | Description | Cost | |--------|-------------|------| | evaluate(content) | Per-claim verification with confidence scoring | $0.02 | | preview(query) | Free truncated research preview | Free | | research(query) | Full research with sources | $0.02 | | deepResearch(query) | Extended analysis via Sonar Pro | $0.10 | | feedback(id, outcome) | Report evaluation accuracy | Free | | reputation() | Source reputation scores | Free | | fingerprints() | Database stats (374+ claims) | Free | | health() | API status and endpoints | Free | | manifest() | x402 payment manifest | Free |

Payment (x402)

Paid methods return a 402 error with payment instructions:

try {
  const result = await oracle.research("What is Stellar?");
} catch (err) {
  if (err.status === 402) {
    console.log(err.challenge);  // x402 payment challenge
    console.log(err.networks);   // ["eip155:8453", "stellar:testnet", ...]
    // Sign payment and retry with paymentHeader option
  }
}

Supports: Base (USDC), SKALE (gasless), Stellar (sponsored fees).

Links

• Website: agentoracle.co
• Trust Docs: agentoracle.co/trust
• GitHub: TKCollective/x402-research-skill
• Fingerprints: agentoracle.co/fingerprints

License

MIT