Quickstart

Record your agent's input, output and model metadata through one API. Return a verification link with every result.

1. Create a key

Name a key for your service. Copy it once and store it in PROOFAPI_API_KEY on your server.

macOS / Linux

export PROOFAPI_API_KEY="YOUR_KEY"

PowerShell

$env:PROOFAPI_API_KEY="YOUR_KEY"

2. Install the SDK

The proofapi package on npm. Install it in your server application, not in the browser.

Terminal · Node.js 20+ · ESM

npm install proofapi

3. Record your first result

Use the response from your existing AI provider. ProofAPI records it; it doesn't generate another answer.

TypeScript

import { ProofAPI } from 'proofapi';

const proofapi = new ProofAPI({
  apiKey: process.env.PROOFAPI_API_KEY,
});

// input and answer come from your application
const record = await proofapi.seal({
  input: 'Classify support ticket #42',
  output: 'billing',
  model: 'your-model',
  label: 'support-agent',
});

console.log(record.status, record.certificateUrl);

Text mode stores input/output in downloadable evidence. Use sample data here; use hash-only mode for confidential content.

Try the Playground

4. Check the record

Open certificateUrl or call verify(id). ANCHORED means the seal was recorded; PENDING_CHAIN means recording is incomplete.

TypeScript

const verification = await proofapi.verify(record.id);
console.log(verification.result.status);
// VERIFIED | FAILED | NOT_ON_CHAIN

const evidence = await proofapi.evidence(record.id);

Keep private content in your infrastructure

Send only locally computed SHA-256 fingerprints. Keep the original bytes and salt so you can prove the content later.

TypeScript

import { createHash, randomBytes } from 'node:crypto';
const salt = randomBytes(32); // keep the salt and original bytes
const hash = (text: string) => createHash('sha256')
  .update(Buffer.concat([salt, Buffer.from(text)])).digest('hex');
const record = await proofapi.sealHashes({
  inputHash: hash(input), outputHash: hash(answer),
  metadataHash: hash(JSON.stringify({ model: 'your-model' })),
});

Connect an AI agent (MCP)

Agents in Claude, Cursor and other MCP clients can seal their own answers without code. Add the server to the client's MCP settings; it gives the agent four tools: seal_answer, seal_hashes, verify_record and get_evidence.

mcp.json

{
  "mcpServers": {
    "proofapi": {
      "command": "npx",
      "args": ["-y", "-p", "proofapi", "proofapi-mcp"],
      "env": { "PROOFAPI_API_KEY": "pk_live_..." }
    }
  }
}

Without a key, anonymous demo limits apply. Agents and tools can also read the API description at /llms.txt and /openapi.json.

API reference

POST /api/v1/seal: seal({ input, output, model?, label? })
POST /api/proofs/hashes: sealHashes({ input_hash, output_hash, metadata_hash })
GET /api/proofs/{id}/verify: verify(id)
GET /api/proofs/{id}/evidence: evidence(id)

REST · bash

curl -X POST https://proofapi.vercel.app/api/v1/seal -H "Authorization: Bearer $PROOFAPI_API_KEY" -H "Content-Type: application/json" -d '{"input":"Classify ticket #42","output":"billing","model":"your-model"}'

Optional: document review demo

Status and errors

Beta runs on Solana devnet. Local memory mode simulates recording. A seal proves integrity, not AI accuracy or the identity of the declared model.

TypeScript

import { ProofAPIError } from 'proofapi';
try {
  const record = await proofapi.seal({ input, output: answer });
  if (record.status === 'PENDING_CHAIN') {
    // Record saved; wait for anchoring before relying on it.
  }
} catch (error) {
  if (error instanceof ProofAPIError) {
    console.error(error.status, error.code);
    // 400 validation, 401 key, 429 limit, 500/503 service
  }
  // No automatic retries for writes: avoid duplicate records.
}

Configuration

TypeScript · local development

const proofapi = new ProofAPI({
  apiKey: process.env.PROOFAPI_API_KEY,
  baseURL: 'http://127.0.0.1:3000', // local server only
  timeoutMs: 65000,
});