agenticproof

Cryptographic proof layer for AI agent work. Ed25519-signed, hash-linked receipt chains.

Every action an AI agent takes -- file reads, API calls, LLM inferences, decisions -- gets a signed receipt. Receipts link into a tamper-evident chain that can be verified, handed off to other agents, and anchored on-chain.

Install

npm install agenticproof

Quick Start

import { ReceiptAgent, verifyChain } from 'agenticproof';

const agent = ReceiptAgent.create('researcher');

agent.callApi('https://api.example.com/data', '{"results": 42}');
agent.callLlm('Summarize the results', 'The data shows 42 results.');
agent.decide('Results are significant', 'Proceed with analysis');
agent.produceOutput('Final report', 'Analysis complete: 42 confirmed.');

const chain = agent.getChain();
const results = verifyChain(agent.getReceipts(), agent.getPublicKey());
console.log('Chain valid:', results.every(r => r.valid)); // true

What It Does

• Receipt chain -- every agent action produces a signed receipt with input/output hashes
• Ed25519 signatures -- each receipt is signed with the agent's private key
• Hash linking -- each receipt references the previous receipt's ID, forming a tamper-evident chain
• Verification -- verify individual receipts or entire chains against a public key
• Multi-agent handoff -- bundle a chain and hand it to another agent with full provenance
• Action types -- file reads, API calls, LLM calls, decisions, tool calls, messages, usefulness reviews
• On-chain anchoring -- anchor receipt chain root hashes to 0G Mainnet

API Reference

Core Exports

import {
  ReceiptAgent,       // High-level agent that records and signs receipts
  ReceiptChain,       // Ordered chain of receipts with hash linking
  createReceipt,      // Low-level receipt creation
  verifyReceipt,      // Verify a single receipt
  verifyChain,        // Verify an entire receipt chain
  hash,               // SHA-256 hash utility
  generateKeyPair,    // Generate Ed25519 key pair
  sign,               // Sign a message
  verify,             // Verify a signature
  publicKeyToHex,     // Convert public key to hex string
} from 'agenticproof';

ReceiptAgent Methods

| Method | Description | |--------|-------------| | ReceiptAgent.create(name) | Create a named agent with fresh keys | | ReceiptAgent.continueFrom(receipts) | Resume from a verified receipt chain | | agent.readFile(path, content) | Record a file read | | agent.callApi(endpoint, response) | Record an API call | | agent.callLlm(prompt, response) | Record an LLM inference | | agent.decide(reasoning, decision) | Record a decision | | agent.produceOutput(desc, output) | Record a produced output | | agent.toolCall(name, input) | Record a tool invocation | | agent.toolResult(name, result) | Record a tool result | | agent.messageSend(recipient, content) | Record a message sent | | agent.reviewUsefulness(summary, result) | Record a usefulness review (quality scoring) | | agent.verifyOwnChain() | Verify the agent's chain integrity | | agent.getReceipts() | Get all receipts in the chain | | agent.getChain() | Get the underlying ReceiptChain |

Integration Exports

import { AxlTransport } from 'agenticproof/integrations/axl';
import { anchorOnChain } from 'agenticproof/integrations/0g-chain';
import { uploadToStorage } from 'agenticproof/integrations/0g-storage';

Multi-Agent Handoff

Agents can hand off work with full cryptographic provenance using ReceiptChain.toHandoffBundle() and Gensyn AXL for P2P transport.

import { ReceiptAgent } from 'agenticproof';
import { AxlTransport } from 'agenticproof/integrations/axl';

// Agent A: researcher
const researcher = ReceiptAgent.create('researcher');
researcher.callApi('https://api.data.org/fetch', '{"items": 100}');
researcher.callLlm('Analyze trends', 'Three key trends identified.');

const bundle = researcher.getChain().toHandoffBundle('researcher');
const axl = new AxlTransport();
const peers = await axl.discoverPeers();
await axl.sendHandoff(peers[0], researcher.getReceipts(), researcher.getPublicKey(), bundle);

// Agent B: builder (receives and continues the chain)
const incoming = await axl.waitForHandoff();
const builder = ReceiptAgent.continueFrom(incoming.bundle.receipts);
builder.decide('Implement trend #1', 'Building feature based on research.');

Claude Code Hooks

Generate hook configuration to automatically produce receipts for every Claude Code session:

npx agenticproof init --claude-code

OpenClaw Plugin

RECEIPT ships with a native OpenClaw plugin for automated receipt generation in OpenClaw-managed agents. See the packages/openclaw-plugin directory in the monorepo.

On-Chain Anchoring

Anchor receipt chain root hashes to 0G Mainnet for permanent, verifiable proof of agent work:

import { anchorOnChain } from 'agenticproof/integrations/0g-chain';

const result = await anchorOnChain(rootHash, storageRootHash, {
  rpc: 'https://evmrpc.0g.ai',
  contractAddress: '0x73B9A7768679B154D7E1eC5F2570a622A3b49651',
  privateKey: process.env.PRIVATE_KEY,
  chainId: 16661,
  usefulnessScore: 82,
});
console.log('Anchored:', result.txHash);

License

MIT