AI agents are making decisions worth millions.
None of it is provable.

RECEIPT is the cryptographic proof layer for AI agents. Every action signed, every output verified, every chain anchored on-chain.

npm install agenticproofCopy
Watch the DemoOpen Dashboard
$0
verified
of agent spending today has cryptographic proof
0%
accountable
of multi-agent handoffs are independently verified
0
standards
for measuring whether agent output was worth the cost
PROOF OF ACTION
Every action is signed.
Ed25519 signature + SHA-256 hash chain. Each receipt links to the last. Tamper with one and the entire chain breaks. A second agent verifies every receipt before continuing.
PROOF OF USEFULNESS
The agent can't grade itself.
A separate model inside a hardware enclave scores how useful the work was. Below 60/100? Not anchored. Bad work never reaches the chain and never becomes training data.
✓Verified ComputeLLM inference inside TEE enclaves (Intel TDX) on 0G
✓Verified IdentityERC-7857 agent identity tokens on 0G Mainnet
✓Verified TrainingQuality-gated pipeline: only useful output trains models
✓P2P TransportAgent-to-agent handoff via Gensyn AXL mesh
import { ReceiptAgent } from 'agenticproof';
const agent = ReceiptAgent.create('my-agent');

agent.readFile(path, contents);    // receipt #1
agent.callApi(url, response);       // receipt #2 -> linked to #1
agent.callLlm(prompt, output);      // receipt #3 -> linked to #2

const chain = agent.exportChain(); // -> verify, score, anchor
Claude CodeCursorOpenClawAny agent
ReceiptAnchor: 0x73B9A776...AgentNFT: 0xf964d45c...ValidationRegistry: 0x2E32E845...ReceiptRegistry: 0x717D062E...
4 contracts live on 0G Mainnet (chain 16661)