Verifiable AI agents

A tamper-proof record of everything your AI agents do.

Zanii gives every AI agent a cryptographic identity and turns each action into a signed, hash-chained receipt — anchored to a public blockchain. Anyone can verify who did what, under whose authority. Without trusting us.

Open protocol · verify offline On-chain anchored · Base TypeScript & Python SDKs ERC-8004 compatible
// one line per tool — every call becomes a provable receipt
import { ZaniiAgent } from '@zanii/sdk';

const zanii = new ZaniiAgent({ serverUrl, agentDid, agentPrivateKey, delegation, apiKey });
const sendEmail = zanii.wrapTool('email.send', realSendEmail);
// sendEmail now works exactly as before — and is cryptographically recorded.
The problem

Your agents act. Nobody can prove what they did.

AI agents now touch real systems — calling tools, moving money, accessing customer data. But the only record is a log you could quietly edit. When an auditor, a customer, or a court asks “what did your agent actually do — and who authorized it?”, “trust our database” isn’t an answer. Zanii makes the answer cryptographic.

How it works

Proof, not promises.

1

Identity & delegation

Every agent gets a cryptographic identity, delegated by its owner with scoped, expiring authority. An agent can never do more than the human behind it — and it’s provable.

2

Signed receipts

Each action becomes an Ed25519-signed, hash-chained receipt in an append-only Merkle transparency log. Deleting, altering, or reordering any receipt is mathematically detectable.

3

Public anchoring

The log’s fingerprint is periodically written to a public blockchain, so not even Zanii can rewrite history. Verify any proof with the open SDK — no trust in us required.

Built in

Everything you need to make agents accountable.

Proof-of-action

Signed receipts attributable to one agent identity. Forgery needs the private key.

Scoped delegation

Owner-signed, expiring authority with one-click revocation. Least privilege, provable.

Merkle proofs

Inclusion & consistency proofs (RFC 6962) — the log is provably append-only.

On-chain anchoring

Tree heads anchored to Base, so history is immutable even to the operator.

Audit bundles

Self-contained, offline-verifiable exports mapped to EU AI Act & SOC 2 needs.

Cross-org receipts

Two agents co-sign one neutral proof. ERC-8004 compatible identities.

Trust no one — including us

Every claim is verifiable without us.

The protocol is open. Every proof is checkable offline with our SDK or your own code. The ledger’s fingerprint lives on a public blockchain. You never have to take our word for anything — that’s the point.

Open the live log and verify a proof →
Get started

Make your agents provable.