Use cases

Proof of what your AI did, and what happened to your data.

One umbrella: a signed receipt for every action your AI takes, and custody of every piece of data it touches. Find the category that sounds like you.

Who it’s for
Personal

Proof of what your AI assistant actually did.

You let Claude or ChatGPT send messages, book things, or run tools for you. The chat shows the talk. It never proves the action.

  • A signed receipt for each action your assistant takes through its tools: what it did, and when.
  • A record that lives outside the chat, that the assistant cannot quietly edit or delete.
  • Available now for developers through the Zanii MCP proxy. One-step setup for Claude Desktop, and ChatGPT support, are coming.
Enterprise

Put agents to work without losing the paper trail.

Your agents answer customers, issue refunds, and write to your systems at machine speed. The day one goes wrong, you must prove which agent did what, under whose permission.

  • A live, tamper-proof feed of every action, with a permission each agent cannot exceed.
  • A one-click audit bundle auditors and insurers accept (SOC 2, HIPAA, PCI).
  • A public receipt for any single action, verifiable by anyone in seconds.
Government

Deploy agents on sensitive work, and prove they stayed in bounds.

Agents touch citizen records and draft decisions about real people. When a decision is challenged, you need an independent record that it stayed in bounds.

  • Proof of least-privilege: the agent only touched what it was lawfully allowed to.
  • A portable evidence pack for oversight or a court, verifiable offline, no call home to a vendor.
  • The sensitive data never leaves your perimeter; only a fingerprint does.
Custody: what happened to your data

Proof covers the data too.

The cards above prove what the agent did. Custody proves what happened to the data it touched, in every category. Log in and check it yourself: every answer is computed in your browser from the public ledger, not taken on our word.

Checked live

It was never trained on

Your data was not used to train, fine-tune, or improve any model.

Checked live

No copy was kept

When the task ended, the data was destroyed. Nothing lingers in a log, a cache, or a backup.

Coming

Nobody read it

No human and no model saw the raw content, only that the work was done. Needs attested hardware.

Every answer is a verdict you can check: proven, not covered, failed, or cannot determine. Never a checkmark you take on faith.

Subject rightsData minimizationResidencyRetentionTraining provenanceData-flow mapOpen Zanii Custody →
Same receipt, whatever the job
PaymentsRefundsDevOpsTradingRecords & PIIRegulated workflows

Do not take our word for it.

Pull any receipt and verify the proof yourself.

See a receipt verify →Open the explorer