How do I prove which AI model answered?

Ask the provider for a signed receipt naming the model. Zanii LLM writes one per call: the model id, the time and hashes of the request and the reply, signed by our key and anchored in a public Merkle log. Anyone you show the hash to can verify it themselves, offline, without trusting us or the log.

Why a log is not enough

A provider's own log is a claim about itself. It can be edited, and an auditor has no way to tell. A receipt is signed by a key the provider controls and published in an append-only log, so an alteration is detectable by arithmetic rather than by trust.

What the receipt contains

The account's key, the model that served the call, the time, and salted hashes of the request and the response. Not the text: the prompt and the answer never leave your systems, and the salt stays with us so a short prompt cannot be guessed from the log.

How to check one

Fetch the Merkle inclusion proof for the receipt hash from the ledger and verify it client-side with an open-source library. The verification does not call us at all.

Open the console Read the docs

More answers: Is there an OpenAI-compatible API with an audit trail? · What does an AI audit trail need to contain? · Can I buy AI inference billed in dirhams? · Which LLM APIs support Arabic properly?