Zanii is a transparency log for AI-agent actions. It borrows the proven machinery of Certificate Transparency (RFC 6962) and applies it to what agents do, so accountability does not depend on trusting an operator’s database.
Every agent has a did:key identity, an Ed25519 keypair where the DID is the public key. The owner issues a delegation certificate granting scoped, expiring authority. An agent can never act beyond its delegation, and that boundary is cryptographically checkable.
For each action the agent builds a receipt: action, target, a hash of the payload (never the payload itself), the timestamp, and a pointer to the previous receipt. It signs the receipt with its key and submits it. Only a hash of your data ever leaves your systems.
Receipts are appended to a Merkle transparency log. The log publishes a Signed Tree Head, a signed fingerprint of its entire history. Inclusion proofs show a receipt is in the log; consistency proofs show the log only ever grew. Deleting, altering, or reordering anything breaks the math.
Periodically the Signed Tree Head root is committed to Base, a public blockchain. After that, not even Zanii can rewrite history without the change being detectable against an immutable external reference.
Anyone can pull a receipt with its inclusion proof and the tree head and verify it offline, with the open SDK or their own code. Export a self-contained audit bundle for any agent. No trust in us required, that is the entire point.