Control
Check authority, policy and human approval before supported actions.
80 source modules; the private verifier build is excluded. Availability and test evidence are separate. No module is labeled hosted/live without recorded operational evidence.
| Module and job | Delivery | Verification evidence |
|---|---|---|
@zanii/brokerThe credential broker that closes Zanii's stated custody hole: the agent never holds raw API keys - the broker does, and it injects them ONLY into calls that produced a receipt first. Receipt-or-nothing, fail-closed (ledger unreachable = no call), credentials bound to URL origins so they cannot be exfiltrated to attacker-chosen hosts, secrets never in hash preimages or logged responses. | Usable source / SDK | not assessed Hosted: unverified |
@zanii/constitutionMachine-checkable behavioral constraints as signed policy - publish an agent's rules (forbid / require_approval / allow over scope patterns), stamp the constitution hash on every receipt via the existing manifest_hash field, then PROVE each action was governed: a forbidden action with the stamp is a provable violation, and a rulebook change mid-stream is visible. Generalizes the fta/sca walls into a framework. | Usable source / SDK | not assessed Hosted: unverified |
@zanii/coreZanii protocol: verifiable identity, signed action receipts, Merkle transparency-log proofs. Pure, no I/O. | Usable source / SDK | not assessed Hosted: unverified |
@zanii/guardStop attacks on AI agents before they land, and prove what was stopped: an inline guard before every action (canaries, tool pinning, exfiltration sequences, taint from untrusted content, rate limits), an evolution tracker that holds high-risk actions until a changed agent is re-tested, and a sandboxed red-team harness. Decisions are signed receipts. | Usable source / SDK | not assessed Hosted: unverified |
@zanii/kmsStop storing agent private keys in plaintext. Seal a key or identity under a passphrase with scrypt + AES-256-GCM, unseal only to sign. Zero dependencies. | Usable source / SDK | not assessed Hosted: unverified |
@zanii/policyDecide whether an agent action is allowed before it runs - amount caps, allow/deny lists, rate limits, and payload conditions that delegation scopes can't express. | Usable source / SDK | not assessed Hosted: unverified |
@zanii/runtimeZanii agent runtime: deterministic rails — the model proposes, tested code disposes. Scoped authority, earned status words, and a human gate for irreversible actions. | Usable source / SDK | not assessed Hosted: unverified |
@zanii/signerdKeep an agent's private key out of the agent's process. A local daemon unseals the key, signs receipts, owns the timestamp and the chain head; agents record through a socket with DaemonAgent. | Usable source / SDK | conformance double Tested artifact 0.2.0 Hosted: unverified |
@zanii/successionAgent lifecycle and key rotation with signed continuity - the owner attests 'B succeeds A', so reputation and CV history transfer verifiably instead of dying with a key. A compromise succession REQUIRES the compromise timestamp, bracketing the trusted window. Continuity of authority, not identity of code. | Usable source / SDK | not assessed Hosted: unverified |
@zanii/swarmN-party (3+) co-signed receipts for agent teams: genuine M-of-N Ed25519 threshold signatures over a canonical body (not a 2-party chain). Recorded on the ledger via POST /v1/swarm (SPEC §14) and verifiable offline. | Usable source / SDK | not assessed Hosted: unverified |
@zanii/wallsUAE vertical 'wall' presets as enforceable, auditable artifacts - scopes never granted + a policy config + a manifest_hash + a receipted eval. SCA trading ('education only, never buy this'), RERA, TDRA, Legal, Consumer, DIFC/ADGM. | Usable source / SDK | not assessed Hosted: unverified |
Blackbox and Certify
These are separate products. Their hosted gateway availability has not been verified by this deployment. Source packages and integration examples do not establish that a hosted workflow is live.