Agent
Guardrails.
Signed capability manifests for AI agents. Spending caps, receiver allowlists, cooldown windows — provably enforced by the protocol. No unsigned policy executes.
Features.
Signed Capability Manifests
Policies are signed by the agent owner. The signature is verifiable on-chain. No unsigned policy can be enforced.
Spending Caps
Each policy specifies a maximum spend per transaction, per day, or per epoch. Exceeding the cap fails at settlement.
Receiver Allowlists
Agents can only pay addresses that appear in their policy manifest. One-time addresses can be whitelisted per transaction.
Cooldown Windows
Policies can enforce minimum time between transactions. During cooldown, settlements are rejected even with a valid proof.
Expiry + Revocation
Policies expire after a UNIX timestamp. The owner can revoke a policy at any time — revoked policies cannot settle.
Hierarchical Policies
Policies can delegate to sub-policies. A top-level policy can grant limited authority to sub-agents without giving full access.
Signed manifests.
Every policy is signed by the agent owner's wallet. The signature is verified on-chain before any settlement is accepted. A policy without a valid signature cannot execute.
Policies can be rotated, revoked, or upgraded without changing the agent's address. The protocol always enforces the current active policy.