Replay protection

An agent that transacts for you while you sleep is powerful — and dangerous if an instruction it signed can be captured and replayed later. Omyra closes that vector at the proof layer.

Time-bounded validity

Every action binds an expiration window into the public inputs of its proof. A verifier checks the current slot against that window: outside it, the proof simply does not verify. A leaked or intercepted instruction is useless after its slot passes.

proof.public_inputs = {
  action_hash,
  not_before: <slot>,
  not_after:  <slot + window>,
  ...
}

Why this is the unlock for autonomy

Without replay protection, "unattended agent-to-agent settlement" is reckless — a single captured message can be cashed in repeatedly. With it, machine-to-machine commerce becomes safe: an agent can buy an inference or a forecast from another agent, and the authorization expires the moment its window closes.