Most agent setups today look like:
LLM → tool call → execution
That means:
no deterministic boundary
no audit guarantee
no protection against prompt injection at execution time
ÆTHERYA changes that.
It evaluates every proposed action against deterministic constraints before execution.
Key properties:
fail-closed (any error → deny)
no LLM in the decision path
signed approvals with replay protection
auditable decision chain
The LLM describes intent. It does not get to act.