Skip to main content
# How Parmana Works Parmana sits between AI systems and real-world execution to ensure that every action is explicitly authorized before it happens. --- ## The simple idea AI can generate actions. But AI should not be trusted to execute actions directly. Parmana ensures every action is checked before execution. --- ## The system in 5 steps ### 1. AI generates an action An AI system proposes something like: - approve a payment - process a refund - deploy infrastructure - approve a claim --- ### 2. The action becomes a signal The proposed action is converted into a structured signal: - standardized format - validated schema - traceable origin --- ### 3. Governance evaluates it Parmana checks: - signed policies - deterministic rules - authorization conditions It answers:
Is this action allowed?
--- ### 4. A decision is produced The system produces one of: - APPROVE - REJECT - REQUIRE_OVERRIDE This decision is deterministic and reproducible. --- ### 5. Execution happens only if approved Only approved decisions are passed to execution systems. Every execution is: - enforced - tracked - signed - verifiable --- ## What makes this different Traditional systems assume trust between layers. Parmana removes trust and replaces it with verification. --- ## Before vs After Parmana ### Before 

AI → Direct Execution



Unverified, probabilistic, and inconsistent.



After

AI → Signals → Governance → Authorization Decision → Execution Runtime → Attestation



Deterministic, verifiable, and enforceable.



The key outcome



Parmana ensures:



AI can propose actions

but cannot execute them

unless explicitly authorized

Simple summary



Parmana answers one question:



“Was this action authorized before it was executed?”



If yes → execute

If no → block