How Parmana compares
Teams usually control AI agent actions in one of three ways. This page compares each with Parmana on the points a security reviewer asks about. It describes categories, not named products.
| Question | API key and logs | Prompt or output filter | Approval workflow without signed records | Parmana |
|---|---|---|---|---|
| Is each action checked against a stated rule before it runs? | No. Logs show what happened after the fact. | Checks the text, not the business action. | Only where a person is asked. | Yes. A named, versioned policy decides, and the policy content is recorded. |
| Does a person approve the action, and is that approval recorded? | No. | No. | Usually recorded as a log entry. | Yes. The approval is a signature, checked before the action is released. |
| Can a third party verify the record without your system? | Only if you export and share the logs. | No. | Usually no. | Yes. Public keys and verification endpoints let a third party check a record. |
| What happens when a check cannot run? | Not applicable. | Varies. | Varies. | The action is refused. |
What Parmana does not do
Parmana does not prove a policy is correct, and it does not prove that the external system carried out an action. It records what was checked, approved, and released. See the security review page for the full list.