How do I connect an AI agent I already run to Parmana?
Give the agent its own key, limited to the actions it may take. Then send each action through Parmana before it reaches your system.
Create a key for the agent that allows only the actions it needs. The operator issues the key, and it is never shared or written into logs.
Send each action to Parmana first. Parmana checks it against your rules and either lets it through or refuses it with a reason.
If a refusal asks for a person's approval, get that approval and send the request again. Parmana records every step. TypeScript and Python libraries are available, and the same steps work over plain HTTP.