Practical guides to AI agent permissions and approvals

Learn how to define what an agent may do, decide when a person should review an action and apply the decision before a tool call runs.

These guides are for developers working with agents and external tools. They explain design patterns with illustrative examples. Nodrail is in development; the guides do not imply that a Nodrail SDK or runtime integration is already available.

Start with permissions

How to manage AI agent permissions

Build a permission matrix for a support agent. Distinguish the caller, operation, resource and conditions, then decide when to allow, deny or request review. Understand where the check belongs and what a hook or SDK cannot cover by itself.

Read the agent permissions guide.

Add review where it matters

How to add human approval to an AI agent workflow

Follow an illustrative release request from pending review to a decision and an execution claim. Learn how to show effective parameters, authenticate the reviewer, handle expiry and prevent an approval from authorizing a different action.

Read the human approval guide.

Put the ideas together

Start with the operation your agent needs to perform. Define its permission scope, identify the point where the tool call can be stopped and make any human approval specific to that request. Check error and denial paths before expanding access.

Learn about Nodrail or join the waitlist when registration is open.