Introduction

Actera is the authority layer for AI agents. It defines what agents can do autonomously, what requires human approval, and what must be blocked.

Your integration asks Actera before an agent acts. Actera evaluates the request against the agent’s assigned policies and returns a deterministic decision:

Decision	What your integration does
Allow	Continue with the action.
Require approval	Pause the action for an eligible reviewer.
Block	Stop the action.

Policies express delegated authority. They can depend on the action, trusted business context, and cumulative usage across requests. One request is not the whole risk: several individually small refunds can exceed a customer’s daily allowance.

Connect, observe, enforce

Start with one agent and one operation you want to control.

1. Connect an agent through Claude Code, an SDK, or HTTP.
2. Review real requests in Decisions and use them to create focused policies.
3. Test the expected outcomes, publish the policies, and assign them to the agent.
4. Enable enforcement in the plugin or application and verify an allowed, blocked, and approval-required request.

Monitoring records what the policies would decide while the integration continues. Enforcing makes the integration apply those decisions. Connecting an agent or assigning policies alone does not switch its integration into enforcement.

Who uses Actera

Platform and security teams define reusable authority boundaries. Application teams enforce them where actions are dispatched. Reviewers handle exceptions in Needs approval, and operations teams use decision evidence and execution reports to understand what happened.

Actera covers the action paths connected to it. Your application must place the check where it can still stop the action and supply trustworthy facts. A decision does not independently prove that a downstream operation completed.

Start with the TypeScript quickstart or Claude Code setup. Read How Actera works for the request lifecycle.