Getting Started
Register an education agent, apply an ArmorEdu bundle, and confirm the first held or blocked action.
This guide takes a single education agent from registration to a first enforced decision. It assumes you already have an ArmorIQ organization and API key. If you do not, start with account setup.
1. Register the agent
ArmorEdu policies target an agent by identity, so the agent must exist in your organization first. Register it from the dashboard under Agents, or from the SDK during bootstrap. Note the agent's ID: the policy binds to it.
See SDK client initialization for the registration call and Agents for the dashboard flow.
2. Apply a bundle
Open Policy Studio and start from one of the bundle templates that matches the agent's role (course management, research, or IT operations). Set the bundle's targetId to your agent, review the allowed tools and the block / hold overrides, then activate it.
A bundle begins in allow_log by default: every tool call is permitted and recorded while you confirm the tool list is complete. Tighten individual tools to hold or block once you have seen real traffic.
3. Integrate the agent with the SDK
Education agents enforce through the ArmorIQ SDK. In the agent's tool loop, capture the plan, obtain an intent token, and let the SDK check each tool call against the active policy:
from armoriq_sdk import ArmorIQClient
client = ArmorIQClient(
api_key="ak_your_key_here",
user_id="your-user-id",
agent_id="course-management-agent",
)The full instrumentation flow (capture_plan → get_intent_token → invoke) is documented in the SDK core methods. The TypeScript SDK exposes the same shape.
4. Confirm enforcement
Trigger an action the bundle restricts. For the course bundle, ask the agent to export student records: the call should be blocked. Ask it to publish final grades: the call should be held for approval and appear in the plans queue.
Every decision, whether allowed, held, or blocked, shows up in the observability trace for that turn. If enforcement does not behave as expected, see Troubleshooting.