Skip to main content
This tutorial creates one policy you can test immediately. Keep the first policy narrow: one real workflow at one trust boundary. If you want the operational product guide for the same task, use Create a Policy.
Before you start, make sure you can sign in to the Anthale dashboard and that you already know which workflow you plan to protect, such as a customer support assistant, a retrieval flow, or an internal copilot.

​
Prerequisites

  • Access to the Policies page for the Organization that will own the policy.
  • One workflow to protect and one boundary to test first.
  • A plan to test the policy with one integration guide or a direct SDK path after the policy is saved.

​
Create the policy

1

Start with one workflow and one boundary

Pick one workflow and the first boundary you want Anthale to govern. For most teams, that is the input path before the model call.
2

Create the policy in the dashboard

Open the Policies page, create a new policy, and give it a name that tells you what it protects, such as support-input-prod or docs-rag-input-staging.
3

Enable a minimal starter set

Start with prompt injection protection, data leakage prevention, and content moderation. That gives you coverage for control takeover, sensitive data exposure, and disallowed content without making the first policy hard to reason about.
4

Choose first actions

For the boundary you plan to test first, start with block where the request should stop immediately, redact where sanitized content is safe to continue, and detect where you want operational signal before stricter enforcement.
5

Save the policy

Save the policy and keep its identifier handy for the SDK quickstart you will run next.

​
Verify the result

  • Confirm the policy appears in the list for the current Organization.
  • Reopen it and check that the intended guardrails are enabled for the boundary you plan to test first.
  • Run one integration guide or the Python Direct SDK Path or TypeScript Direct SDK Path with that policy identifier. A blocked test prompt should return or raise block behavior before the model continues.
You now have a first policy that is narrow enough to test and concrete enough to tune.

​
Next steps

If you need the dashboard how-to, read Create a Policy. Otherwise start with the integration guide that matches your stack, or use the Python Direct SDK Path or TypeScript Direct SDK Path for a lower-level implementation.
Last modified on April 13, 2026