Creating Your First Guardrail

This tutorial helps you build a reusable guardrail profile that protects both inputs and outputs for a production-facing deployment.

​

What You Will Build

A guardrail profile that:

1. Uses Bud Sentinel probes for safety and PII categories.
2. Applies both input and output guard types.
3. Uses a tuned severity threshold.
4. Deploys to an existing project deployment.

​

Prerequisites

• Access to Bud AI Foundry.
• A project with at least one active deployment.
• Permission to create and deploy guardrails.

​

Step 1: Start the Workflow

1. Go to Guardrails.
2. Click + Add Guardrail.
3. Confirm you can access the provider selection screen.

​

Step 2: Choose Provider and Probes

1. Choose a provider (for example, Bud Sentinel).
2. Click Next.
3. Search probe catalog with tags (for example, Safety and PII).
4. Select the probes you need.
5. Review probe details to validate coverage.

​

Step 3: Choose Deployment Type

1. Choose Guardrail Endpoint to deploy as a standalone guardrail endpoint.
2. Select Add to Existing Deployment so the profile protects an already-running endpoint.

​

Step 4: Choose Project and Deployment

1. Pick your target project.
2. Select the deployment/endpoint you want to protect.
3. Continue to profile configuration.

​

Step 5: Configure the Profile

Set:

• Name: production-safety-guardrail
• Guard Types: input, output
• Severity: start at medium and tune based on observed traffic
• Description: add policy intent and owner team

​

Step 6: Deploy

1. Click Deploy.
2. Wait for the success state.
3. Open the guardrail detail page and validate:
   • selected probes
   • active deployment mapping
   • status and timestamps

​

Step 7: Validate with Test Prompts

Use known-safe and known-risk prompts to verify behavior and reduce false positives/negatives before broad rollout.

​

Best Practices

• Start with medium severity and tune with real traffic.
• Combine input and output checks for user-facing workloads.
• Document the owner, intent, and rollout scope in the description.
• Validate with known-safe and known-risk prompts before broad rollout.

​

Optional: Create Your First Governance Policy

If your project includes agents, create a governance policy after the endpoint guardrail is in place. Open Guardrails, select Governance Policy, create a policy, choose a conflict strategy, add checks or tool-access patterns, set lifecycle hooks, select an outcome, add guidance and optional filters, then deploy the policy to the project. After deployment, open the project agent, choose the agent version, and attach the policy from the agent builder Governance section.