Create Guardrail
This page explains how to create a new guardrail in the SPLX Platform. A guardrail in the Platform defines the policy you want to enforce, serves as the reference for viewing results and monitoring, and connects to the backend runtime instance that performs the actual protection on your AI system in production. The Platform acts as the control plane for the guardrail service, allowing you to configure and manage enforcement while the runtime instance executes it.
Add Guardrail
When you click Add Guardrail in the UI, you will see the following fields:
Guardrail Name Enter a descriptive name for the guardrail. Example: Finance Chatbot Guardrail.
Guardrail Description Provide a short explanation of the purpose of this guardrail. Example: Prevents system prompt leakage and blocks PII exposure in responses.
Custom Guardrail Instance Enable this option if the guardrail service is not running as SPLX SaaS or as the part of our VPC deployment. Use it when the guardrail service is deployed separate from the platform in your own infrastructure (e.g. on-premises). It connects the Platform control plane with your self-hosted guardrail service, so you can still configure and monitor it through the UI. When enabled, you will need to provide:
URL - the endpoint of your deployed guardrail service.
API Key - the authentication key used by the Platform to securely connect to your guardrail service.
Guardrail ID
Once you create a guardrail, the Platform assigns a Guardrail ID. It acts as the reference key that ties messages to their policy configuration and ensures results appear in the correct guardrail view within the Platform.
How it works
The Guardrail ID tells the guardrail service which policies to enforce for each message.
It also tells the Platform which guardrail instance should display the flagged/unflagged messages, metrics, and monitoring results.
Multiple AI systems can share the same Guardrail ID, allowing them to use identical protection policies while their results appear under the same guardrail in the UI.
You must include the Guardrail ID with every input and output message. This ensures that both user prompts and AI responses are evaluated against the right guardrail configuration.
Next Steps
After creating a guardrail, define your policy in the Platform to specify what should be flagged.
For details on how to send messages to the guardrail service, see Connect Guardrail Service page.
Last updated