Assign an input guardrail to an individual specialized agent when separate agents handle distinct conversations and need their own content policies. A per-agent guardrail enforces its policy only on messages sent to that agent. This gives you targeted control beyond the single instance-level guardrail.
Per-agent guardrails operate independently of the instance-level active guardrail. An agent has its own guardrail even when no instance-level guardrail is active, and the reverse is also true. When you assign a guardrail to a specialized agent, Opal evaluates each input the agent receives. Opal blocks the message when any input violates a policy.
Prerequisites
The guardrail must already exist and have a trained model with the status Ready. To create and train one, see Create and train an input guardrail.
Assign a guardrail to an agent
To assign a guardrail to an agent, complete the following steps:
- Open a specialized agent in the agent editor. See Manage agents for information on selecting or creating an agent.
-
Find the Input Guardrails setting.
- Select the trained guardrail you want to enforce for this agent from the drop-down list.
-
Select the version of the guardrail you want to enforce for this agent from the Select version drop-down list.
- Do one of the following to save your changes:
- Click Update if you are editing an existing agent.
- Click Save if you are creating an agent.
The guardrail takes effect immediately for conversations with that agent. To change the policy later, select a different guardrail version or clear the setting to remove per-agent enforcement.
Related documentation
Learn more about input guardrails in the following articles:
Article is closed for comments.