Guardrails

Guardrails configure input and output guard settings to ensure a safe and controlled AI chat experience. There are two types of guards:

• Input Guard – Controls and validates user inputs.
• Output Guard – Regulates the agent's responses.

Buttons	Description
Reset	Restore the original condition
Save Changes	Save the guardrails changes

Table: Buttons in Guardrail Settings

Figure: Guardrail Interface

Input Guard

Input Guard	Description
Jailbreak Detection	Detects and blocks attempts to bypass or manipulate the agent's safety mechanisms
PII Detection	Identifies any Personally Identifiable Information (PII) in the user input
Secret Detection	Detects sensitive data like credentials, API keys, or custom-defined secrets in the input

Table: Input Guard Settings

Figure: All Input Guards Enabled in Guardrail Settings

Output Guard

Output Guard	Description
PII Detection	Detects and removes any Personally Identifiable Information from the agent's response
Secrets Detection	Identifies and blocks responses containing confidential or custom-defined sensitive information
Competitor Mention	Prevents the agent from mentioning competitor names or related information
Profanity	Detects and filters out offensive or vulgar language
Toxicity	Flags and blocks toxic, harmful, or abusive content in the agent's response

Table: Output Guard Settings

Figure: All Output Guards Enabled in Guardrail Settings