Guardrail Customization gives administrators the tools to manage how the Orchestrator enforces safety protocols. By defining these rules, you can control which actions the AI Agent is allowed to take and which ones should be flagged or blocked. This feature works alongside standard System Guardrails to provide business-specific oversight, channel-level controls, and tailored messaging for your users.
- Pre-Requisites
- How to Access Guardrail Customization
- System Guardrails
- Enabling Custom Guardrail Settings
- Configuration Options
Pre-Requisites
To configure Guardrail Customization, you must have Administrator access. Your workspace must also be running an Agentic AI-enabled plan.
How to Access Guardrail Customization
To begin configuring your settings, follow these steps:
1. Click the gear icon (⚙) located on the right side of the top Agentic bar to open the Settings.
2. In the Orchestrator Configuration locate the Guardrails Configuration section.
3. Click Guardrail Customization to open the configuration panel.
System Guardrails
The System Guardrails section is located at the top of the page and displays an "Active" badge. These protocols are enforced by Talkdesk at the platform level; they remain active at all times to provide a baseline of security, regardless of your specific customizations.
Note: System Guardrails cannot be disabled. They protect all agentic interactions against harmful content, policy violations, and privacy breaches by default.
Enabling Custom Guardrail Settings
Switch the Enable Guardrail Settings toggle to the ON position to activate your custom configurations. Once this is enabled, the options detailed in the following sections will take effect.
Configuration Options
Guardrail Prompt
Note: When this toggle is OFF, only the System Guardrails are enforced. Custom rules, channel configurations, and personalized messages will not take effect until the toggle is enabled.
The Guardrail Prompt acts as a secondary safety layer. Use this field to set business-specific boundaries for the AI Agent’s responses, ensuring they align with your brand voice and regulatory requirements.
For instance, you might use this to prevent the agent from quoting specific financial rates or making unauthorized legal commitments on behalf of the company.
The field accepts a maximum of 300 characters. Be concise and specific to ensure consistent enforcement across conversations.
Active Channels
The Active Channels section defines which communication channels are subject to your custom guardrail configuration.
Default: All channels using Agentic AI are protected with Talkdesk’s default guardrail settings. No additional configuration is required.
Custom channel: Click Add Channel to select specific channels for custom configuration. Channels that are not explicitly added will continue to operate under the default guardrail settings.
Number of Triggers Before Closing the Conversation
This setting controls how many times a user is allowed to send messages that trigger the guardrail before the conversation is automatically closed. Enter a value between 1 and 5.
This limit applies to both System and Custom Guardrails. A value of 1 closes the conversation immediately after the first violation, while a value of 5 allows the user up to five attempts before the session ends.
Message Type
When a custom message is configured, you can set the channel Type to control how guardrail-triggered responses are delivered:
- Default: Uses the standard Talkdesk guardrail response when a guardrail is triggered.
- Custom: Allows you to define your own response messages and add language variants. Selecting this option enables the Guardrail Triggered Messages and Limit Reached Messages fields.
Language
Each active channel supports multilingual guardrail messages. By default, the Default language is applied. Click Add Language to configure guardrail messages in additional languages for your audience.
Guardrail Triggered Messages
This message is delivered to the user when their input activates the guardrail. Use this field to inform users that their request cannot be processed, without revealing the specific rule that was triggered.
To set up text for each of your supported languages, click + Add More Message.
Limit Reached Messages
When a user reaches the Number of Triggers threshold, this message is delivered before the conversation is closed. It should clearly indicate that the session has ended due to repeated policy violations, while maintaining a professional tone.
Use the + Add More Message button to provide localized versions of this closing message.
Once you have finished adjusting your settings, click Save to apply the changes.