How to manage Zeus AI Guardrails
Last updated September 14, 2026
Overview
The Guardrails section in Zeus AI allows administrators to define rules that help control, validate, and monitor how Zeus AI handles customer and agent messages.
What is a guardrail?
A guardrail is a rule that helps ensure Zeus AI behaves as expected when handling messages. Depending on the guardrail type, it can instruct the AI on how to respond, limit the length of a message, detect specific patterns, or validate URLs.
Guardrails can be applied to either "Agent Message" or "Customer Message", depending on what you need to control.
Key Features
- Multiple guardrail types: Create "Prompt Guardrail", "Content Limit Guardrail", "Regex Pattern Guardrail", or "URL Guardrail" depending on what you need to control.
- "Guardrail visibility": Choose between "External (Editable by all users)" and "Internal (Atlas team only)".
- "Message side": For supported guardrail types, select whether the rule applies to "Agent Message" or "Customer Message".
- "Custom configuration": Configure the prompt, character limit, regex pattern, or URL validation according to the selected guardrail type.
- "Guardrail status": Manage whether a guardrail is active, inactive, or set to "Log Only".
How to create a new guardrail
- Navigate to " Zeus AI > Agent > Guardrails " and click "+ Add guardrail".
- Enter a clear and descriptive "Guardrail name" that explains what the guardrail controls.
- Select a "Guardrail Type". The available options are:
- "Prompt Guardrail": Enter instructions that define how Zeus AI should handle or evaluate messages. The editor also supports formatting and content options such as headings, lists, code, quotes, tables, images, canned messages, Help Center articles, mentions, variables, and emojis.
- "Content Limit Guardrail": Enter a "Character Limit" to restrict the maximum number of characters allowed in a message.
- "Regex Pattern Guardrail": Enter a valid regular expression to detect a specific pattern in a message.
- "URL Guardrail": Validates URLs in agent messages. This option is only visible to Atlas users.
- Select "Guardrail Visibility" to choose who can edit the guardrail:
- "External (Editable by all users)": All users can view and edit the guardrail.
- "Internal (Atlas team only)": Only the Atlas team can edit the guardrail.
- Select "Message Side":
- "Agent Message": Applies the guardrail to messages generated by Zeus AI.
- "Customer Message": Applies the guardrail to messages received from customers.
- Configure the fields required for the selected guardrail type.
- Review the configuration and click "Create" in the upper-right corner. The guardrail is saved with "Inactive" status.
Guardrail setup example The following example helps prevent the Zeus AI agent from giving customers incorrect information about refunds. When a customer asks about a refund, Zeus AI should explain the refund policy instead of making up exceptions or promising a specific outcome.
Configuration settings
| Setting | Value |
| "Name" | Refund Policy Guardrail |
| "Guardrail Type" | Prompt Guardrail |
| "Guardrail Visibility" | Internal (Atlas team only) |
| "Message Side" | Customer Message |
| "Prompt" | When a customer asks about a refund, always explain the applicable refund policy. Do not promise a refund, make exceptions, or provide information that is not included in the refund policy. If the policy does not provide an answer, explain that the request requires further assistance. |
Important: This configuration gives Zeus AI a clear instruction for handling refund questions. In simple terms the guardrail tells Zeus AI what to do when a customer asks about refunds: explain the policy, avoid making promises, and ask for further assistance when the policy does not provide an answer.
How to manage a guardrail
After creating a guardrail, you can edit, activate inactivate or delete.
How to edit a guardrail
- Find the guardrail you want to update, click the ellipsis icon and select "Edit", or click directly on the guardrail.
- Update the guardrail configuration as needed.
Important: The "Guardrail Type" is not editable.
How to set a guardrail as active
To activate an inactive or log only guardrail, follow the steps:
- Find the guardrail you want to activate, click the ellipsis icon and select "Set Active".
Important: The active guardrail is enforced. When a message meets the guardrail's configured conditions, the guardrail is applied according to its configuration.
How to set a guardrail to "Log Only"
- Find the guardrail you want to monitor, click the ellipsis icon and select "Log Only".
How to set a guardrail as inactive
To stop a guardrail from running:
- Find the guardrail you want to disable, click the ellipsis icon and select "Set Inactive".
Important: An inactive guardrail is disabled until it is set to active again.
How to delete a guardrail
- Find the guardrail you want to delete, click the ellipsis icon, and select "Delete."
- Click "OK" to confirm the deletion.
Important: This action cannot be undone. If you may need the guardrail again later, use "Set Inactive" instead of deleting it.
FAQs & Troubleshooting
What is a guardrail?
A guardrail is a rule used to control, validate, or monitor how Zeus AI handles messages. Depending on the type, it can provide instructions, limit message length, detect specific patterns, or validate URLs.
Which guardrail type should I use?
Choose the type based on what you need to control:
- "Prompt Guardrail": Define instructions or rules for how Zeus AI should handle or evaluate a message.
- "Content Limit Guardrail": Limit the maximum number of characters allowed in a message.
- "Regex Pattern Guardrail": Detect a specific text pattern using a regular expression.
- "URL Guardrail": Validate URLs in agent messages. This option is only available to Atlas users.
What is the difference between "Agent Message" and "Customer Message"?
- "Agent Message" applies the guardrail to messages generated by Zeus AI.
- "Customer Message" applies the guardrail to messages received from customers.
For example, you can use an "Agent Message" guardrail to instruct Zeus AI not to discuss invoices, or a "Customer Message" guardrail to identify a specific type of customer message.
Who can edit a guardrail?
This is controlled by "Guardrail Visibility":
- "External (Editable by all users)": All users can view and edit the guardrail.
- "Internal (Atlas team only)": Only Atlas team members can edit the guardrail.
Can I use variables in a Prompt Guardrail?
Yes. You can use available custom variables, enter {{, to include dynamic information in the prompt.
Can I format a prompt guardrail?
Yes. The prompt editor supports formatting and content options such as headings, lists, code, quotes, tables, images, canned messages, Help Center articles, mentions, variables, and emojis.
What does "Log Only" mean?
"Log Only" allows you to monitor a guardrail without enforcing it. It can be useful for testing a new guardrail and understanding when it would be triggered before making it active.
Can I change a guardrail after creating it?
Yes. Click the guardrail's ellipsis icon, select "Edit", update the configuration, and save your changes.
I can't create a guardrail. What should I check?
Make sure you have entered a "Guardrail name", selected a "Guardrail Type", and completed all required fields for that type.
My Regex Pattern Guardrail is not working.
Check that the value entered in "Regex Pattern" is a valid regular expression and that the pattern matches the content you want to detect.
For example, if your goal is to detect a specific type of number, make sure the regex actually matches the format used in the messages you are testing.
My guardrail is not being applied to the expected messages.
Check the "Message Side" setting.
An "Agent Message" guardrail applies to messages generated by Zeus AI, while a "Customer Message" guardrail applies to messages received from customers.
Also check that the guardrail is not set to "Inactive" if you expect it to be applied.
Other users cannot edit my guardrail. Why?
Check the "Guardrail Visibility" setting.
If it is set to "Internal (Atlas team only)", only Atlas team members can edit the guardrail.
My Prompt Guardrail is not producing the expected result.
Review the prompt and make sure the instructions are specific and unambiguous.
Clearly describe:
- What Zeus AI should do.
- What Zeus AI should avoid.
- What information Zeus AI should use.
- What should happen when the required information is not available.
If you are using a Prompt Guardrail to review another AI's response, make sure the prompt clearly defines what should be approved or rejected and includes any required variables, such as {{response_to_verify}}.
My Content Limit Guardrail is not behaving as expected.
Check the "Character Limit" value and make sure it represents the maximum number of characters you want to allow.
I cannot see the "URL Guardrail" option.
The "URL Guardrail" is only visible to Atlas users. If you need this option and cannot see it, check whether your account has the required access.
I want to test a guardrail without enforcing it.
Set the guardrail to "Log Only". This allows you to monitor when the guardrail would be triggered before enabling it.
I no longer want a guardrail to run, but I may need it later.
Use "Set Inactive" instead of "Delete". This disables the guardrail while keeping its configuration available for future use.