How to enable Guardrails

Turn on guardrails for an assistant: safety filters that keep its conversations on topic, block harmful requests and jailbreak attempts, reduce made-up answers, and hide personal information. You choose the guardrails in Staging, set them up, and deploy them so they protect the live assistant.

How guardrails work

Ejento AI has five guardrails. You turn each one on or off per assistant, in the Guardrails section of the assistant's Configuration panel:

Guardrail	What it does	Settings
Topical Control	Controls what topics the assistant can discuss, based on guidelines you write.	Yes: your guidelines
Ethical Moderation	Keeps responses ethically sound, for example by refusing harmful or illegal requests.	None, on or off
Jailbreak Detection	Stops attempts to trick the assistant into ignoring its rules (prompt injection and jailbreaks).	None, on or off
Hallucination	Reduces made-up information in responses.	None, on or off
PII Redaction	Hides personally identifiable information, such as names, phone numbers, and email addresses.	Yes: which kinds of information to hide

Guardrail changes work like any other change to an assistant: each switch you turn on is saved to Staging right away, and people only get the protection after you deploy. For a detailed explanation of each guardrail, see Understanding Guardrails.

Use cases

IT help desk: Turn on Topical Control with guidelines such as "Only answer questions about IT support" and "Do not give financial, legal, or medical advice," so the assistant stays in its lane. Add Jailbreak Detection so users can't talk it out of those rules.

HR assistant handling personal data: Turn on PII Redaction so names, addresses, and bank details are hidden. Exclude your company name so it isn't redacted in every answer.

Public-facing assistant: Before making an assistant available to everyone, turn on Ethical Moderation and Jailbreak Detection as a baseline.

1. Open your assistant

In the sidebar, click Assistants. Search for your assistant, then click the chat icon (Chat with assistant) in its Actions column.

2. Switch to Staging

At the top right, click Staging. You change guardrails in Staging; they reach the live assistant when you deploy. If the assistant already opens in Staging, skip this step.

3. Turn on the guardrails you need

In the Configuration panel on the right, click Guardrails, then turn on the switch next to each guardrail you want. The message Enabled guardrail successfully appears each time, and the change counter next to Deploy goes up.

Ethical Moderation, Jailbreak Detection, and Hallucination need nothing else. Topical Control and PII Redaction show a + icon once they're on; set them up in the next two steps.

4. Write the Topical Control guidelines

Click the + icon next to Topical Control. In Add Guidelines, type one rule per line, for example "Only answer questions about IT support" or "Do not give financial, legal, or medical advice." Press Enter to start a new line, or click the × next to a line to remove it. Then click Save. The message Topical Control guidelines updated successfully appears.

Write guidelines as clear instructions about what the assistant may and may not discuss. Questions outside them are blocked.

5. Choose what PII Redaction hides

Click the + icon next to PII Redaction. The window has two tabs:

Redact categories: check the kinds of information to hide. All categories are selected at first: Person, PersonType, LicensePlate, SortCode, PhoneNumber, Organization, Address, Email, URL, IPAddress, DateTime, BankAccountNumber, DriversLicenseNumber, and PassportNumber. Clear the ones you want to keep visible; for example, clear Organization and DateTime if answers need company names and dates. Click View supported PII entities to see what each category covers.

Exclude specific terms: under Values to never redact, enter words that should always stay visible, such as your company or product name.

Click Confirm. The message Updated categories for PII guardrail appears.

6. Deploy the guardrails

Click Deploy. The Deploy to Production window lists the guardrails under Guardrails. Optionally add a Description, then click Deploy Changes. The message Successfully deployed to production appears, and the guardrails now protect the live assistant.

How to enable Guardrails

How guardrails work#

Use cases#

1. Open your assistant#

2. Switch to Staging#

3. Turn on the guardrails you need#

4. Write the Topical Control guidelines#

5. Choose what PII Redaction hides#

6. Deploy the guardrails#