AI Guardrails Generator

AI Guardrails Generator: Rules and Injection Tests

Describe your AI app and get a guardrail block for your system prompt, a JSON policy file and a set of prompt injection test cases tailored to your app. Free, runs in your browser.

Your AI app

Sensitive data to protect

1. System prompt guardrail block

2. JSON policy file

3. Prompt injection test cases

How to use the AI Guardrails Generator

  1. Enter your app name and choose the type of app.
  2. List the topics the AI may and may not discuss, one per line.
  3. Add the tools the AI can call, if any, and when it should hand off to a human.
  4. Tick the kinds of sensitive data to protect.
  5. Copy the guardrail block into your system prompt, keep the JSON policy in your repo, and run the test prompts against your app.

Frequently asked questions

What are AI guardrails?

Guardrails are rules and checks that keep an AI app inside its intended behaviour: what it may discuss, what data it may reveal, which tools it may call, and when it should hand off to a person. They can live in the prompt, in code around the model, or both.

What is prompt injection?

OWASP, in its Top 10 for LLM applications (LLM01:2025), says a prompt injection vulnerability occurs when user prompts alter the LLM’s behavior or output in unintended ways. Indirect injection is the same idea arriving through content the model reads, such as a document or web page.

Are guardrails in the system prompt enough?

No. Instructions in a prompt can be bypassed, so treat them as one layer. Enforce the important limits in code: allow-list tools, require confirmation for risky actions, filter outputs for secrets and personal data, and give the app only the access it needs.

What is the JSON policy file for?

It is a plain description of your rules by stage (input, context, output, tool and conversation). Use it as documentation, as configuration for your own checks, or as a starting point for a guardrail library. The format is this tool’s own, not an industry standard.

How do I use the test cases?

Send each prompt to your app and compare the reply with the expected behaviour. Run them again whenever you change the prompt, the model or the tools.

Is my app description uploaded?

No. Everything is generated in your browser and nothing is stored.

Everything runs in your browser. Nothing you enter is uploaded or stored by Nexatoolbox.