Guardrails

Guardrails are a security system offered by Wikit that analyze the content of requests sent to the large language models used in your applications. Their objective is to filter user requests to ensure proper functioning and safe use of the application.

In practice, guardrails serve to:

1. Optimize "small talk" exchanges by processing them with a lighter, less resource-intensive model.
2. Block toxic messages to protect the conversational environment.
3. Prevent prompt injection attacks, which aim to manipulate the model to obtain unwanted responses.

These guardrails can be easily activated via your application's settings, thereby enabling smarter and safer management of user interactions.

With a guardrail in place, your application knows when to redirect an innocuous conversation to a simpler model, when to detect inappropriate content, and how to react to prevent abuse. It's a simple yet effective solution to improve both the security and efficiency of your applications.