Knowledge Base
Knowledge Base
  • Login
Font Size:
Ask Joget AI

Guards

Overview

Guards check messages and responses for safety or policy violations, then block, redact, or redirect when something crosses a line. They sit around a task to keep it in bounds. When a violation is detected, three actions are possible:

  • Block: Abort the task
  • Sanitize: Replace content
  • Redirect: Jump to another task

The following is a quick summary on how each guard type works:

Guard How It Works
LLM-Based Guard Delegates safety decisions to a secondary LLM (e.g., Llama Guard) using a configurable safety prompt.
Rule-Based Guard Evaluates content against regex patterns and keyword blocklists; redacts matches with [REDACTED] by default.
Webhook Guard POSTs content to an external HTTP endpoint that returns a {"safe": true/false} decision.

How to use

Drag and drop a Decision from the Decisions section in the palette into the Drop a decision rule to here section of the task in the canvas.

Available guards

Created by Debanraj Ravindran Last modified by Debanraj Ravindran on Jul 27, 2026
Tags:
Previous BeanShell Decision NextRule-Based Guard