VGE AIDRInstall

Summary

Analyze and guard text content either as user input prompts before sending to an LLM or as LLM-generated outputs before returning them to users, applying security policies and optionally sanitizing or blocking content.

Inputs

• Stream Direction (required) — Specify whether to guard user input prompts ('input') before they reach the LLM, or to guard LLM responses ('output') before returning to the user.
• Text (required) — The text to analyze and guard; interpreted as the user prompt when guarding input, or as the LLM response when guarding output.
• Original Prompt — (Output mode only) The original user prompt supplied to the LLM, used to detect context manipulation in the LLM's output.
• Passthrough Fields — Comma-separated JSON field names from the input item to pass through unchanged to the output item.
• Options — Additional optional parameters including request timeout in milliseconds, whether to fail open on errors (continue with original text), inclusion of the full VGE response in output, and custom JSON metadata to send with the request.

Output shape

a list of JSON objects, one per input item, each including guarded text and associated decision data

Each output item contains the guarded or sanitized text under 'guardedText', the VGE decision (ALLOWED, SANITIZED, BLOCKED) and associated fields such as score, threat level, categories, and request ID. If enabled, includes the original text, redacted or sanitized variants, block messages, and optionally the full API response. Passthrough fields specified are copied unchanged. On errors, depending on 'failOpen', the node either returns the original text with error details or throws an error to halt execution.

Links

• n8n community nodes installation guide